API tools faq
paste
lamiastella

nan loss

lamiastella
Oct 16th, 2020 (edited)
465
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
text 2.22 MB | None | 0 0
raw download clone embed print report
  1. 1
  2.  
  3. network = Network()
  4.  
  5. 2
  6.  
  7. network.cuda()
  8.  
  9. 3
  10.  
  11. ​
  12.  
  13. 4
  14.  
  15. criterion = nn.MSELoss()
  16.  
  17. 5
  18.  
  19. optimizer = optim.Adam(network.parameters(), lr=0.0001)
  20.  
  21. 6
  22.  
  23. ​
  24.  
  25. 7
  26.  
  27. loss_min = np.inf
  28.  
  29. 8
  30.  
  31. num_epochs = 10
  32.  
  33. 9
  34.  
  35. ​
  36.  
  37. 10
  38.  
  39. start_time = time.time()
  40.  
  41. 11
  42.  
  43. for epoch in range(1,num_epochs+1):
  44.  
  45. 12
  46.  
  47.  
  48.  
  49. 13
  50.  
  51. loss_train = 0
  52.  
  53. 14
  54.  
  55. loss_test = 0
  56.  
  57. 15
  58.  
  59. running_loss = 0
  60.  
  61. 16
  62.  
  63.  
  64.  
  65. 17
  66.  
  67.  
  68.  
  69. 18
  70.  
  71. network.train()
  72.  
  73. 19
  74.  
  75. print('size of train loader is: ', len(train_loader))
  76.  
  77. 20
  78.  
  79. ​
  80.  
  81. 21
  82.  
  83. for step in range(1, len(train_loader)+1):
  84.  
  85. 22
  86.  
  87. ​
  88.  
  89. 23
  90.  
  91.  
  92.  
  93. 24
  94.  
  95. batch = next(iter(train_loader))
  96.  
  97. 25
  98.  
  99. images, landmarks = batch['image'], batch['landmarks']
  100.  
  101. 26
  102.  
  103. print(images.shape)
  104.  
  105. 27
  106.  
  107.  
  108.  
  109. 28
  110.  
  111. images = images.unsqueeze_(1)
  112.  
  113. 29
  114.  
  115. ​
  116.  
  117. 30
  118.  
  119. images = torch.cat((images,images,images),1)
  120.  
  121. 31
  122.  
  123. images = images.cuda()
  124.  
  125. 32
  126.  
  127.  
  128.  
  129. 33
  130.  
  131. landmarks = landmarks.view(landmarks.size(0),-1).cuda()
  132.  
  133. 34
  134.  
  135. norm_image = transforms.Normalize(0.3812, 0.1123)
  136.  
  137. 35
  138.  
  139. for image in images:
  140.  
  141. 36
  142.  
  143. image = image.float()
  144.  
  145. 37
  146.  
  147. ##image = to_tensor(image) #TypeError: pic should be PIL Image or ndarray. Got <class 'torch.Tensor'>
  148.  
  149. 38
  150.  
  151. image = norm_image(image)
  152.  
  153. 39
  154.  
  155.  
  156.  
  157. 40
  158.  
  159. ###removing landmarks normalize because of the following error
  160.  
  161. 41
  162.  
  163. ###ValueError: Expected tensor to be a tensor image of size (C, H, W). Got tensor.size() = torch.Size([8, 8])
  164.  
  165. 42
  166.  
  167.  
  168.  
  169. 43
  170.  
  171.  
  172.  
  173. 44
  174.  
  175. for i in range(8):
  176.  
  177. 45
  178.  
  179. if(i%2==0):
  180.  
  181. 46
  182.  
  183. landmarks[:,i] = landmarks[:,i]/800
  184.  
  185. 47
  186.  
  187. else:
  188.  
  189. 48
  190.  
  191. landmarks[:,i] = landmarks[:,i]/600
  192.  
  193. 49
  194.  
  195.  
  196.  
  197. 50
  198.  
  199. print(landmarks.shape)
  200.  
  201. 51
  202.  
  203. print(landmarks)
  204.  
  205. 52
  206.  
  207.  
  208.  
  209. 53
  210.  
  211.  
  212.  
  213. 54
  214.  
  215. ​
  216.  
  217. 55
  218.  
  219.  
  220.  
  221. 56
  222.  
  223. ##norm_landmarks = transforms.Normalize(0.4949, 0.2165)
  224.  
  225. 57
  226.  
  227. landmarks [landmarks != landmarks] = 0
  228.  
  229. 58
  230.  
  231. landmarks = landmarks.unsqueeze_(0)
  232.  
  233. 59
  234.  
  235. landmarks = norm_landmarks(landmarks)
  236.  
  237. 60
  238.  
  239.  
  240.  
  241. 61
  242.  
  243. predictions = network(images)
  244.  
  245. 62
  246.  
  247.  
  248.  
  249. 63
  250.  
  251. # clear all the gradients before calculating them
  252.  
  253. 64
  254.  
  255. optimizer.zero_grad()
  256.  
  257. 65
  258.  
  259.  
  260.  
  261. 66
  262.  
  263. print('predictions are: ', predictions.float())
  264.  
  265. 67
  266.  
  267. print('landmarks are: ', landmarks.float())
  268.  
  269. 68
  270.  
  271. # find the loss for the current step
  272.  
  273. 69
  274.  
  275. loss_train_step = criterion(predictions.float(), landmarks.float())
  276.  
  277. 70
  278.  
  279.  
  280.  
  281. 71
  282.  
  283.  
  284.  
  285. 72
  286.  
  287. loss_train_step = loss_train_step.to(torch.float32)
  288.  
  289. 73
  290.  
  291. print("loss_train_step before backward: ", loss_train_step)
  292.  
  293. 74
  294.  
  295.  
  296.  
  297. 75
  298.  
  299. # calculate the gradients
  300.  
  301. 76
  302.  
  303. loss_train_step.backward()
  304.  
  305. 77
  306.  
  307.  
  308.  
  309. 78
  310.  
  311. # update the parameters
  312.  
  313. 79
  314.  
  315. optimizer.step()
  316.  
  317. 80
  318.  
  319.  
  320.  
  321. 81
  322.  
  323. print("loss_train_step after backward: ", loss_train_step)
  324.  
  325. 82
  326.  
  327. ​
  328.  
  329. 83
  330.  
  331.  
  332.  
  333. 84
  334.  
  335. loss_train += loss_train_step.item()
  336.  
  337. 85
  338.  
  339.  
  340.  
  341. 86
  342.  
  343. print("loss_train: ", loss_train)
  344.  
  345. 87
  346.  
  347. running_loss = loss_train/step
  348.  
  349. 88
  350.  
  351. print('step: ', step)
  352.  
  353. 89
  354.  
  355. print('running loss: ', running_loss)
  356.  
  357. 90
  358.  
  359.  
  360.  
  361. 91
  362.  
  363. print_overwrite(step, len(train_loader), running_loss, 'train')
  364.  
  365. 92
  366.  
  367.  
  368.  
  369. 93
  370.  
  371. network.eval()
  372.  
  373. 94
  374.  
  375. with torch.no_grad():
  376.  
  377. 95
  378.  
  379.  
  380.  
  381. 96
  382.  
  383. for step in range(1,len(test_loader)+1):
  384.  
  385. 97
  386.  
  387.  
  388.  
  389. 98
  390.  
  391. batch = next(iter(train_loader))
  392.  
  393. 99
  394.  
  395. images, landmarks = batch['image'], batch['landmarks']
  396.  
  397. 100
  398.  
  399. images = images.cuda()
  400.  
  401. 101
  402.  
  403. landmarks = landmarks.view(landmarks.size(0),-1).cuda()
  404.  
  405. 102
  406.  
  407. ##[8, 600, 800] --> [8,3,600,800]
  408.  
  409. 103
  410.  
  411. images = images.unsqueeze(1)
  412.  
  413. 104
  414.  
  415. images = torch.cat((images, images, images), 1)
  416.  
  417. 105
  418.  
  419. predictions = network(images)
  420.  
  421. 106
  422.  
  423. ​
  424.  
  425. 107
  426.  
  427. # find the loss for the current step
  428.  
  429. 108
  430.  
  431. loss_test_step = criterion(predictions, landmarks)
  432.  
  433. 109
  434.  
  435. ​
  436.  
  437. 110
  438.  
  439. loss_test += loss_test_step.item()
  440.  
  441. 111
  442.  
  443. running_loss = loss_test/step
  444.  
  445. 112
  446.  
  447. ​
  448.  
  449. 113
  450.  
  451. print_overwrite(step, len(test_loader), running_loss, 'Validation')
  452.  
  453. 114
  454.  
  455.  
  456.  
  457. 115
  458.  
  459. loss_train /= len(train_loader)
  460.  
  461. 116
  462.  
  463. loss_test /= len(test_loader)
  464.  
  465. 117
  466.  
  467.  
  468.  
  469. 118
  470.  
  471. print('\n--------------------------------------------------')
  472.  
  473. 119
  474.  
  475. print('Epoch: {} Train Loss: {:.4f} Valid Loss: {:.4f}'.format(epoch, loss_train, loss_test))
  476.  
  477. 120
  478.  
  479. print('--------------------------------------------------')
  480.  
  481. 121
  482.  
  483.  
  484.  
  485. 122
  486.  
  487. if loss_test < loss_min:
  488.  
  489. 123
  490.  
  491. loss_min = loss_test
  492.  
  493. 124
  494.  
  495. torch.save(network.state_dict(), '../moth_landmarks.pth')
  496.  
  497. 125
  498.  
  499. print("\nMinimum Valid Loss of {:.4f} at epoch {}/{}".format(loss_min, epoch, num_epochs))
  500.  
  501. 126
  502.  
  503. print('Model Saved\n')
  504.  
  505. 127
  506.  
  507.  
  508.  
  509. 128
  510.  
  511. print('Training Complete')
  512.  
  513. 129
  514.  
  515. print("Total Elapsed Time : {} s".format(time.time()-start_time))
  516.  
  517. size of train loader is: 90
  518. torch.Size([8, 600, 800])
  519. torch.Size([8, 8])
  520. tensor([[0.6168, 0.4029, 0.8523, 0.3417, 0.3588, 0.5000, 0.6125, 0.5400],
  521. [0.6213, 0.4131, 0.8438, 0.3550, 0.3513, 0.4400, 0.5716, 0.5123],
  522. [0.6286, 0.4078, 0.8063, 0.2267, 0.4788, 0.1533, 0.5953, 0.4913],
  523. [ nan, nan, 0.6512, 0.1717, 0.4100, 0.1983, 0.5253, 0.5240],
  524. [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170],
  525. [0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317],
  526. [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459],
  527. [0.6202, 0.4066, 0.8746, 0.3376, 0.3717, 0.3090, 0.5842, 0.5165]],
  528. device='cuda:0', dtype=torch.float64)
  529. predictions are: tensor([[-0.1423, -0.1056, -0.3138, 0.4684, -0.1054, -0.5563, -0.0186, 0.1072],
  530. [-0.1047, -0.1028, -0.2962, 0.4469, -0.1573, -0.5332, -0.0197, 0.1038],
  531. [-0.1244, -0.0833, -0.2863, 0.4311, -0.1496, -0.4844, -0.0656, 0.0925],
  532. [-0.1570, -0.1024, -0.2959, 0.4236, -0.1198, -0.4870, -0.0458, 0.1049],
  533. [-0.1260, -0.1189, -0.3429, 0.4834, -0.1040, -0.5703, 0.0156, 0.0999],
  534. [-0.1380, -0.0681, -0.3151, 0.4013, -0.1561, -0.5097, -0.0721, 0.0928],
  535. [-0.1592, -0.1133, -0.2992, 0.4642, -0.1194, -0.5710, -0.0054, 0.0882],
  536. [-0.1140, -0.0634, -0.3053, 0.4357, -0.1321, -0.5395, -0.0382, 0.0839]],
  537. device='cuda:0', grad_fn=<AddmmBackward>)
  538. landmarks are: tensor([[[ 0.5631, -0.4249, 1.6509, -0.7078, -0.6289, 0.0236, 0.5432,
  539. 0.2083],
  540. [ 0.5841, -0.3778, 1.6113, -0.6462, -0.6635, -0.2536, 0.3542,
  541. 0.0802],
  542. [ 0.6177, -0.4022, 1.4381, -1.2390, -0.0746, -1.5777, 0.4639,
  543. -0.0168],
  544. [-2.2859, -2.2859, 0.7222, -1.4930, -0.3921, -1.3698, 0.1404,
  545. 0.1343],
  546. [ 0.6471, -0.3840, 1.5767, -1.0311, -0.0457, -1.5007, 0.6889,
  547. 0.1020],
  548. [ 0.5849, -0.3836, 1.2649, -1.2236, -0.3460, -1.2313, 0.4508,
  549. 0.1698],
  550. [ 0.6520, -0.3912, 1.7095, -0.9079, -0.0284, -1.3621, 0.8096,
  551. 0.2356],
  552. [ 0.5787, -0.4077, 1.7537, -0.7263, -0.5692, -0.8586, 0.4126,
  553. 0.1000]]], device='cuda:0')
  554. loss_train_step before backward: tensor(1.0104, device='cuda:0', grad_fn=<MseLossBackward>)
  555. loss_train_step after backward: tensor(1.0104, device='cuda:0', grad_fn=<MseLossBackward>)
  556. loss_train: 1.0104049444198608
  557. step: 1
  558. running loss: 1.0104049444198608
  559. Train Steps: 1/90 Loss: 1.0104 torch.Size([8, 600, 800])
  560. torch.Size([8, 8])
  561. tensor([[0.6336, 0.4191, 0.8938, 0.5167, 0.3937, 0.3517, 0.7343, 0.5748],
  562. [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235],
  563. [0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118],
  564. [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413],
  565. [0.6125, 0.4035, 0.7825, 0.3100, 0.3463, 0.4900, 0.5832, 0.5637],
  566. [0.6114, 0.4018, 0.7213, 0.1967, 0.3763, 0.2700, 0.5875, 0.5533],
  567. [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183],
  568. [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617]],
  569. device='cuda:0', dtype=torch.float64)
  570. predictions are: tensor([[-0.0659, -0.2580, 0.0962, 0.1213, -0.2025, -0.6994, 0.0772, 0.1072],
  571. [-0.0033, -0.2372, 0.0826, 0.1461, -0.1880, -0.6930, 0.0741, 0.1244],
  572. [-0.0202, -0.2130, 0.0678, 0.1583, -0.1568, -0.6912, 0.0771, 0.1296],
  573. [ 0.0287, -0.2443, 0.0870, 0.1132, -0.1829, -0.6959, 0.0773, 0.1000],
  574. [-0.0197, -0.2014, 0.0873, 0.1103, -0.1730, -0.6639, 0.0322, 0.1340],
  575. [-0.0283, -0.2539, 0.0533, 0.1278, -0.1841, -0.6857, 0.0922, 0.1333],
  576. [ 0.0008, -0.2950, 0.0865, 0.1903, -0.2152, -0.7125, 0.1525, 0.1757],
  577. [ 0.0252, -0.2935, 0.0619, 0.1467, -0.1725, -0.6754, 0.0774, 0.1400]],
  578. device='cuda:0', grad_fn=<AddmmBackward>)
  579. landmarks are: tensor([[[ 0.6405, -0.3503, 1.8423, 0.1005, -0.4672, -0.6616, 1.1057,
  580. 0.3692],
  581. [ 0.5762, -0.4706, 1.7754, -0.0984, -0.3680, 0.2380, 0.6277,
  582. 0.1322],
  583. [ 0.6077, -0.3226, 0.9993, -0.6462, -0.2651, -1.3082, 0.2946,
  584. 0.5401],
  585. [ 0.6108, -0.4201, 1.8711, -0.7848, -0.0053, -1.2236, 1.0362,
  586. 0.2142],
  587. [ 0.5432, -0.4224, 1.3284, -0.8540, -0.6866, -0.0226, 0.4077,
  588. 0.3177],
  589. [ 0.5380, -0.4302, 1.0455, -1.3775, -0.5480, -1.0388, 0.4277,
  590. 0.2699],
  591. [ 0.5934, -0.4276, 1.8192, -0.1458, -0.4499, 0.3777, 0.6702,
  592. 0.1082],
  593. [ 0.5637, -0.4143, 1.7519, -0.0787, -0.3055, -0.0149, 0.3758,
  594. 0.3084]]], device='cuda:0')
  595. loss_train_step before backward: tensor(0.5292, device='cuda:0', grad_fn=<MseLossBackward>)
  596. loss_train_step after backward: tensor(0.5292, device='cuda:0', grad_fn=<MseLossBackward>)
  597. loss_train: 1.539566159248352
  598. step: 2
  599. running loss: 0.769783079624176
  600. Train Steps: 2/90 Loss: 0.7698 torch.Size([8, 600, 800])
  601. torch.Size([8, 8])
  602. tensor([[0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609],
  603. [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114],
  604. [0.6162, 0.4134, 0.6700, 0.2467, 0.3962, 0.2533, 0.5737, 0.5467],
  605. [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170],
  606. [0.6202, 0.4079, 0.8025, 0.2500, 0.3763, 0.3217, 0.6125, 0.5533],
  607. [0.6100, 0.4016, 0.8600, 0.5067, 0.4612, 0.5233, 0.5086, 0.5519],
  608. [0.6257, 0.4034, 0.8287, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901],
  609. [0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585]],
  610. device='cuda:0', dtype=torch.float64)
  611. predictions are: tensor([[ 0.1696, -0.3065, 0.4627, -0.0944, -0.2657, -0.7406, 0.1930, 0.1774],
  612. [ 0.1604, -0.3258, 0.4616, -0.0499, -0.2615, -0.7130, 0.2208, 0.1799],
  613. [ 0.1484, -0.3291, 0.4707, -0.0997, -0.2455, -0.7404, 0.1867, 0.1688],
  614. [ 0.1728, -0.3067, 0.4303, -0.0530, -0.2103, -0.7886, 0.2801, 0.1768],
  615. [ 0.1909, -0.2937, 0.4370, -0.1293, -0.2625, -0.7441, 0.1705, 0.1318],
  616. [ 0.1561, -0.3214, 0.4482, -0.0593, -0.2596, -0.7309, 0.2042, 0.2013],
  617. [ 0.1806, -0.3209, 0.4534, -0.0854, -0.2666, -0.7370, 0.2352, 0.1802],
  618. [ 0.1912, -0.2630, 0.4746, -0.0326, -0.2686, -0.7027, 0.2144, 0.1895]],
  619. device='cuda:0', grad_fn=<AddmmBackward>)
  620. landmarks are: tensor([[[ 0.6026, -0.4032, 1.7326, -0.7771, -0.3691, -0.8617, 0.9704,
  621. 0.3050],
  622. [ 0.5450, -0.4730, 1.7420, 0.1372, -0.1919, 0.2614, 0.4976,
  623. 0.0764],
  624. [ 0.5603, -0.3764, 0.8088, -1.1466, -0.4557, -1.1158, 0.3642,
  625. 0.2391],
  626. [ 0.6471, -0.3840, 1.5767, -1.0311, -0.0457, -1.5007, 0.6889,
  627. 0.1020],
  628. [ 0.5788, -0.4020, 1.4208, -1.1312, -0.5480, -0.8002, 0.5432,
  629. 0.2699],
  630. [ 0.5318, -0.4310, 1.6864, 0.0543, -0.1554, 0.1313, 0.0635,
  631. 0.2634],
  632. [ 0.6042, -0.4225, 1.5420, -1.2082, -0.4730, -1.0311, 0.6380,
  633. -0.0220],
  634. [ 0.5115, -0.4332, 1.7557, -0.3152, -0.2536, 0.3931, 0.4139,
  635. 0.2936]]], device='cuda:0')
  636. loss_train_step before backward: tensor(0.3366, device='cuda:0', grad_fn=<MseLossBackward>)
  637. loss_train_step after backward: tensor(0.3366, device='cuda:0', grad_fn=<MseLossBackward>)
  638. loss_train: 1.8762048482894897
  639. step: 3
  640. running loss: 0.6254016160964966
  641. Train Steps: 3/90 Loss: 0.6254 torch.Size([8, 600, 800])
  642. torch.Size([8, 8])
  643. tensor([[0.6163, 0.4114, 0.7650, 0.2017, 0.3763, 0.2867, 0.5631, 0.5071],
  644. [0.6109, 0.4041, 0.6975, 0.3167, 0.3513, 0.3383, 0.5153, 0.5319],
  645. [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667],
  646. [0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320],
  647. [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869],
  648. [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6188, 0.5283],
  649. [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750],
  650. [0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222]],
  651. device='cuda:0', dtype=torch.float64)
  652. predictions are: tensor([[ 0.3182, -0.3804, 0.7228, -0.2070, -0.2713, -0.7219, 0.3276, 0.2051],
  653. [ 0.3052, -0.3400, 0.7528, -0.3017, -0.3315, -0.7304, 0.2778, 0.2039],
  654. [ 0.3572, -0.3411, 0.7817, -0.2362, -0.2603, -0.7317, 0.3130, 0.1719],
  655. [ 0.3135, -0.3787, 0.8097, -0.2249, -0.3170, -0.6919, 0.3421, 0.2800],
  656. [ 0.3222, -0.3184, 0.7258, -0.2893, -0.2752, -0.7587, 0.2927, 0.1911],
  657. [ 0.3452, -0.3446, 0.7560, -0.2866, -0.3250, -0.7399, 0.3119, 0.1912],
  658. [ 0.3920, -0.3923, 0.8089, -0.1643, -0.3038, -0.6605, 0.3647, 0.2397],
  659. [ 0.2950, -0.3404, 0.7301, -0.2619, -0.2894, -0.7364, 0.2727, 0.1653]],
  660. device='cuda:0', grad_fn=<AddmmBackward>)
  661. landmarks are: tensor([[[ 0.5606, -0.3856, 1.2476, -1.3544, -0.5480, -0.9618, 0.3152,
  662. 0.0562],
  663. [ 0.5359, -0.4193, 0.9358, -0.8232, -0.6635, -0.7232, 0.0943,
  664. 0.1710],
  665. [ 0.5768, -0.3857, 1.5305, -0.7694, -0.6462, -0.6308, 0.3988,
  666. 0.3315],
  667. [ 0.6174, -0.4490, 1.8885, -0.0996, -0.4845, -0.3691, 0.9814,
  668. 0.1715],
  669. [ 0.6119, -0.3927, 1.6979, -0.7925, -0.2536, -1.3698, 0.4282,
  670. -0.0368],
  671. [ 0.6062, -0.3778, 1.8423, -0.6462, -0.4383, -1.0465, 0.5721,
  672. 0.1544],
  673. [ 0.5525, -0.4463, 1.7326, 0.0313, -0.2536, -0.0688, 0.1968,
  674. 0.3700],
  675. [ 0.6237, -0.2983, 1.3919, -0.4691, -0.0457, -1.2313, 0.2453,
  676. 0.5882]]], device='cuda:0')
  677. loss_train_step before backward: tensor(0.1705, device='cuda:0', grad_fn=<MseLossBackward>)
  678. loss_train_step after backward: tensor(0.1705, device='cuda:0', grad_fn=<MseLossBackward>)
  679. loss_train: 2.046713799238205
  680. step: 4
  681. running loss: 0.5116784498095512
  682.  
  683. Train Steps: 4/90 Loss: 0.5117 torch.Size([8, 600, 800])
  684. torch.Size([8, 8])
  685. tensor([[0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795],
  686. [ nan, nan, 0.8938, 0.2850, 0.4662, 0.3117, 0.7406, 0.5528],
  687. [0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117],
  688. [0.6185, 0.4067, 0.8838, 0.4450, 0.4037, 0.4733, 0.5213, 0.5142],
  689. [0.6110, 0.3984, 0.8750, 0.4933, 0.4625, 0.4950, 0.5578, 0.5676],
  690. [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834],
  691. [0.6293, 0.4097, 0.8800, 0.2517, 0.5262, 0.2600, 0.7430, 0.5378],
  692. [ nan, nan, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609]],
  693. device='cuda:0', dtype=torch.float64)
  694. predictions are: tensor([[ 0.4885, -0.4018, 1.0979, -0.3366, -0.3702, -0.6909, 0.3982, 0.2501],
  695. [ 0.4422, -0.3510, 1.0551, -0.4015, -0.3460, -0.7835, 0.3912, 0.2216],
  696. [ 0.4735, -0.3952, 1.0856, -0.3061, -0.3689, -0.6766, 0.3673, 0.2366],
  697. [ 0.4811, -0.3667, 1.0538, -0.3488, -0.3313, -0.6762, 0.3805, 0.2586],
  698. [ 0.4752, -0.3900, 1.0767, -0.3471, -0.3499, -0.7226, 0.3713, 0.2491],
  699. [ 0.4760, -0.4058, 1.0386, -0.4084, -0.3484, -0.8394, 0.4057, 0.2369],
  700. [ 0.4448, -0.3698, 1.0587, -0.4265, -0.3316, -0.8082, 0.3901, 0.2314],
  701. [ 0.4433, -0.3462, 1.0494, -0.4267, -0.3098, -0.7908, 0.3494, 0.1983]],
  702. device='cuda:0', grad_fn=<AddmmBackward>)
  703. landmarks are: tensor([[[ 6.3554e-01, -4.0805e-01, 1.6113e+00, 1.8522e-01, -4.7298e-01,
  704. 1.4673e-01, 9.9965e-01, 3.9055e-01],
  705. [-2.2859e+00, -2.2859e+00, 1.8423e+00, -9.6952e-01, -1.3233e-01,
  706. -8.4634e-01, 1.1349e+00, 2.6764e-01],
  707. [ 6.1184e-01, -3.9831e-01, 1.5824e+00, 3.4688e-01, -4.2679e-01,
  708. -6.8822e-02, 3.4688e-01, 5.3934e-01],
  709. [ 5.7079e-01, -4.0747e-01, 1.7961e+00, -2.3048e-01, -4.2102e-01,
  710. -9.9615e-02, 1.2187e-01, 8.9251e-02],
  711. [ 5.3637e-01, -4.4573e-01, 1.7557e+00, -7.2363e-03, -1.4965e-01,
  712. 4.6189e-04, 2.9049e-01, 3.3573e-01],
  713. [ 6.1264e-01, -4.0570e-01, 1.4439e+00, -1.3159e+00, -1.1501e-01,
  714. -1.5777e+00, 5.5366e-01, -5.2974e-02],
  715. [ 6.2072e-01, -3.9360e-01, 1.7788e+00, -1.1235e+00, 1.4480e-01,
  716. -1.0850e+00, 1.1459e+00, 1.9818e-01],
  717. [-2.2859e+00, -2.2859e+00, 1.8249e+00, -8.0015e-01, 4.0878e-02,
  718. -1.2543e+00, 8.0590e-01, 3.0505e-01]]], device='cuda:0')
  719. loss_train_step before backward: tensor(0.5289, device='cuda:0', grad_fn=<MseLossBackward>)
  720. loss_train_step after backward: tensor(0.5289, device='cuda:0', grad_fn=<MseLossBackward>)
  721. loss_train: 2.5755676329135895
  722. step: 5
  723. running loss: 0.5151135265827179
  724. Train Steps: 5/90 Loss: 0.5151 torch.Size([8, 600, 800])
  725. torch.Size([8, 8])
  726. tensor([[0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750],
  727. [0.6200, 0.3999, 0.8653, 0.5207, 0.4100, 0.5125, 0.5975, 0.5103],
  728. [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459],
  729. [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364],
  730. [0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162],
  731. [0.6104, 0.4029, 0.8738, 0.4900, 0.4088, 0.4533, 0.5070, 0.5510],
  732. [ nan, nan, 0.7240, 0.2722, 0.3900, 0.2567, 0.5168, 0.5933],
  733. [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717]],
  734. device='cuda:0', dtype=torch.float64)
  735. predictions are: tensor([[ 0.5403, -0.4755, 1.3845, -0.3674, -0.3488, -0.6346, 0.4944, 0.2667],
  736. [ 0.4828, -0.5086, 1.3642, -0.3836, -0.3648, -0.6673, 0.4748, 0.2593],
  737. [ 0.4669, -0.4383, 1.3275, -0.4923, -0.3288, -0.8227, 0.5028, 0.2269],
  738. [ 0.4408, -0.4837, 1.3020, -0.5519, -0.3340, -0.8562, 0.4644, 0.2288],
  739. [ 0.4844, -0.4597, 1.3352, -0.3919, -0.3426, -0.6659, 0.4823, 0.2951],
  740. [ 0.5130, -0.4774, 1.3368, -0.4231, -0.3680, -0.7182, 0.4799, 0.2842],
  741. [ 0.4028, -0.4552, 1.2646, -0.5310, -0.3368, -0.8135, 0.4102, 0.2107],
  742. [ 0.4311, -0.4643, 1.2980, -0.5692, -0.2955, -0.8370, 0.4452, 0.2128]],
  743. device='cuda:0', grad_fn=<AddmmBackward>)
  744. landmarks are: tensor([[[ 0.5525, -0.4463, 1.7326, 0.0313, -0.2536, -0.0688, 0.1968,
  745. 0.3700],
  746. [ 0.5778, -0.4389, 1.7107, 0.1192, -0.3921, 0.0815, 0.4741,
  747. 0.0711],
  748. [ 0.6520, -0.3912, 1.7095, -0.9079, -0.0284, -1.3621, 0.8096,
  749. 0.2356],
  750. [ 0.6158, -0.3960, 1.4092, -1.2774, -0.2074, -1.1851, 0.8491,
  751. 0.1917],
  752. [ 0.6059, -0.4442, 1.8921, -0.3422, -0.6230, -0.1461, 0.7489,
  753. 0.0985],
  754. [ 0.5334, -0.4252, 1.7499, -0.0226, -0.3979, -0.1920, 0.0558,
  755. 0.2589],
  756. [-2.2859, -2.2859, 1.0580, -1.0288, -0.4845, -1.1004, 0.1011,
  757. 0.4543],
  758. [ 0.5746, -0.3882, 1.1436, -1.2005, -0.4903, -1.0157, 0.4393,
  759. 0.3546]]], device='cuda:0')
  760. loss_train_step before backward: tensor(0.2597, device='cuda:0', grad_fn=<MseLossBackward>)
  761. loss_train_step after backward: tensor(0.2597, device='cuda:0', grad_fn=<MseLossBackward>)
  762. loss_train: 2.8352708220481873
  763. step: 6
  764. running loss: 0.4725451370080312
  765. Train Steps: 6/90 Loss: 0.4725 torch.Size([8, 600, 800])
  766. torch.Size([8, 8])
  767. tensor([[0.6289, 0.4032, 0.8419, 0.5446, 0.4075, 0.5017, 0.6312, 0.5117],
  768. [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973],
  769. [0.6236, 0.3967, 0.8675, 0.5400, 0.3862, 0.4517, 0.5825, 0.5200],
  770. [0.6201, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044],
  771. [0.6229, 0.4066, 0.8513, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350],
  772. [0.6043, 0.4022, 0.6887, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136],
  773. [0.6212, 0.4033, 0.8938, 0.4167, 0.3813, 0.4267, 0.5613, 0.5583],
  774. [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837]],
  775. device='cuda:0', dtype=torch.float64)
  776. predictions are: tensor([[ 0.4695, -0.5488, 1.5943, -0.4788, -0.3550, -0.6922, 0.5460, 0.2842],
  777. [ 0.4509, -0.5320, 1.5898, -0.3879, -0.3391, -0.6629, 0.5457, 0.2746],
  778. [ 0.4613, -0.5351, 1.5649, -0.4374, -0.3564, -0.6521, 0.5522, 0.2870],
  779. [ 0.3650, -0.5891, 1.4587, -0.6514, -0.3122, -0.8934, 0.5336, 0.2440],
  780. [ 0.4334, -0.5463, 1.5269, -0.4893, -0.3581, -0.7008, 0.5142, 0.2635],
  781. [ 0.3762, -0.5329, 1.4488, -0.6472, -0.3316, -0.8446, 0.5024, 0.2438],
  782. [ 0.4500, -0.5802, 1.5537, -0.4733, -0.2985, -0.6812, 0.5522, 0.2756],
  783. [ 0.3715, -0.5650, 1.4511, -0.6158, -0.3295, -0.7749, 0.4584, 0.2101]],
  784. device='cuda:0', grad_fn=<AddmmBackward>)
  785. landmarks are: tensor([[[ 0.6189, -0.4238, 1.6026, 0.2295, -0.4037, 0.0313, 0.6298,
  786. 0.0774],
  787. [ 0.6240, -0.4321, 1.8423, 0.1852, -0.5885, -0.1612, 0.6962,
  788. 0.0111],
  789. [ 0.5944, -0.4538, 1.7210, 0.2083, -0.5018, -0.1997, 0.4046,
  790. 0.1159],
  791. [ 0.5785, -0.4129, 1.2785, -1.3996, -0.3227, -1.3259, 0.4258,
  792. 0.0438],
  793. [ 0.5911, -0.4080, 1.6460, 0.3546, -0.2074, 0.0467, 0.4970,
  794. 0.1852],
  795. [ 0.5053, -0.4281, 0.8954, -1.3698, -0.5423, -1.1389, 0.2453,
  796. 0.0862],
  797. [ 0.5832, -0.4231, 1.8423, -0.3614, -0.5249, -0.3152, 0.3065,
  798. 0.2930],
  799. [ 0.5629, -0.3871, 0.7243, -0.9581, -0.5827, -0.9849, 0.1288,
  800. 0.4103]]], device='cuda:0')
  801. loss_train_step before backward: tensor(0.1155, device='cuda:0', grad_fn=<MseLossBackward>)
  802. loss_train_step after backward: tensor(0.1155, device='cuda:0', grad_fn=<MseLossBackward>)
  803. loss_train: 2.950811892747879
  804. step: 7
  805. running loss: 0.4215445561068399
  806. Train Steps: 7/90 Loss: 0.4215 torch.Size([8, 600, 800])
  807. torch.Size([8, 8])
  808. tensor([[0.6205, 0.4016, 0.8350, 0.2717, 0.3987, 0.2550, 0.5787, 0.5133],
  809. [ nan, nan, 0.8888, 0.3100, 0.5262, 0.2817, 0.7145, 0.6003],
  810. [0.6286, 0.4086, 0.8408, 0.2801, 0.4163, 0.2800, 0.6725, 0.5393],
  811. [0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950],
  812. [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274],
  813. [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332],
  814. [0.6164, 0.4076, 0.8838, 0.4117, 0.3713, 0.5550, 0.6238, 0.5350],
  815. [0.6357, 0.4097, 0.9038, 0.3883, 0.4213, 0.2950, 0.6686, 0.5390]],
  816. device='cuda:0', dtype=torch.float64)
  817. predictions are: tensor([[ 0.3134, -0.6794, 1.6194, -0.7281, -0.3209, -0.8706, 0.5486, 0.2226],
  818. [ 0.3383, -0.6451, 1.6516, -0.6063, -0.3142, -0.8186, 0.5737, 0.2116],
  819. [ 0.2937, -0.6742, 1.6026, -0.7307, -0.3141, -0.8592, 0.5482, 0.2223],
  820. [ 0.4637, -0.6103, 1.7207, -0.4602, -0.3870, -0.5815, 0.5783, 0.2602],
  821. [ 0.4362, -0.6387, 1.6842, -0.5264, -0.3400, -0.6412, 0.5542, 0.2823],
  822. [ 0.5325, -0.5639, 1.7223, -0.4289, -0.3985, -0.5341, 0.5724, 0.2672],
  823. [ 0.4656, -0.5311, 1.6839, -0.4444, -0.3957, -0.5205, 0.5554, 0.2717],
  824. [ 0.4447, -0.5605, 1.6902, -0.5166, -0.4024, -0.6472, 0.5584, 0.2629]],
  825. device='cuda:0', grad_fn=<AddmmBackward>)
  826. landmarks are: tensor([[[ 5.7997e-01, -4.3118e-01, 1.5709e+00, -1.0311e+00, -4.4411e-01,
  827. -1.1081e+00, 3.8730e-01, 8.5142e-02],
  828. [-2.2859e+00, -2.2859e+00, 1.8192e+00, -8.5404e-01, 1.4480e-01,
  829. -9.8491e-01, 1.0143e+00, 4.8673e-01],
  830. [ 6.1742e-01, -3.9842e-01, 1.5975e+00, -9.9214e-01, -3.6328e-01,
  831. -9.9261e-01, 8.2047e-01, 2.0505e-01],
  832. [ 5.8199e-01, -4.7544e-01, 1.7095e+00, 1.9292e-01, -2.5358e-01,
  833. 5.4350e-02, 4.9700e-01, 4.6189e-04],
  834. [ 6.1083e-01, -4.0082e-01, 1.9088e+00, -2.5294e-02, -5.7691e-01,
  835. -3.0747e-01, 7.9054e-01, 1.4989e-01],
  836. [ 6.5201e-01, -4.0564e-01, 1.9173e+00, -7.6520e-02, -5.5958e-01,
  837. -4.5373e-01, 7.9493e-01, 1.7680e-01],
  838. [ 5.6143e-01, -4.0323e-01, 1.7961e+00, -3.8445e-01, -5.7113e-01,
  839. 2.7760e-01, 5.9515e-01, 1.8522e-01],
  840. [ 6.5036e-01, -3.9360e-01, 1.8885e+00, -4.9222e-01, -3.4018e-01,
  841. -9.2333e-01, 8.0224e-01, 2.0352e-01]]], device='cuda:0')
  842. loss_train_step before backward: tensor(0.2142, device='cuda:0', grad_fn=<MseLossBackward>)
  843. loss_train_step after backward: tensor(0.2142, device='cuda:0', grad_fn=<MseLossBackward>)
  844. loss_train: 3.1650566458702087
  845. step: 8
  846. running loss: 0.3956320807337761
  847.  
  848. Train Steps: 8/90 Loss: 0.3956 torch.Size([8, 600, 800])
  849. torch.Size([8, 8])
  850. tensor([[0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212],
  851. [0.6361, 0.4076, 0.8862, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297],
  852. [0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567],
  853. [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817],
  854. [0.6122, 0.3993, 0.8738, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927],
  855. [0.6248, 0.4032, 0.7738, 0.1900, 0.4813, 0.1400, 0.5941, 0.4904],
  856. [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5837, 0.5500],
  857. [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5963, 0.6217]],
  858. device='cuda:0', dtype=torch.float64)
  859. predictions are: tensor([[ 0.3986, -0.6094, 1.7731, -0.5675, -0.4028, -0.6541, 0.6191, 0.1876],
  860. [ 0.5058, -0.5929, 1.8639, -0.3546, -0.4482, -0.4050, 0.6035, 0.2369],
  861. [ 0.3617, -0.6422, 1.7939, -0.5044, -0.3761, -0.5660, 0.5857, 0.2077],
  862. [ 0.2151, -0.7175, 1.6810, -0.7442, -0.3176, -0.7890, 0.5508, 0.1978],
  863. [ 0.4400, -0.6495, 1.8597, -0.4622, -0.3749, -0.5228, 0.6240, 0.2258],
  864. [ 0.2712, -0.7017, 1.7019, -0.7678, -0.3379, -0.8403, 0.5856, 0.1985],
  865. [ 0.3907, -0.6001, 1.8238, -0.4108, -0.4309, -0.5027, 0.5874, 0.2351],
  866. [ 0.3679, -0.6428, 1.7431, -0.5741, -0.3706, -0.6884, 0.6126, 0.1843]],
  867. device='cuda:0', grad_fn=<AddmmBackward>)
  868. landmarks are: tensor([[[ 0.6125, -0.4345, 1.4308, -1.1384, -0.4213, -1.0031, 0.7190,
  869. 0.1214],
  870. [ 0.6520, -0.4032, 1.8076, 0.1852, -0.5711, -0.1381, 0.7876,
  871. 0.1608],
  872. [ 0.5538, -0.4139, 1.7557, -0.1843, -0.0459, 0.1242, 0.4219,
  873. 0.2853],
  874. [ 0.5964, -0.3380, 1.4785, -0.8386, -0.2420, -1.0619, 0.3238,
  875. 0.4008],
  876. [ 0.5417, -0.4417, 1.7499, -0.1304, -0.1994, -0.0324, 0.0951,
  877. -0.0099],
  878. [ 0.5999, -0.4236, 1.2880, -1.4083, -0.0630, -1.6393, 0.4584,
  879. -0.0208],
  880. [ 0.5604, -0.4620, 1.7961, -0.1997, -0.3229, 0.1082, 0.4104,
  881. 0.2545],
  882. [ 0.5687, -0.4088, 1.0397, -1.1466, -0.3113, -1.1928, 0.4681,
  883. 0.5855]]], device='cuda:0')
  884. loss_train_step before backward: tensor(0.1039, device='cuda:0', grad_fn=<MseLossBackward>)
  885. loss_train_step after backward: tensor(0.1039, device='cuda:0', grad_fn=<MseLossBackward>)
  886. loss_train: 3.268991008400917
  887. step: 9
  888. running loss: 0.36322122315565747
  889. Train Steps: 9/90 Loss: 0.3632 torch.Size([8, 600, 800])
  890. torch.Size([8, 8])
  891. tensor([[0.6164, 0.4102, 0.8850, 0.4867, 0.4213, 0.5983, 0.5609, 0.5038],
  892. [0.6268, 0.4102, 0.8938, 0.3667, 0.4025, 0.2833, 0.6275, 0.5183],
  893. [0.6284, 0.4127, 0.8538, 0.5867, 0.4363, 0.5083, 0.6038, 0.5433],
  894. [0.6255, 0.4017, 0.8688, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901],
  895. [0.6200, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148],
  896. [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377],
  897. [0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6362, 0.5200],
  898. [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116]],
  899. device='cuda:0', dtype=torch.float64)
  900. predictions are: tensor([[ 0.4023, -0.5970, 1.8370, -0.4402, -0.3702, -0.4494, 0.5963, 0.2152],
  901. [ 0.2735, -0.7746, 1.8249, -0.7597, -0.3093, -0.8099, 0.6603, 0.1850],
  902. [ 0.3854, -0.6540, 1.8492, -0.5377, -0.3915, -0.5254, 0.5822, 0.1906],
  903. [ 0.3886, -0.5986, 1.8082, -0.5924, -0.3838, -0.5982, 0.6233, 0.1837],
  904. [ 0.3802, -0.6962, 1.8287, -0.5726, -0.3557, -0.6003, 0.5991, 0.2111],
  905. [ 0.4066, -0.6400, 1.8792, -0.4734, -0.3624, -0.4860, 0.6284, 0.1999],
  906. [ 0.4266, -0.5321, 1.8191, -0.5020, -0.4016, -0.4485, 0.5741, 0.2038],
  907. [ 0.3916, -0.6078, 1.8294, -0.5132, -0.4115, -0.4810, 0.5802, 0.2019]],
  908. device='cuda:0', grad_fn=<AddmmBackward>)
  909. landmarks are: tensor([[[ 0.5614, -0.3912, 1.8018, -0.0380, -0.3402, 0.4778, 0.3049,
  910. 0.0412],
  911. [ 0.6092, -0.3913, 1.8423, -0.5923, -0.4268, -0.9772, 0.6125,
  912. 0.1082],
  913. [ 0.6166, -0.3795, 1.6575, 0.4239, -0.2709, 0.0620, 0.5028,
  914. 0.2237],
  915. [ 0.6031, -0.4307, 1.7268, -0.8002, -0.6058, -0.6462, 0.6442,
  916. -0.0220],
  917. [ 0.5781, -0.4205, 1.8159, -0.0691, -0.6116, -0.3069, 0.4236,
  918. 0.0919],
  919. [ 0.5775, -0.4607, 1.6741, 0.1962, -0.4036, -0.1212, 0.4588,
  920. 0.1979],
  921. [ 0.5841, -0.4199, 1.7961, -0.6693, -0.6231, 0.1082, 0.6529,
  922. 0.1159],
  923. [ 0.5376, -0.3903, 1.7095, -0.3229, -0.4730, 0.4701, 0.3871,
  924. 0.0772]]], device='cuda:0')
  925. loss_train_step before backward: tensor(0.0941, device='cuda:0', grad_fn=<MseLossBackward>)
  926. loss_train_step after backward: tensor(0.0941, device='cuda:0', grad_fn=<MseLossBackward>)
  927. loss_train: 3.3630512952804565
  928. step: 10
  929. running loss: 0.33630512952804564
  930. Train Steps: 10/90 Loss: 0.3363 torch.Size([8, 600, 800])
  931. torch.Size([8, 8])
  932. tensor([[0.6156, 0.4125, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084],
  933. [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600],
  934. [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837],
  935. [0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567],
  936. [0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461],
  937. [0.6250, 0.3993, 0.9138, 0.4333, 0.3763, 0.5217, 0.6995, 0.5320],
  938. [0.6259, 0.4156, 0.8812, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960],
  939. [0.6310, 0.4017, 0.8563, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006]],
  940. device='cuda:0', dtype=torch.float64)
  941. predictions are: tensor([[ 0.4281, -0.6018, 1.9033, -0.4220, -0.3854, -0.3424, 0.6084, 0.1981],
  942. [ 0.4875, -0.5669, 1.9027, -0.3804, -0.4182, -0.3505, 0.6296, 0.1911],
  943. [ 0.3455, -0.6332, 1.7253, -0.7356, -0.3643, -0.6066, 0.5329, 0.1503],
  944. [ 0.3817, -0.6205, 1.8508, -0.5359, -0.3424, -0.4687, 0.5927, 0.1690],
  945. [ 0.4571, -0.5370, 1.8616, -0.5107, -0.3985, -0.4145, 0.6011, 0.1742],
  946. [ 0.5165, -0.5026, 1.9352, -0.3225, -0.4410, -0.2379, 0.6228, 0.1941],
  947. [ 0.2141, -0.7852, 1.7427, -0.8960, -0.2852, -0.8919, 0.6142, 0.1452],
  948. [ 0.5665, -0.5681, 1.9401, -0.3287, -0.4210, -0.2708, 0.6404, 0.1673]],
  949. device='cuda:0', grad_fn=<AddmmBackward>)
  950. landmarks are: tensor([[[ 0.5573, -0.3808, 1.8018, -0.0534, -0.1843, 0.4008, 0.3769,
  951. 0.0622],
  952. [ 0.5883, -0.3594, 1.7557, 0.2545, -0.4152, -0.0611, 0.3353,
  953. 0.3007],
  954. [ 0.5629, -0.3871, 0.7243, -0.9581, -0.5827, -0.9849, 0.1288,
  955. 0.4103],
  956. [ 0.5538, -0.4139, 1.7557, -0.1843, -0.0459, 0.1242, 0.4219,
  957. 0.2853],
  958. [ 0.5124, -0.4483, 1.5420, -0.6231, -0.7155, -0.4152, 0.2679,
  959. 0.2365],
  960. [ 0.6010, -0.4417, 1.9346, -0.2844, -0.5480, 0.1236, 0.9448,
  961. 0.1715],
  962. [ 0.6052, -0.3663, 1.7845, -0.8156, -0.0804, -1.4237, 0.5866,
  963. 0.0051],
  964. [ 0.6286, -0.4303, 1.6691, 0.3931, -0.5827, -0.0919, 0.6359,
  965. 0.0261]]], device='cuda:0')
  966. loss_train_step before backward: tensor(0.0903, device='cuda:0', grad_fn=<MseLossBackward>)
  967. loss_train_step after backward: tensor(0.0903, device='cuda:0', grad_fn=<MseLossBackward>)
  968. loss_train: 3.453396238386631
  969. step: 11
  970. running loss: 0.3139451125806028
  971. Train Steps: 11/90 Loss: 0.3139 torch.Size([8, 600, 800])
  972. torch.Size([8, 8])
  973. tensor([[0.6202, 0.4064, 0.7879, 0.2179, 0.4567, 0.1725, 0.5955, 0.5478],
  974. [0.6132, 0.4037, 0.6963, 0.2217, 0.4100, 0.1950, 0.5395, 0.5175],
  975. [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649],
  976. [0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162],
  977. [0.6127, 0.4066, 0.8550, 0.5567, 0.4662, 0.5141, 0.5070, 0.5412],
  978. [0.6179, 0.4118, 0.7278, 0.4237, 0.3588, 0.3400, 0.5675, 0.5917],
  979. [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773],
  980. [0.6126, 0.3954, 0.8538, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350]],
  981. device='cuda:0', dtype=torch.float64)
  982. predictions are: tensor([[ 0.3583, -0.6343, 1.7170, -0.7579, -0.3092, -0.6544, 0.5731, 0.1568],
  983. [ 0.3602, -0.6007, 1.7199, -0.7404, -0.3476, -0.6163, 0.5342, 0.1324],
  984. [ 0.5714, -0.4951, 1.9335, -0.2031, -0.4453, -0.1089, 0.6063, 0.1638],
  985. [ 0.6233, -0.4319, 1.9334, -0.2226, -0.4486, -0.0354, 0.5995, 0.1807],
  986. [ 0.4814, -0.5525, 1.8654, -0.3785, -0.4045, -0.2285, 0.5742, 0.1675],
  987. [ 0.4247, -0.6203, 1.7907, -0.5895, -0.3542, -0.4821, 0.5728, 0.1632],
  988. [ 0.4121, -0.5504, 1.7687, -0.6334, -0.3926, -0.4873, 0.5501, 0.1550],
  989. [ 0.5594, -0.5201, 1.9404, -0.2369, -0.3908, -0.1411, 0.6089, 0.1694]],
  990. device='cuda:0', grad_fn=<AddmmBackward>)
  991. landmarks are: tensor([[[ 0.5787, -0.4086, 1.3535, -1.2794, -0.1764, -1.4891, 0.4645,
  992. 0.2442],
  993. [ 0.5465, -0.4214, 0.9300, -1.2620, -0.3921, -1.3852, 0.2062,
  994. 0.1043],
  995. [ 0.6105, -0.4360, 1.6171, 0.5162, -0.5076, 0.0159, 0.5171,
  996. -0.1385],
  997. [ 0.6059, -0.4442, 1.8921, -0.3422, -0.6230, -0.1461, 0.7489,
  998. 0.0985],
  999. [ 0.5443, -0.4079, 1.6633, 0.2853, -0.1325, 0.0888, 0.0558,
  1000. 0.2138],
  1001. [ 0.5680, -0.3840, 1.0756, -0.3290, -0.6289, -0.7155, 0.3353,
  1002. 0.4470],
  1003. [ 0.5841, -0.3574, 1.0859, -0.9541, -0.2824, -1.2851, 0.3460,
  1004. 0.3808],
  1005. [ 0.5435, -0.4597, 1.6575, 0.0159, -0.3229, -0.1150, 0.1877,
  1006. 0.1854]]], device='cuda:0')
  1007. loss_train_step before backward: tensor(0.1143, device='cuda:0', grad_fn=<MseLossBackward>)
  1008. loss_train_step after backward: tensor(0.1143, device='cuda:0', grad_fn=<MseLossBackward>)
  1009. loss_train: 3.5676762238144875
  1010. step: 12
  1011. running loss: 0.29730635198454064
  1012.  
  1013. Train Steps: 12/90 Loss: 0.2973 torch.Size([8, 600, 800])
  1014. torch.Size([8, 8])
  1015. tensor([[0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650],
  1016. [0.6233, 0.4091, 0.8100, 0.2950, 0.3563, 0.3883, 0.6013, 0.5200],
  1017. [0.6098, 0.3991, 0.8638, 0.4717, 0.4263, 0.4967, 0.5212, 0.5650],
  1018. [0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090],
  1019. [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402],
  1020. [0.6189, 0.4029, 0.8375, 0.5767, 0.4745, 0.4829, 0.5551, 0.5598],
  1021. [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167],
  1022. [0.6069, 0.3975, 0.8625, 0.5083, 0.4388, 0.5483, 0.5650, 0.4967]],
  1023. device='cuda:0', dtype=torch.float64)
  1024. predictions are: tensor([[ 0.5343, -0.4726, 1.7997, -0.3158, -0.3930, -0.2037, 0.5199, 0.1621],
  1025. [ 0.5070, -0.4471, 1.7056, -0.5312, -0.4011, -0.3798, 0.4930, 0.1416],
  1026. [ 0.5230, -0.4723, 1.8404, -0.3098, -0.4120, -0.1843, 0.5487, 0.1675],
  1027. [ 0.5441, -0.4712, 1.8092, -0.4498, -0.4106, -0.3381, 0.5668, 0.1592],
  1028. [ 0.5783, -0.4525, 1.8416, -0.2854, -0.3991, -0.1710, 0.5524, 0.1664],
  1029. [ 0.4761, -0.5077, 1.7601, -0.4455, -0.3867, -0.3223, 0.5207, 0.1484],
  1030. [ 0.4512, -0.5921, 1.7266, -0.6192, -0.3187, -0.5162, 0.4994, 0.1540],
  1031. [ 0.5362, -0.4790, 1.8213, -0.3137, -0.4172, -0.1955, 0.5498, 0.1414]],
  1032. device='cuda:0', grad_fn=<AddmmBackward>)
  1033. landmarks are: tensor([[[ 0.5770, -0.4036, 1.7095, 0.3084, -0.3691, -0.2690, 0.2314,
  1034. 0.3238],
  1035. [ 0.5932, -0.3962, 1.4554, -0.9233, -0.6404, -0.4922, 0.4912,
  1036. 0.1159],
  1037. [ 0.5308, -0.4425, 1.7037, -0.1073, -0.3171, 0.0082, 0.1217,
  1038. 0.3238],
  1039. [ 0.6165, -0.4249, 1.7961, -0.5384, -0.4499, -0.9695, 0.6401,
  1040. 0.0652],
  1041. [ 0.5780, -0.4565, 1.6221, 0.2532, -0.3728, -0.1718, 0.4357,
  1042. 0.2091],
  1043. [ 0.5726, -0.4249, 1.5824, 0.3777, -0.0942, -0.0556, 0.2782,
  1044. 0.2997],
  1045. [ 0.5768, -0.3899, 1.3861, -0.7771, -0.5885, -0.5461, 0.5028,
  1046. 0.5624],
  1047. [ 0.5174, -0.4497, 1.6979, 0.0620, -0.2594, 0.2468, 0.3238,
  1048. 0.0082]]], device='cuda:0')
  1049. loss_train_step before backward: tensor(0.0588, device='cuda:0', grad_fn=<MseLossBackward>)
  1050. loss_train_step after backward: tensor(0.0588, device='cuda:0', grad_fn=<MseLossBackward>)
  1051. loss_train: 3.626456379890442
  1052. step: 13
  1053. running loss: 0.2789581830684955
  1054. Train Steps: 13/90 Loss: 0.2790 torch.Size([8, 600, 800])
  1055. torch.Size([8, 8])
  1056. tensor([[0.6200, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391],
  1057. [0.6263, 0.4065, 0.9038, 0.4317, 0.3588, 0.4550, 0.6325, 0.5250],
  1058. [0.6277, 0.4103, 0.8087, 0.5717, 0.4188, 0.4750, 0.5663, 0.6083],
  1059. [0.6346, 0.4086, 0.7938, 0.5500, 0.3962, 0.4867, 0.7343, 0.5702],
  1060. [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5413, 0.5717],
  1061. [0.6183, 0.4076, 0.8838, 0.4517, 0.3813, 0.4483, 0.5775, 0.5633],
  1062. [0.6204, 0.4049, 0.7975, 0.2700, 0.3937, 0.2567, 0.5700, 0.5183],
  1063. [0.6124, 0.4069, 0.8314, 0.5001, 0.3738, 0.4650, 0.5167, 0.5402]],
  1064. device='cuda:0', dtype=torch.float64)
  1065. predictions are: tensor([[ 0.3386, -0.5945, 1.4882, -0.8442, -0.2679, -0.8243, 0.4488, 0.1746],
  1066. [ 0.6488, -0.4722, 1.7840, -0.1984, -0.4139, -0.1456, 0.5088, 0.2118],
  1067. [ 0.6712, -0.3812, 1.7589, -0.1449, -0.4616, -0.0147, 0.4819, 0.1966],
  1068. [ 0.5852, -0.3525, 1.7167, -0.2342, -0.4497, -0.0917, 0.4490, 0.1965],
  1069. [ 0.6722, -0.3628, 1.7972, -0.1442, -0.4551, -0.0454, 0.5214, 0.2173],
  1070. [ 0.6903, -0.3728, 1.7680, -0.1855, -0.4479, -0.0402, 0.4770, 0.2045],
  1071. [ 0.4033, -0.5384, 1.5374, -0.7993, -0.3055, -0.7459, 0.4317, 0.1558],
  1072. [ 0.5944, -0.3282, 1.7185, -0.2286, -0.4429, -0.0724, 0.4695, 0.1935]],
  1073. device='cuda:0', grad_fn=<AddmmBackward>)
  1074. landmarks are: tensor([[[ 0.5776, -0.3987, 1.6005, -0.9121, -0.3844, -1.2358, 0.4247,
  1075. 0.2043],
  1076. [ 0.6070, -0.4085, 1.8885, -0.2921, -0.6289, -0.1843, 0.6356,
  1077. 0.1390],
  1078. [ 0.6134, -0.3910, 1.4497, 0.3546, -0.3517, -0.0919, 0.3296,
  1079. 0.5239],
  1080. [ 0.6454, -0.3984, 1.3804, 0.2545, -0.4557, -0.0380, 1.1057,
  1081. 0.3478],
  1082. [ 0.5432, -0.4462, 1.7557, -0.0380, -0.3113, -0.0765, 0.2141,
  1083. 0.3546],
  1084. [ 0.5700, -0.4032, 1.7961, -0.1997, -0.5249, -0.2151, 0.3815,
  1085. 0.3161],
  1086. [ 0.5798, -0.4156, 1.3977, -1.0388, -0.4672, -1.1004, 0.3469,
  1087. 0.1082],
  1088. [ 0.5425, -0.4067, 1.5543, 0.0241, -0.5596, -0.1381, 0.1005,
  1089. 0.2093]]], device='cuda:0')
  1090. loss_train_step before backward: tensor(0.0389, device='cuda:0', grad_fn=<MseLossBackward>)
  1091. loss_train_step after backward: tensor(0.0389, device='cuda:0', grad_fn=<MseLossBackward>)
  1092. loss_train: 3.6653256751596928
  1093. step: 14
  1094. running loss: 0.2618089767971209
  1095. Train Steps: 14/90 Loss: 0.2618 torch.Size([8, 600, 800])
  1096. torch.Size([8, 8])
  1097. tensor([[0.6346, 0.4144, 0.9088, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899],
  1098. [0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272],
  1099. [0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567],
  1100. [0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533],
  1101. [0.6223, 0.4028, 0.8988, 0.4200, 0.3763, 0.5733, 0.6375, 0.5167],
  1102. [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378],
  1103. [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389],
  1104. [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901]],
  1105. device='cuda:0', dtype=torch.float64)
  1106. predictions are: tensor([[ 0.6552, -0.3837, 1.7376, -0.1305, -0.4385, -0.1260, 0.4828, 0.2422],
  1107. [ 0.5077, -0.4758, 1.5026, -0.5479, -0.3470, -0.5731, 0.4024, 0.2062],
  1108. [ 0.6341, -0.3702, 1.6776, -0.1107, -0.4211, -0.0452, 0.4231, 0.2498],
  1109. [ 0.6070, -0.3405, 1.5338, -0.4015, -0.4672, -0.2579, 0.4081, 0.2168],
  1110. [ 0.8312, -0.2328, 1.8114, 0.1777, -0.5426, 0.2986, 0.4698, 0.2533],
  1111. [ 0.7075, -0.2959, 1.7456, 0.0659, -0.5056, 0.1518, 0.4310, 0.2443],
  1112. [ 0.4628, -0.4883, 1.4341, -0.7040, -0.3237, -0.7027, 0.4005, 0.2153],
  1113. [ 0.4701, -0.4907, 1.4618, -0.6527, -0.3600, -0.6216, 0.3729, 0.2062]],
  1114. device='cuda:0', grad_fn=<AddmmBackward>)
  1115. landmarks are: tensor([[[ 0.6454, -0.3719, 1.9115, -0.1304, -0.5076, -0.2844, 1.0033,
  1116. 0.4386],
  1117. [ 0.5711, -0.4015, 1.6979, -0.6770, -0.5365, -1.0619, 0.1712,
  1118. 0.1494],
  1119. [ 0.5538, -0.4139, 1.7557, -0.1843, -0.0459, 0.1242, 0.4219,
  1120. 0.2853],
  1121. [ 0.5878, -0.4052, 1.0229, -1.2855, -0.5596, -0.8232, 0.5316,
  1122. 0.2699],
  1123. [ 0.5886, -0.4253, 1.8654, -0.3460, -0.5480, 0.3623, 0.6587,
  1124. 0.1005],
  1125. [ 0.6141, -0.4345, 1.6864, -0.0303, -0.2882, 0.1544, 0.9521,
  1126. 0.1982],
  1127. [ 0.5697, -0.4466, 1.1973, -1.1871, -0.4571, -0.9965, 0.5219,
  1128. 0.2032],
  1129. [ 0.5515, -0.4129, 1.4785, -1.0080, -0.5192, -1.1004, 0.1034,
  1130. -0.0220]]], device='cuda:0')
  1131. loss_train_step before backward: tensor(0.0639, device='cuda:0', grad_fn=<MseLossBackward>)
  1132. loss_train_step after backward: tensor(0.0639, device='cuda:0', grad_fn=<MseLossBackward>)
  1133. loss_train: 3.729193616658449
  1134. step: 15
  1135. running loss: 0.24861290777722994
  1136. Train Steps: 15/90 Loss: 0.2486 torch.Size([8, 600, 800])
  1137. torch.Size([8, 8])
  1138. tensor([[0.6102, 0.4020, 0.8638, 0.3717, 0.3625, 0.5017, 0.6038, 0.5500],
  1139. [0.6371, 0.4092, 0.8337, 0.5850, 0.3950, 0.5117, 0.6559, 0.5262],
  1140. [ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600],
  1141. [0.6277, 0.4083, 0.8350, 0.2717, 0.4562, 0.1800, 0.5918, 0.4878],
  1142. [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667],
  1143. [0.6203, 0.4073, 0.8189, 0.2398, 0.4400, 0.2054, 0.5929, 0.5501],
  1144. [0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240],
  1145. [0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136]],
  1146. device='cuda:0', dtype=torch.float64)
  1147. predictions are: tensor([[ 0.8263, -0.1902, 1.7374, 0.2545, -0.5898, 0.3922, 0.4587, 0.2920],
  1148. [ 0.8305, -0.2324, 1.7869, 0.3114, -0.5759, 0.3759, 0.4640, 0.2764],
  1149. [ 0.5266, -0.4159, 1.3851, -0.5850, -0.3763, -0.5650, 0.3748, 0.2421],
  1150. [ 0.4363, -0.4686, 1.3255, -0.7476, -0.3138, -0.7884, 0.3587, 0.2176],
  1151. [ 0.7096, -0.3306, 1.6011, -0.1989, -0.4514, -0.1596, 0.4091, 0.2694],
  1152. [ 0.5078, -0.4304, 1.3690, -0.6331, -0.3530, -0.6531, 0.3775, 0.2460],
  1153. [ 0.5176, -0.3927, 1.3475, -0.6248, -0.3892, -0.6089, 0.3561, 0.2448],
  1154. [ 0.6879, -0.3213, 1.6219, -0.0797, -0.4829, -0.0799, 0.4250, 0.2820]],
  1155. device='cuda:0', grad_fn=<AddmmBackward>)
  1156. landmarks are: tensor([[[ 0.5324, -0.4293, 1.7037, -0.5692, -0.6115, 0.0313, 0.5028,
  1157. 0.2545],
  1158. [ 0.6569, -0.3960, 1.5651, 0.4162, -0.4614, 0.0774, 0.7438,
  1159. 0.1447],
  1160. [-2.2859, -2.2859, 1.4266, -1.1389, -0.4499, -1.1235, 0.2891,
  1161. 0.3007],
  1162. [ 0.6132, -0.4002, 1.5709, -1.0311, -0.1785, -1.4545, 0.4474,
  1163. -0.0328],
  1164. [ 0.5768, -0.3857, 1.5305, -0.7694, -0.6462, -0.6308, 0.3988,
  1165. 0.3315],
  1166. [ 0.5792, -0.4048, 1.4965, -1.1781, -0.2534, -1.3371, 0.4528,
  1167. 0.2549],
  1168. [ 0.5676, -0.4112, 1.1898, -1.2467, -0.2940, -1.4622, 0.2103,
  1169. 0.1343],
  1170. [ 0.5235, -0.4273, 1.7499, -0.4306, -0.5827, -0.4614, 0.1651,
  1171. 0.0862]]], device='cuda:0')
  1172. loss_train_step before backward: tensor(0.2667, device='cuda:0', grad_fn=<MseLossBackward>)
  1173. loss_train_step after backward: tensor(0.2667, device='cuda:0', grad_fn=<MseLossBackward>)
  1174. loss_train: 3.9958606250584126
  1175. step: 16
  1176. running loss: 0.24974128906615078
  1177.  
  1178. Train Steps: 16/90 Loss: 0.2497 torch.Size([8, 600, 800])
  1179. torch.Size([8, 8])
  1180. tensor([[0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578],
  1181. [0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904],
  1182. [0.6256, 0.4199, 0.8638, 0.5800, 0.3987, 0.4383, 0.5600, 0.5950],
  1183. [0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136],
  1184. [0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162],
  1185. [0.6161, 0.4099, 0.8738, 0.4383, 0.3788, 0.5483, 0.5605, 0.5019],
  1186. [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400],
  1187. [0.6187, 0.4104, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683]],
  1188. device='cuda:0', dtype=torch.float64)
  1189. predictions are: tensor([[ 0.5092, -0.4276, 1.2993, -0.6912, -0.3900, -0.7287, 0.3428, 0.2390],
  1190. [ 0.4638, -0.5024, 1.3391, -0.7674, -0.3217, -0.9413, 0.3813, 0.2378],
  1191. [ 0.7581, -0.2839, 1.6582, 0.2064, -0.5467, 0.2310, 0.4485, 0.2857],
  1192. [ 0.6093, -0.3809, 1.5054, -0.2810, -0.4603, -0.3393, 0.3837, 0.2761],
  1193. [ 0.7945, -0.2427, 1.6784, 0.1989, -0.5599, 0.2828, 0.4505, 0.2914],
  1194. [ 0.6689, -0.2769, 1.5637, 0.0263, -0.5515, 0.1618, 0.3862, 0.2881],
  1195. [ 0.5313, -0.4060, 1.3328, -0.6785, -0.4032, -0.7108, 0.3818, 0.2442],
  1196. [ 0.5118, -0.4049, 1.3256, -0.6318, -0.4154, -0.6396, 0.3457, 0.2628]],
  1197. device='cuda:0', grad_fn=<AddmmBackward>)
  1198. landmarks are: tensor([[[ 0.5232, -0.4189, 0.9358, -1.4006, -0.3921, -1.3698, 0.2555,
  1199. 0.2906],
  1200. [ 0.6126, -0.3871, 1.6229, -1.0773, -0.2132, -1.3698, 0.5829,
  1201. -0.0209],
  1202. [ 0.6038, -0.3464, 1.7037, 0.3931, -0.4441, -0.2613, 0.3007,
  1203. 0.4624],
  1204. [ 0.5235, -0.4273, 1.7499, -0.4306, -0.5827, -0.4614, 0.1651,
  1205. 0.0862],
  1206. [ 0.6059, -0.4442, 1.8921, -0.3422, -0.6230, -0.1461, 0.7489,
  1207. 0.0985],
  1208. [ 0.5597, -0.3928, 1.7499, -0.2613, -0.5365, 0.2468, 0.3028,
  1209. 0.0321],
  1210. [ 0.5842, -0.3848, 1.2938, -1.2159, -0.2132, -1.4160, 0.5778,
  1211. 0.2083],
  1212. [ 0.5718, -0.3905, 1.0053, -1.3305, -0.4614, -1.1235, 0.4450,
  1213. 0.3392]]], device='cuda:0')
  1214. loss_train_step before backward: tensor(0.0752, device='cuda:0', grad_fn=<MseLossBackward>)
  1215. loss_train_step after backward: tensor(0.0752, device='cuda:0', grad_fn=<MseLossBackward>)
  1216. loss_train: 4.071053955703974
  1217. step: 17
  1218. running loss: 0.23947376210023374
  1219. Train Steps: 17/90 Loss: 0.2395 torch.Size([8, 600, 800])
  1220. torch.Size([8, 8])
  1221. tensor([[0.6085, 0.4008, 0.8588, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283],
  1222. [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834],
  1223. [0.6135, 0.3994, 0.7913, 0.3050, 0.3625, 0.3050, 0.5837, 0.5050],
  1224. [0.6127, 0.4084, 0.8700, 0.4467, 0.3987, 0.4317, 0.5013, 0.5471],
  1225. [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167],
  1226. [0.6034, 0.4011, 0.7350, 0.2533, 0.3438, 0.3367, 0.5516, 0.5084],
  1227. [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466],
  1228. [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600]],
  1229. device='cuda:0', dtype=torch.float64)
  1230. predictions are: tensor([[ 0.6578, -0.3683, 1.5521, -0.1059, -0.4751, -0.2149, 0.4410, 0.2723],
  1231. [ 0.5114, -0.4456, 1.2885, -0.7996, -0.3856, -0.9587, 0.4281, 0.2349],
  1232. [ 0.5884, -0.3397, 1.3325, -0.5015, -0.4970, -0.4707, 0.4000, 0.2406],
  1233. [ 0.6309, -0.3858, 1.5195, -0.1167, -0.5022, -0.1658, 0.3629, 0.2780],
  1234. [ 0.5000, -0.4365, 1.3342, -0.7227, -0.3862, -0.8498, 0.3772, 0.2283],
  1235. [ 0.5706, -0.3431, 1.3098, -0.5633, -0.5075, -0.5184, 0.3733, 0.2563],
  1236. [ 0.5632, -0.3163, 1.3676, -0.3814, -0.5250, -0.3097, 0.3615, 0.2432],
  1237. [ 0.5914, -0.3618, 1.5231, -0.0976, -0.4550, -0.1584, 0.4225, 0.2744]],
  1238. device='cuda:0', grad_fn=<AddmmBackward>)
  1239. landmarks are: tensor([[[ 0.5245, -0.4347, 1.6806, 0.1159, 0.0046, 0.0129, 0.1044,
  1240. 0.1544],
  1241. [ 0.6126, -0.4057, 1.4439, -1.3159, -0.1150, -1.5777, 0.5537,
  1242. -0.0530],
  1243. [ 0.5477, -0.4413, 1.3688, -0.8771, -0.6115, -0.8771, 0.4104,
  1244. 0.0467],
  1245. [ 0.5441, -0.3997, 1.7326, -0.2228, -0.4441, -0.2921, 0.0296,
  1246. 0.2409],
  1247. [ 0.5483, -0.4249, 1.5305, -0.7386, -0.6115, -0.8694, 0.3353,
  1248. 0.1005],
  1249. [ 0.5009, -0.4333, 1.1090, -1.1158, -0.6982, -0.7309, 0.2617,
  1250. 0.0622],
  1251. [ 0.5076, -0.4443, 1.2337, -0.5023, -0.6808, -0.3614, 0.0866,
  1252. 0.2386],
  1253. [ 0.5652, -0.4325, 1.7388, -0.1429, -0.0688, 0.3469, 0.4970,
  1254. 0.3007]]], device='cuda:0')
  1255. loss_train_step before backward: tensor(0.0491, device='cuda:0', grad_fn=<MseLossBackward>)
  1256. loss_train_step after backward: tensor(0.0491, device='cuda:0', grad_fn=<MseLossBackward>)
  1257. loss_train: 4.120140295475721
  1258. step: 18
  1259. running loss: 0.22889668308198452
  1260. Train Steps: 18/90 Loss: 0.2289 torch.Size([8, 600, 800])
  1261. torch.Size([8, 8])
  1262. tensor([[0.6252, 0.4158, 0.8988, 0.4083, 0.3788, 0.4783, 0.6225, 0.5633],
  1263. [0.6250, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6088, 0.5183],
  1264. [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183],
  1265. [0.6137, 0.4038, 0.8563, 0.4050, 0.3813, 0.2550, 0.5106, 0.4954],
  1266. [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123],
  1267. [0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6063, 0.6083],
  1268. [0.6154, 0.4112, 0.7037, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600],
  1269. [0.6147, 0.4112, 0.7988, 0.3200, 0.3775, 0.2767, 0.5150, 0.5550]],
  1270. device='cuda:0', dtype=torch.float64)
  1271. predictions are: tensor([[ 0.6256, -0.3496, 1.5313, -0.1727, -0.4812, -0.2339, 0.4121, 0.2335],
  1272. [ 0.5802, -0.3599, 1.4961, -0.1003, -0.4505, -0.1714, 0.3991, 0.2416],
  1273. [ 0.6846, -0.3257, 1.6299, 0.1926, -0.5661, 0.1919, 0.4176, 0.2676],
  1274. [ 0.4458, -0.4585, 1.2494, -0.8908, -0.3609, -1.0730, 0.3473, 0.1982],
  1275. [ 0.4966, -0.3877, 1.1893, -0.8811, -0.4293, -0.9376, 0.3631, 0.2087],
  1276. [ 0.6634, -0.3550, 1.5272, -0.1704, -0.4972, -0.2177, 0.3968, 0.2328],
  1277. [ 0.4826, -0.3862, 1.2012, -0.8707, -0.4220, -0.9600, 0.3600, 0.1954],
  1278. [ 0.4057, -0.4449, 1.1502, -0.9486, -0.3773, -1.0787, 0.3214, 0.1925]],
  1279. device='cuda:0', grad_fn=<AddmmBackward>)
  1280. landmarks are: tensor([[[ 0.6017, -0.3654, 1.8654, -0.3998, -0.5365, -0.0765, 0.5894,
  1281. 0.3161],
  1282. [ 0.6012, -0.3846, 1.7326, -0.0457, -0.1381, 0.2853, 0.5259,
  1283. 0.1082],
  1284. [ 0.5934, -0.4276, 1.8192, -0.1458, -0.4499, 0.3777, 0.6702,
  1285. 0.1082],
  1286. [ 0.5485, -0.4209, 1.6691, -0.4152, -0.5249, -1.1081, 0.0725,
  1287. 0.0021],
  1288. [ 0.5342, -0.4280, 1.0051, -1.0619, -0.5711, -1.0388, 0.2226,
  1289. 0.0802],
  1290. [ 0.5774, -0.3868, 1.6286, -0.5692, -0.6462, -0.2767, 0.5143,
  1291. 0.5239],
  1292. [ 0.5565, -0.3864, 0.9647, -1.2159, -0.3286, -1.4391, 0.1852,
  1293. 0.3007],
  1294. [ 0.5532, -0.3864, 1.4035, -0.8079, -0.5423, -1.0080, 0.0928,
  1295. 0.2776]]], device='cuda:0')
  1296. loss_train_step before backward: tensor(0.0405, device='cuda:0', grad_fn=<MseLossBackward>)
  1297. loss_train_step after backward: tensor(0.0405, device='cuda:0', grad_fn=<MseLossBackward>)
  1298. loss_train: 4.160642504692078
  1299. step: 19
  1300. running loss: 0.21898118445747777
  1301. Train Steps: 19/90 Loss: 0.2190 torch.Size([8, 600, 800])
  1302. torch.Size([8, 8])
  1303. tensor([[0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136],
  1304. [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483],
  1305. [0.6248, 0.4185, 0.8500, 0.5767, 0.4463, 0.4550, 0.5613, 0.5917],
  1306. [0.6151, 0.4085, 0.8750, 0.4367, 0.3887, 0.4367, 0.5066, 0.5846],
  1307. [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947],
  1308. [0.6228, 0.4004, 0.8750, 0.5250, 0.3825, 0.5233, 0.6362, 0.5000],
  1309. [0.6115, 0.4005, 0.8838, 0.3867, 0.3763, 0.4700, 0.5800, 0.5550],
  1310. [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383]],
  1311. device='cuda:0', dtype=torch.float64)
  1312. predictions are: tensor([[ 0.4995, -0.4501, 1.2798, -0.9045, -0.3884, -1.1201, 0.3507, 0.1686],
  1313. [ 0.5374, -0.3718, 1.3481, -0.4737, -0.4760, -0.4540, 0.3643, 0.2129],
  1314. [ 0.5177, -0.4185, 1.3734, -0.5420, -0.4201, -0.6426, 0.3885, 0.1762],
  1315. [ 0.5296, -0.4432, 1.3546, -0.6588, -0.4503, -0.7486, 0.3094, 0.1954],
  1316. [ 0.5990, -0.3650, 1.4727, -0.2500, -0.4999, -0.2574, 0.3996, 0.2076],
  1317. [ 0.5845, -0.3787, 1.4616, -0.2690, -0.4853, -0.3027, 0.3903, 0.1992],
  1318. [ 0.5524, -0.3988, 1.3570, -0.6329, -0.4578, -0.6935, 0.3651, 0.1924],
  1319. [ 0.4944, -0.4201, 1.3606, -0.5418, -0.4362, -0.6470, 0.3477, 0.1965]],
  1320. device='cuda:0', grad_fn=<AddmmBackward>)
  1321. landmarks are: tensor([[[ 5.2355e-01, -4.2731e-01, 1.7499e+00, -4.3064e-01, -5.8268e-01,
  1322. -4.6143e-01, 1.6505e-01, 8.6245e-02],
  1323. [ 5.5456e-01, -4.6205e-01, 1.7788e+00, -4.2294e-01, -5.1917e-01,
  1324. -2.2633e-02, 4.2771e-01, 2.4681e-01],
  1325. [ 5.9988e-01, -3.5304e-01, 1.6402e+00, 3.7768e-01, -2.2471e-01,
  1326. -1.8430e-01, 3.0647e-01, 4.4696e-01],
  1327. [ 5.5525e-01, -3.9923e-01, 1.7557e+00, -2.6898e-01, -4.9030e-01,
  1328. -2.6898e-01, 5.4227e-02, 4.1446e-01],
  1329. [ 5.8031e-01, -4.3672e-01, 1.8423e+00, -3.0331e-02, -5.9423e-01,
  1330. 2.3557e-02, 6.5034e-01, -9.2270e-04],
  1331. [ 5.9082e-01, -4.3664e-01, 1.7557e+00, 1.3903e-01, -5.1917e-01,
  1332. 1.3133e-01, 6.5289e-01, 2.3557e-02],
  1333. [ 5.3839e-01, -4.3610e-01, 1.7961e+00, -4.9992e-01, -5.4804e-01,
  1334. -1.1501e-01, 3.9307e-01, 2.7760e-01],
  1335. [ 5.6637e-01, -4.3212e-01, 1.8249e+00, -2.0739e-01, -2.6513e-01,
  1336. 4.1617e-01, 5.6628e-01, 2.0062e-01]]], device='cuda:0')
  1337. loss_train_step before backward: tensor(0.0984, device='cuda:0', grad_fn=<MseLossBackward>)
  1338. loss_train_step after backward: tensor(0.0984, device='cuda:0', grad_fn=<MseLossBackward>)
  1339. loss_train: 4.259060598909855
  1340. step: 20
  1341. running loss: 0.21295302994549276
  1342.  
  1343. Train Steps: 20/90 Loss: 0.2130 torch.Size([8, 600, 800])
  1344. torch.Size([8, 8])
  1345. tensor([[0.6109, 0.4041, 0.6975, 0.3167, 0.3513, 0.3383, 0.5153, 0.5319],
  1346. [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500],
  1347. [0.6230, 0.4152, 0.7588, 0.2283, 0.4012, 0.2883, 0.6200, 0.5767],
  1348. [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967],
  1349. [0.6339, 0.4102, 0.8588, 0.3133, 0.4425, 0.2117, 0.6417, 0.5089],
  1350. [0.6299, 0.4008, 0.8450, 0.5350, 0.4213, 0.5000, 0.6350, 0.5100],
  1351. [0.6289, 0.4024, 0.9088, 0.4567, 0.3937, 0.5633, 0.7058, 0.5609],
  1352. [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683]],
  1353. device='cuda:0', dtype=torch.float64)
  1354. predictions are: tensor([[ 0.4723, -0.4247, 1.1850, -0.9063, -0.4518, -0.8865, 0.2939, 0.1639],
  1355. [ 0.5025, -0.3635, 1.3192, -0.5956, -0.5070, -0.4938, 0.3355, 0.1808],
  1356. [ 0.4405, -0.4470, 1.2356, -1.0312, -0.3698, -1.2157, 0.3035, 0.1463],
  1357. [ 0.3921, -0.5284, 1.1823, -1.2778, -0.2926, -1.5964, 0.2982, 0.1206],
  1358. [ 0.5190, -0.3989, 1.3849, -0.7171, -0.4370, -0.7962, 0.4063, 0.1556],
  1359. [ 0.6043, -0.3868, 1.5303, -0.0823, -0.4990, -0.1121, 0.3608, 0.1888],
  1360. [ 0.6078, -0.3671, 1.6489, 0.1029, -0.5688, 0.1931, 0.3999, 0.2117],
  1361. [ 0.5852, -0.4089, 1.5890, -0.1718, -0.4224, -0.2408, 0.3769, 0.1869]],
  1362. device='cuda:0', grad_fn=<AddmmBackward>)
  1363. landmarks are: tensor([[[ 5.3591e-01, -4.1932e-01, 9.3580e-01, -8.2325e-01, -6.6351e-01,
  1364. -7.2317e-01, 9.4325e-02, 1.7099e-01],
  1365. [ 5.7633e-01, -4.1470e-01, 1.3226e+00, -1.0619e+00, -6.6351e-01,
  1366. -4.1524e-01, 5.3741e-01, 2.5450e-01],
  1367. [ 5.9151e-01, -3.6821e-01, 1.2187e+00, -1.2313e+00, -4.3256e-01,
  1368. -9.5412e-01, 5.7783e-01, 3.7768e-01],
  1369. [ 6.0479e-01, -3.7229e-01, 1.6517e+00, -1.0773e+00, 4.6189e-04,
  1370. -1.5161e+00, 5.8660e-01, 8.0947e-03],
  1371. [ 6.4212e-01, -3.9120e-01, 1.6806e+00, -8.3865e-01, -2.4203e-01,
  1372. -1.3082e+00, 6.7795e-01, 6.4585e-02],
  1373. [ 6.2361e-01, -4.3441e-01, 1.6171e+00, 1.8522e-01, -3.4018e-01,
  1374. 2.3557e-02, 6.4711e-01, 6.9746e-02],
  1375. [ 6.1907e-01, -4.2731e-01, 1.9115e+00, -1.7660e-01, -4.6721e-01,
  1376. 3.1609e-01, 9.7406e-01, 3.0505e-01],
  1377. [ 5.3585e-01, -4.3703e-01, 1.7095e+00, -3.0331e-02, -8.0370e-02,
  1378. -3.8029e-02, 1.0439e-01, 3.3918e-01]]], device='cuda:0')
  1379. loss_train_step before backward: tensor(0.0438, device='cuda:0', grad_fn=<MseLossBackward>)
  1380. loss_train_step after backward: tensor(0.0438, device='cuda:0', grad_fn=<MseLossBackward>)
  1381. loss_train: 4.30288702994585
  1382. step: 21
  1383. running loss: 0.20489938237837382
  1384. Train Steps: 21/90 Loss: 0.2049 torch.Size([8, 600, 800])
  1385. torch.Size([8, 8])
  1386. tensor([[0.6188, 0.4099, 0.7400, 0.2433, 0.3962, 0.2750, 0.6162, 0.5467],
  1387. [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586],
  1388. [0.6304, 0.4024, 0.8925, 0.4800, 0.3937, 0.4817, 0.7485, 0.5297],
  1389. [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773],
  1390. [0.6148, 0.4076, 0.8666, 0.4820, 0.4138, 0.5067, 0.5250, 0.5767],
  1391. [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533],
  1392. [0.6185, 0.4042, 0.7700, 0.2250, 0.4062, 0.2117, 0.5763, 0.5150],
  1393. [0.6275, 0.4081, 0.8063, 0.2017, 0.4825, 0.1583, 0.6156, 0.4869]],
  1394. device='cuda:0', dtype=torch.float64)
  1395. predictions are: tensor([[ 0.4667, -0.4029, 1.2225, -0.9475, -0.4710, -0.9117, 0.3017, 0.1536],
  1396. [ 0.4264, -0.4826, 1.3187, -1.0577, -0.3452, -1.2749, 0.3562, 0.1209],
  1397. [ 0.5903, -0.3429, 1.5855, -0.0366, -0.5336, 0.1025, 0.4147, 0.1930],
  1398. [ 0.5301, -0.4158, 1.5460, -0.2706, -0.4314, -0.2919, 0.3681, 0.1735],
  1399. [ 0.5612, -0.3537, 1.6040, 0.0042, -0.5038, 0.1165, 0.3997, 0.1812],
  1400. [ 0.5596, -0.3945, 1.5997, -0.1237, -0.4685, -0.0593, 0.3873, 0.1867],
  1401. [ 0.4114, -0.4599, 1.1883, -1.2378, -0.3738, -1.3679, 0.2803, 0.1201],
  1402. [ 0.4364, -0.4916, 1.2448, -1.2068, -0.3356, -1.4575, 0.3233, 0.1142]],
  1403. device='cuda:0', grad_fn=<AddmmBackward>)
  1404. landmarks are: tensor([[[ 0.5725, -0.3926, 1.1321, -1.1620, -0.4557, -1.0157, 0.5605,
  1405. 0.2391],
  1406. [ 0.6075, -0.4129, 1.6055, -1.0080, -0.2420, -1.0080, 0.9704,
  1407. 0.2944],
  1408. [ 0.6257, -0.4273, 1.8365, -0.0688, -0.4672, -0.0611, 1.1715,
  1409. 0.1608],
  1410. [ 0.5428, -0.4244, 1.7095, -0.0380, 0.0232, 0.0725, 0.0866,
  1411. 0.3806],
  1412. [ 0.5539, -0.4031, 1.7168, -0.0594, -0.3748, 0.0543, 0.1390,
  1413. 0.3777],
  1414. [ 0.5499, -0.4225, 1.8018, -0.2921, -0.3055, 0.0543, 0.4046,
  1415. 0.2699],
  1416. [ 0.5707, -0.4189, 1.2707, -1.2467, -0.4095, -1.3082, 0.3758,
  1417. 0.0928],
  1418. [ 0.6124, -0.4010, 1.4381, -1.3544, -0.0573, -1.5546, 0.5573,
  1419. -0.0369]]], device='cuda:0')
  1420. loss_train_step before backward: tensor(0.0400, device='cuda:0', grad_fn=<MseLossBackward>)
  1421. loss_train_step after backward: tensor(0.0400, device='cuda:0', grad_fn=<MseLossBackward>)
  1422. loss_train: 4.342839952558279
  1423. step: 22
  1424. running loss: 0.19740181602537632
  1425. Train Steps: 22/90 Loss: 0.1974 torch.Size([8, 600, 800])
  1426. torch.Size([8, 8])
  1427. tensor([[0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083],
  1428. [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5687, 0.5567],
  1429. [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466],
  1430. [0.6128, 0.4118, 0.8638, 0.5333, 0.4625, 0.5267, 0.5193, 0.5475],
  1431. [0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200],
  1432. [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100],
  1433. [0.6224, 0.4179, 0.8700, 0.5683, 0.4037, 0.4683, 0.5650, 0.5600],
  1434. [0.6233, 0.4091, 0.8100, 0.2950, 0.3563, 0.3883, 0.6013, 0.5200]],
  1435. device='cuda:0', dtype=torch.float64)
  1436. predictions are: tensor([[ 0.5874, -0.3813, 1.6545, -0.0507, -0.4417, 0.0812, 0.4304, 0.1834],
  1437. [ 0.3994, -0.4998, 1.2229, -1.3992, -0.2548, -1.6338, 0.2909, 0.1109],
  1438. [ 0.4773, -0.3984, 1.3342, -0.7806, -0.4434, -0.6878, 0.3460, 0.1358],
  1439. [ 0.5034, -0.4020, 1.5718, -0.4009, -0.3795, -0.3917, 0.4109, 0.1579],
  1440. [ 0.4831, -0.4854, 1.3986, -1.1440, -0.2725, -1.4507, 0.3654, 0.1005],
  1441. [ 0.6052, -0.4153, 1.6555, -0.1839, -0.4558, -0.1716, 0.4230, 0.1649],
  1442. [ 0.6080, -0.3871, 1.6687, -0.1327, -0.4663, -0.0808, 0.4349, 0.1612],
  1443. [ 0.5008, -0.3815, 1.3538, -0.8037, -0.4405, -0.6999, 0.3763, 0.1449]],
  1444. device='cuda:0', grad_fn=<AddmmBackward>)
  1445. landmarks are: tensor([[[ 0.5713, -0.4105, 1.7557, 0.0467, -0.6520, -0.2690, 0.3988,
  1446. 0.5239],
  1447. [ 0.5824, -0.3625, 1.0405, -0.9325, -0.4268, -1.1389, 0.3411,
  1448. 0.2853],
  1449. [ 0.5076, -0.4443, 1.2337, -0.5023, -0.6808, -0.3614, 0.0866,
  1450. 0.2386],
  1451. [ 0.5447, -0.3838, 1.7037, 0.1775, -0.1497, 0.1467, 0.1128,
  1452. 0.2431],
  1453. [ 0.6089, -0.3948, 1.7383, -0.8617, -0.2536, -1.2390, 0.6009,
  1454. 0.1159],
  1455. [ 0.6125, -0.4273, 1.6402, 0.2006, -0.4788, -0.0303, 0.6182,
  1456. 0.0697],
  1457. [ 0.5891, -0.3557, 1.7326, 0.3392, -0.4210, -0.1227, 0.3238,
  1458. 0.3007],
  1459. [ 0.5932, -0.3962, 1.4554, -0.9233, -0.6404, -0.4922, 0.4912,
  1460. 0.1159]]], device='cuda:0')
  1461. loss_train_step before backward: tensor(0.0464, device='cuda:0', grad_fn=<MseLossBackward>)
  1462. loss_train_step after backward: tensor(0.0464, device='cuda:0', grad_fn=<MseLossBackward>)
  1463. loss_train: 4.389246199280024
  1464. step: 23
  1465. running loss: 0.1908367912730445
  1466. Train Steps: 23/90 Loss: 0.1908 torch.Size([8, 600, 800])
  1467. torch.Size([8, 8])
  1468. tensor([[0.6261, 0.3987, 0.8688, 0.4917, 0.4300, 0.5333, 0.7010, 0.5309],
  1469. [0.6218, 0.4185, 0.7338, 0.2650, 0.4625, 0.1950, 0.5687, 0.5800],
  1470. [0.6202, 0.4064, 0.7879, 0.2179, 0.4567, 0.1725, 0.5955, 0.5478],
  1471. [0.6307, 0.4029, 0.8650, 0.5200, 0.3763, 0.4017, 0.7311, 0.5366],
  1472. [0.6076, 0.3953, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954],
  1473. [0.6127, 0.4115, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495],
  1474. [0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367],
  1475. [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550]],
  1476. device='cuda:0', dtype=torch.float64)
  1477. predictions are: tensor([[ 0.6133, -0.3910, 1.8013, 0.1958, -0.3964, 0.3140, 0.4751, 0.2048],
  1478. [ 0.4495, -0.4382, 1.2529, -1.0922, -0.3345, -1.1163, 0.2963, 0.1465],
  1479. [ 0.4924, -0.4259, 1.3304, -1.0964, -0.2980, -1.2180, 0.3399, 0.1222],
  1480. [ 0.5933, -0.4085, 1.6925, -0.3447, -0.3869, -0.4754, 0.4793, 0.1284],
  1481. [ 0.5779, -0.3745, 1.5604, -0.4256, -0.4401, -0.2684, 0.4439, 0.1614],
  1482. [ 0.4712, -0.4171, 1.3140, -0.8936, -0.4123, -0.7876, 0.3292, 0.1524],
  1483. [ 0.4787, -0.4535, 1.4525, -0.9943, -0.2762, -1.1924, 0.3752, 0.1188],
  1484. [ 0.5423, -0.4336, 1.6983, -0.3355, -0.3292, -0.3300, 0.4789, 0.1588]],
  1485. device='cuda:0', grad_fn=<AddmmBackward>)
  1486. landmarks are: tensor([[[ 0.6059, -0.4442, 1.7268, -0.0149, -0.2998, 0.1775, 0.9521,
  1487. 0.1661],
  1488. [ 0.5862, -0.3530, 1.1032, -1.0619, -0.1497, -1.3852, 0.3411,
  1489. 0.3931],
  1490. [ 0.5787, -0.4086, 1.3535, -1.2794, -0.1764, -1.4891, 0.4645,
  1491. 0.2442],
  1492. [ 0.6273, -0.4249, 1.7095, 0.1159, -0.5480, -0.4306, 1.0910,
  1493. 0.1928],
  1494. [ 0.5206, -0.4603, 1.4670, -0.4768, -0.6577, -0.3998, 0.1836,
  1495. 0.0021],
  1496. [ 0.5442, -0.3855, 1.0224, -0.9541, -0.6115, -0.9233, 0.1745,
  1497. 0.2522],
  1498. [ 0.6174, -0.4201, 1.7309, -0.8784, -0.4735, -0.9524, 0.6242,
  1499. 0.1931],
  1500. [ 0.5351, -0.4321, 1.7326, -0.1381, 0.0463, 0.1049, 0.4162,
  1501. 0.2776]]], device='cuda:0')
  1502. loss_train_step before backward: tensor(0.0366, device='cuda:0', grad_fn=<MseLossBackward>)
  1503. loss_train_step after backward: tensor(0.0366, device='cuda:0', grad_fn=<MseLossBackward>)
  1504. loss_train: 4.425804238766432
  1505. step: 24
  1506. running loss: 0.18440850994860133
  1507.  
  1508. Train Steps: 24/90 Loss: 0.1844 torch.Size([8, 600, 800])
  1509. torch.Size([8, 8])
  1510. tensor([[0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5787, 0.5550],
  1511. [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116],
  1512. [0.6117, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103],
  1513. [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609],
  1514. [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901],
  1515. [0.6154, 0.4112, 0.7037, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600],
  1516. [0.6197, 0.4118, 0.8688, 0.5517, 0.4037, 0.5233, 0.5875, 0.5600],
  1517. [0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967]],
  1518. device='cuda:0', dtype=torch.float64)
  1519. predictions are: tensor([[ 0.5954, -0.4218, 1.5997, -0.6919, -0.3504, -0.7712, 0.4362, 0.1723],
  1520. [ 0.5064, -0.3984, 1.6899, -0.0856, -0.4166, 0.1454, 0.4647, 0.2048],
  1521. [ 0.5156, -0.4074, 1.3041, -1.1373, -0.3325, -1.1563, 0.3512, 0.1370],
  1522. [ 0.5488, -0.4500, 1.6568, -0.8152, -0.2328, -1.0531, 0.4730, 0.1296],
  1523. [ 0.5115, -0.4477, 1.4305, -1.0701, -0.2929, -1.2223, 0.3507, 0.1469],
  1524. [ 0.5264, -0.4210, 1.3500, -1.0950, -0.3389, -1.1687, 0.3748, 0.1390],
  1525. [ 0.5776, -0.4009, 1.8139, 0.1687, -0.4191, 0.2928, 0.5090, 0.2054],
  1526. [ 0.5844, -0.3730, 1.7113, -0.0934, -0.4380, 0.1155, 0.4841, 0.1837]],
  1527. device='cuda:0', grad_fn=<AddmmBackward>)
  1528. landmarks are: tensor([[[ 0.5647, -0.4129, 1.4901, -1.0619, -0.6462, -0.5846, 0.3873,
  1529. 0.2776],
  1530. [ 0.5376, -0.3903, 1.7095, -0.3229, -0.4730, 0.4701, 0.3871,
  1531. 0.0772],
  1532. [ 0.5393, -0.4299, 0.7453, -1.3775, -0.5596, -1.1081, 0.1527,
  1533. 0.0712],
  1534. [ 0.6026, -0.4032, 1.7326, -0.7771, -0.3691, -0.8617, 0.9704,
  1535. 0.3050],
  1536. [ 0.5515, -0.4129, 1.4785, -1.0080, -0.5192, -1.1004, 0.1034,
  1537. -0.0220],
  1538. [ 0.5565, -0.3864, 0.9647, -1.2159, -0.3286, -1.4391, 0.1852,
  1539. 0.3007],
  1540. [ 0.5762, -0.3840, 1.7268, 0.2622, -0.4210, 0.1313, 0.4277,
  1541. 0.3007],
  1542. [ 0.4974, -0.4482, 1.6633, -0.3306, -0.6173, 0.1313, 0.2925,
  1543. 0.0081]]], device='cuda:0')
  1544. loss_train_step before backward: tensor(0.0310, device='cuda:0', grad_fn=<MseLossBackward>)
  1545. loss_train_step after backward: tensor(0.0310, device='cuda:0', grad_fn=<MseLossBackward>)
  1546. loss_train: 4.456804607063532
  1547. step: 25
  1548. running loss: 0.17827218428254127
  1549. Train Steps: 25/90 Loss: 0.1783 torch.Size([8, 600, 800])
  1550. torch.Size([8, 8])
  1551. tensor([[ nan, nan, 0.7648, 0.2722, 0.3962, 0.2183, 0.5060, 0.5422],
  1552. [0.6236, 0.4084, 0.7738, 0.2133, 0.3663, 0.3233, 0.5813, 0.5567],
  1553. [0.6115, 0.4081, 0.6725, 0.2433, 0.4088, 0.1933, 0.5167, 0.5544],
  1554. [0.6168, 0.4029, 0.8523, 0.3417, 0.3588, 0.5000, 0.6125, 0.5400],
  1555. [0.6273, 0.4105, 0.8988, 0.4517, 0.3912, 0.2550, 0.5894, 0.4811],
  1556. [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391],
  1557. [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5687, 0.5567],
  1558. [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600]],
  1559. device='cuda:0', dtype=torch.float64)
  1560. predictions are: tensor([[ 0.5044, -0.4351, 1.3826, -1.0189, -0.3350, -1.0562, 0.3842, 0.1582],
  1561. [ 0.5649, -0.4414, 1.5426, -0.8654, -0.3159, -0.9570, 0.4214, 0.1911],
  1562. [ 0.5411, -0.4083, 1.3608, -1.0084, -0.3764, -1.0115, 0.3643, 0.1792],
  1563. [ 0.5614, -0.4106, 1.7066, -0.1109, -0.4423, 0.2117, 0.5177, 0.2240],
  1564. [ 0.6142, -0.4105, 1.7577, -0.4887, -0.3124, -0.6148, 0.5242, 0.1637],
  1565. [ 0.5997, -0.4014, 1.5340, -0.7335, -0.3835, -0.6683, 0.4661, 0.1797],
  1566. [ 0.4949, -0.4439, 1.4315, -1.0887, -0.2691, -1.1977, 0.3793, 0.1786],
  1567. [ 0.6371, -0.4062, 1.9366, 0.3314, -0.4247, 0.4813, 0.5960, 0.2351]],
  1568. device='cuda:0', grad_fn=<AddmmBackward>)
  1569. landmarks are: tensor([[[-2.2859, -2.2859, 1.2469, -1.0288, -0.4557, -1.2774, 0.0511,
  1570. 0.2183],
  1571. [ 0.5946, -0.3995, 1.2880, -1.3005, -0.5942, -0.7925, 0.3988,
  1572. 0.2853],
  1573. [ 0.5385, -0.4010, 0.8203, -1.1620, -0.3979, -1.3929, 0.1005,
  1574. 0.2747],
  1575. [ 0.5631, -0.4249, 1.6509, -0.7078, -0.6289, 0.0236, 0.5432,
  1576. 0.2083],
  1577. [ 0.6116, -0.3898, 1.8654, -0.1997, -0.4788, -1.1081, 0.4367,
  1578. -0.0637],
  1579. [ 0.5785, -0.4259, 1.4228, -1.0261, -0.4190, -1.2189, 0.4763,
  1580. 0.2043],
  1581. [ 0.5824, -0.3625, 1.0405, -0.9325, -0.4268, -1.1389, 0.3411,
  1582. 0.2853],
  1583. [ 0.5883, -0.3594, 1.7557, 0.2545, -0.4152, -0.0611, 0.3353,
  1584. 0.3007]]], device='cuda:0')
  1585. loss_train_step before backward: tensor(0.2222, device='cuda:0', grad_fn=<MseLossBackward>)
  1586. loss_train_step after backward: tensor(0.2222, device='cuda:0', grad_fn=<MseLossBackward>)
  1587. loss_train: 4.679033596068621
  1588. step: 26
  1589. running loss: 0.17996283061802387
  1590. Train Steps: 26/90 Loss: 0.1800 torch.Size([8, 600, 800])
  1591. torch.Size([8, 8])
  1592. tensor([[0.6274, 0.4003, 0.8638, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661],
  1593. [0.6068, 0.3963, 0.8650, 0.4317, 0.4037, 0.5083, 0.5253, 0.4999],
  1594. [0.6199, 0.4065, 0.7598, 0.2385, 0.4317, 0.1981, 0.5933, 0.5221],
  1595. [0.6042, 0.3990, 0.6831, 0.2875, 0.3500, 0.3133, 0.5143, 0.5510],
  1596. [0.6182, 0.3972, 0.8552, 0.5914, 0.3683, 0.4181, 0.5688, 0.5378],
  1597. [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5413, 0.5433],
  1598. [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967],
  1599. [0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166]],
  1600. device='cuda:0', dtype=torch.float64)
  1601. predictions are: tensor([[ 0.5752, -0.4557, 1.7538, -0.0873, -0.4162, 0.0275, 0.5230, 0.2209],
  1602. [ 0.5054, -0.4485, 1.6481, -0.3998, -0.3812, -0.2287, 0.4958, 0.1973],
  1603. [ 0.5210, -0.4305, 1.3226, -1.2333, -0.3546, -1.2685, 0.3925, 0.1713],
  1604. [ 0.4906, -0.4524, 1.2683, -1.1478, -0.3965, -1.0973, 0.3616, 0.2053],
  1605. [ 0.5731, -0.4653, 1.7510, -0.3070, -0.4253, -0.3510, 0.5229, 0.1982],
  1606. [ 0.5210, -0.4544, 1.4968, -0.9434, -0.3615, -1.0090, 0.4108, 0.1978],
  1607. [ 0.5749, -0.4515, 1.7346, -0.6102, -0.3249, -0.7786, 0.5225, 0.1889],
  1608. [ 0.5185, -0.4411, 1.7667, -0.0767, -0.4038, 0.1146, 0.5136, 0.2516]],
  1609. device='cuda:0', grad_fn=<AddmmBackward>)
  1610. landmarks are: tensor([[[ 0.6120, -0.4371, 1.7037, 0.4701, -0.5827, -0.0226, 0.5354,
  1611. -0.1331],
  1612. [ 0.5168, -0.4556, 1.7095, -0.2921, -0.4210, 0.0620, 0.1404,
  1613. 0.0231],
  1614. [ 0.5774, -0.4082, 1.2235, -1.1844, -0.2919, -1.3709, 0.4544,
  1615. 0.1256],
  1616. [ 0.5049, -0.4428, 0.8692, -0.9581, -0.6693, -0.8386, 0.0897,
  1617. 0.2589],
  1618. [ 0.5697, -0.4514, 1.6642, 0.4459, -0.5850, -0.3546, 0.3414,
  1619. 0.1982],
  1620. [ 0.5894, -0.3479, 1.7730, -0.6847, -0.5538, -0.7155, 0.2141,
  1621. 0.2237],
  1622. [ 0.6054, -0.3767, 1.7788, 0.0774, -0.5711, -0.7694, 0.5887,
  1623. 0.0081],
  1624. [ 0.5944, -0.4490, 1.8643, -0.0659, -0.5147, 0.1235, 0.7684,
  1625. 0.1004]]], device='cuda:0')
  1626. loss_train_step before backward: tensor(0.0442, device='cuda:0', grad_fn=<MseLossBackward>)
  1627. loss_train_step after backward: tensor(0.0442, device='cuda:0', grad_fn=<MseLossBackward>)
  1628. loss_train: 4.7232260555028915
  1629. step: 27
  1630. running loss: 0.17493429835195895
  1631. Train Steps: 27/90 Loss: 0.1749 torch.Size([8, 600, 800])
  1632. torch.Size([8, 8])
  1633. tensor([[0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742],
  1634. [0.6193, 0.4108, 0.7438, 0.2700, 0.3650, 0.3683, 0.6238, 0.5717],
  1635. [0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933],
  1636. [0.6198, 0.4164, 0.8700, 0.5067, 0.4625, 0.5650, 0.5464, 0.5197],
  1637. [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309],
  1638. [0.6163, 0.4006, 0.8788, 0.4683, 0.3663, 0.4883, 0.5887, 0.5017],
  1639. [0.6142, 0.3982, 0.8650, 0.4883, 0.3912, 0.4317, 0.5315, 0.5350],
  1640. [0.6198, 0.4130, 0.8762, 0.4117, 0.3650, 0.4900, 0.5707, 0.5103]],
  1641. device='cuda:0', dtype=torch.float64)
  1642. predictions are: tensor([[ 0.5405, -0.4394, 1.4507, -0.9549, -0.4286, -0.9721, 0.4815, 0.1903],
  1643. [ 0.4607, -0.4560, 1.2716, -0.9976, -0.4577, -0.8585, 0.3750, 0.2287],
  1644. [ 0.4619, -0.4713, 1.5905, -0.4083, -0.3709, -0.4039, 0.4716, 0.2178],
  1645. [ 0.4783, -0.4746, 1.6230, -0.4580, -0.3821, -0.4283, 0.4770, 0.2247],
  1646. [ 0.5517, -0.4558, 1.6601, -0.6311, -0.3759, -0.8260, 0.4948, 0.2009],
  1647. [ 0.4894, -0.4661, 1.5988, -0.3804, -0.4628, -0.1826, 0.4667, 0.2203],
  1648. [ 0.5504, -0.4706, 1.7239, -0.2985, -0.4422, -0.2883, 0.5128, 0.1965],
  1649. [ 0.5535, -0.4622, 1.7266, -0.4120, -0.4370, -0.4260, 0.4954, 0.2181]],
  1650. device='cuda:0', grad_fn=<AddmmBackward>)
  1651. landmarks are: tensor([[[ 0.6134, -0.4218, 1.7268, -0.6154, -0.4730, -1.0850, 0.5463,
  1652. -0.0957],
  1653. [ 0.5747, -0.3886, 1.1494, -1.0388, -0.6000, -0.5846, 0.5952,
  1654. 0.3546],
  1655. [ 0.5959, -0.3579, 1.6055, 0.3623, -0.0573, -0.2074, 0.3122,
  1656. 0.4547],
  1657. [ 0.5770, -0.3624, 1.7326, 0.0543, -0.1497, 0.3238, 0.2378,
  1658. 0.1146],
  1659. [ 0.6520, -0.3623, 1.8885, 0.0313, -0.5538, -0.5384, 0.6926,
  1660. 0.1661],
  1661. [ 0.5609, -0.4354, 1.7730, -0.1227, -0.5942, -0.0303, 0.4335,
  1662. 0.0313],
  1663. [ 0.5513, -0.4467, 1.7095, -0.0303, -0.4788, -0.2921, 0.1692,
  1664. 0.1854],
  1665. [ 0.5771, -0.3784, 1.7614, -0.3844, -0.6000, -0.0226, 0.3501,
  1666. 0.0712]]], device='cuda:0')
  1667. loss_train_step before backward: tensor(0.0562, device='cuda:0', grad_fn=<MseLossBackward>)
  1668. loss_train_step after backward: tensor(0.0562, device='cuda:0', grad_fn=<MseLossBackward>)
  1669. loss_train: 4.779460072517395
  1670. step: 28
  1671. running loss: 0.17069500258990697
  1672.  
  1673. Train Steps: 28/90 Loss: 0.1707 torch.Size([8, 600, 800])
  1674. torch.Size([8, 8])
  1675. tensor([[0.6311, 0.3998, 0.7975, 0.5767, 0.3838, 0.4850, 0.7327, 0.5343],
  1676. [0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904],
  1677. [0.6100, 0.4016, 0.8600, 0.5067, 0.4612, 0.5233, 0.5086, 0.5519],
  1678. [0.6200, 0.3978, 0.8900, 0.4550, 0.3775, 0.5200, 0.6150, 0.5367],
  1679. [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200],
  1680. [0.6257, 0.4034, 0.8287, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901],
  1681. [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869],
  1682. [ nan, nan, 0.8525, 0.2217, 0.5413, 0.2367, 0.7367, 0.5482]],
  1683. device='cuda:0', dtype=torch.float64)
  1684. predictions are: tensor([[ 0.4621, -0.4910, 1.5000, -0.4019, -0.4888, -0.2934, 0.4030, 0.2112],
  1685. [ 0.4917, -0.4781, 1.5036, -0.9912, -0.3765, -1.2404, 0.4376, 0.1793],
  1686. [ 0.4635, -0.4929, 1.6376, -0.2115, -0.4377, -0.1783, 0.4493, 0.2203],
  1687. [ 0.4968, -0.4848, 1.6916, -0.1059, -0.5143, 0.0720, 0.5134, 0.2184],
  1688. [ 0.5540, -0.4735, 1.7266, -0.2006, -0.5030, -0.1378, 0.4941, 0.2128],
  1689. [ 0.5009, -0.4610, 1.3333, -1.0000, -0.4870, -1.0072, 0.4258, 0.1925],
  1690. [ 0.5170, -0.4835, 1.7170, -0.1887, -0.4760, -0.2195, 0.4769, 0.1899],
  1691. [ 0.4410, -0.5048, 1.4669, -1.0029, -0.3608, -1.2285, 0.4652, 0.1677]],
  1692. device='cuda:0', grad_fn=<AddmmBackward>)
  1693. landmarks are: tensor([[[ 0.6289, -0.4393, 1.3977, 0.3777, -0.5134, -0.0457, 1.0984,
  1694. 0.1821],
  1695. [ 0.6126, -0.3871, 1.6229, -1.0773, -0.2132, -1.3698, 0.5829,
  1696. -0.0209],
  1697. [ 0.5318, -0.4310, 1.6864, 0.0543, -0.1554, 0.1313, 0.0635,
  1698. 0.2634],
  1699. [ 0.5776, -0.4484, 1.8249, -0.1843, -0.5423, 0.1159, 0.5547,
  1700. 0.1929],
  1701. [ 0.5865, -0.3973, 1.8423, -0.0688, -0.5192, -0.2305, 0.4162,
  1702. 0.1159],
  1703. [ 0.6042, -0.4225, 1.5420, -1.2082, -0.4730, -1.0311, 0.6380,
  1704. -0.0220],
  1705. [ 0.5320, -0.4488, 1.6633, 0.3315, -0.5018, -0.2459, 0.0828,
  1706. -0.0370],
  1707. [-2.2859, -2.2859, 1.6517, -1.2620, 0.2141, -1.1928, 1.1166,
  1708. 0.2463]]], device='cuda:0')
  1709. loss_train_step before backward: tensor(0.2187, device='cuda:0', grad_fn=<MseLossBackward>)
  1710. loss_train_step after backward: tensor(0.2187, device='cuda:0', grad_fn=<MseLossBackward>)
  1711. loss_train: 4.998158469796181
  1712. step: 29
  1713. running loss: 0.17235029206193728
  1714. Train Steps: 29/90 Loss: 0.1724 torch.Size([8, 600, 800])
  1715. torch.Size([8, 8])
  1716. tensor([[0.6216, 0.4167, 0.8588, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667],
  1717. [0.6321, 0.4048, 0.8738, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927],
  1718. [0.6060, 0.3924, 0.8450, 0.5717, 0.4200, 0.5217, 0.5253, 0.4752],
  1719. [0.6222, 0.3957, 0.8838, 0.5017, 0.3937, 0.4600, 0.5900, 0.5017],
  1720. [0.6224, 0.4179, 0.8700, 0.5683, 0.4037, 0.4683, 0.5650, 0.5600],
  1721. [0.6357, 0.4118, 0.8400, 0.2500, 0.5413, 0.1633, 0.6725, 0.5586],
  1722. [0.6261, 0.4066, 0.8325, 0.2150, 0.4763, 0.2667, 0.7002, 0.5633],
  1723. [0.6182, 0.4058, 0.8738, 0.4350, 0.3563, 0.3400, 0.5290, 0.5822]],
  1724. device='cuda:0', dtype=torch.float64)
  1725. predictions are: tensor([[ 0.4439, -0.4882, 1.6552, -0.1041, -0.5164, -0.0036, 0.4752, 0.1893],
  1726. [ 0.4620, -0.4752, 1.6635, -0.2930, -0.4908, -0.3561, 0.4884, 0.1639],
  1727. [ 0.4005, -0.5161, 1.5989, -0.2423, -0.4633, -0.2455, 0.4451, 0.1616],
  1728. [ 0.4913, -0.5129, 1.6991, -0.2184, -0.4814, -0.2698, 0.4712, 0.1792],
  1729. [ 0.4668, -0.4844, 1.6592, -0.1376, -0.4977, -0.1037, 0.4597, 0.1926],
  1730. [ 0.3816, -0.5161, 1.4410, -0.9878, -0.3631, -1.2132, 0.4439, 0.1703],
  1731. [ 0.4210, -0.4985, 1.4144, -1.0300, -0.4295, -1.1849, 0.4380, 0.1659],
  1732. [ 0.4693, -0.5386, 1.6060, -0.4653, -0.4800, -0.5654, 0.4283, 0.1936]],
  1733. device='cuda:0', grad_fn=<AddmmBackward>)
  1734. landmarks are: tensor([[[ 0.5853, -0.3614, 1.6806, 0.2930, -0.4499, 0.1005, 0.3815,
  1735. 0.3315],
  1736. [ 0.6335, -0.4162, 1.7499, 0.3084, -0.4961, -0.2459, 0.6524,
  1737. -0.0102],
  1738. [ 0.5132, -0.4736, 1.6171, 0.3546, -0.3460, 0.1236, 0.1404,
  1739. -0.0911],
  1740. [ 0.5879, -0.4583, 1.7961, 0.0313, -0.4672, -0.1612, 0.4393,
  1741. 0.0313],
  1742. [ 0.5891, -0.3557, 1.7326, 0.3392, -0.4210, -0.1227, 0.3238,
  1743. 0.3007],
  1744. [ 0.6504, -0.3840, 1.5940, -1.1312, 0.2141, -1.5315, 0.8205,
  1745. 0.2944],
  1746. [ 0.6059, -0.4080, 1.5594, -1.2928, -0.0861, -1.0542, 0.9485,
  1747. 0.3157],
  1748. [ 0.5697, -0.4113, 1.7499, -0.2767, -0.6404, -0.7155, 0.1576,
  1749. 0.4032]]], device='cuda:0')
  1750. loss_train_step before backward: tensor(0.0521, device='cuda:0', grad_fn=<MseLossBackward>)
  1751. loss_train_step after backward: tensor(0.0521, device='cuda:0', grad_fn=<MseLossBackward>)
  1752. loss_train: 5.050214387476444
  1753. step: 30
  1754. running loss: 0.16834047958254814
  1755. Train Steps: 30/90 Loss: 0.1683 torch.Size([8, 600, 800])
  1756. torch.Size([8, 8])
  1757. tensor([[0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334],
  1758. [0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804],
  1759. [0.6083, 0.3957, 0.8638, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980],
  1760. [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383],
  1761. [0.6218, 0.4185, 0.7338, 0.2650, 0.4625, 0.1950, 0.5687, 0.5800],
  1762. [0.6079, 0.3964, 0.7420, 0.2958, 0.3563, 0.2917, 0.5351, 0.4980],
  1763. [0.6162, 0.4134, 0.6700, 0.2467, 0.3962, 0.2533, 0.5737, 0.5467],
  1764. [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726]],
  1765. device='cuda:0', dtype=torch.float64)
  1766. predictions are: tensor([[ 0.4256, -0.5307, 1.4230, -0.6611, -0.4992, -0.7421, 0.4294, 0.1491],
  1767. [ 0.4814, -0.4975, 1.7566, -0.1036, -0.4995, -0.1922, 0.5431, 0.1585],
  1768. [ 0.4664, -0.5085, 1.9594, 0.3104, -0.4150, 0.1499, 0.5579, 0.1315],
  1769. [ 0.4296, -0.5129, 1.9311, 0.3247, -0.4423, 0.3089, 0.5455, 0.1614],
  1770. [ 0.3443, -0.5417, 1.3398, -0.8960, -0.4093, -1.0721, 0.3534, 0.1745],
  1771. [ 0.4356, -0.5197, 1.4351, -0.7563, -0.4948, -0.8742, 0.4220, 0.1475],
  1772. [ 0.3822, -0.5374, 1.4107, -0.7609, -0.4680, -0.9120, 0.3871, 0.1718],
  1773. [ 0.4443, -0.4906, 1.4567, -0.7145, -0.4900, -0.8297, 0.4329, 0.1366]],
  1774. device='cuda:0', grad_fn=<AddmmBackward>)
  1775. landmarks are: tensor([[[ 5.7296e-01, -4.5619e-01, 1.2195e+00, -1.2440e+00, -5.4966e-01,
  1776. -7.7109e-01, 5.7045e-01, 1.7788e-01],
  1777. [ 6.0710e-01, -4.1186e-01, 1.7788e+00, -5.1532e-01, -6.0000e-01,
  1778. -5.6921e-01, 6.5857e-01, -6.7050e-02],
  1779. [ 5.2379e-01, -4.5797e-01, 1.7037e+00, 4.6189e-04, -2.7090e-01,
  1780. 6.2048e-02, 1.8356e-01, 1.4106e-02],
  1781. [ 5.6637e-01, -4.3212e-01, 1.8249e+00, -2.0739e-01, -2.6513e-01,
  1782. 4.1617e-01, 5.6628e-01, 2.0062e-01],
  1783. [ 5.8620e-01, -3.5296e-01, 1.1032e+00, -1.0619e+00, -1.4965e-01,
  1784. -1.3852e+00, 3.4111e-01, 3.9307e-01],
  1785. [ 5.2194e-01, -4.5504e-01, 1.1415e+00, -9.1962e-01, -6.4042e-01,
  1786. -9.3872e-01, 1.8562e-01, 1.4106e-02],
  1787. [ 5.6028e-01, -3.7637e-01, 8.0878e-01, -1.1466e+00, -4.5566e-01,
  1788. -1.1158e+00, 3.6420e-01, 2.3911e-01],
  1789. [ 4.9971e-01, -4.4465e-01, 1.1610e+00, -9.7721e-01, -6.0577e-01,
  1790. -1.0311e+00, 1.4038e-01, -1.0312e-01]]], device='cuda:0')
  1791. loss_train_step before backward: tensor(0.0491, device='cuda:0', grad_fn=<MseLossBackward>)
  1792. loss_train_step after backward: tensor(0.0491, device='cuda:0', grad_fn=<MseLossBackward>)
  1793. loss_train: 5.099356964230537
  1794. step: 31
  1795. running loss: 0.16449538594292057
  1796. Train Steps: 31/90 Loss: 0.1645 torch.Size([8, 600, 800])
  1797. torch.Size([8, 8])
  1798. tensor([[0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297],
  1799. [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092],
  1800. [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667],
  1801. [0.6128, 0.4115, 0.7934, 0.3778, 0.3450, 0.4033, 0.5337, 0.5456],
  1802. [0.6055, 0.4015, 0.7425, 0.2033, 0.4113, 0.1883, 0.5217, 0.4823],
  1803. [0.6120, 0.4014, 0.6863, 0.2817, 0.3700, 0.2783, 0.5513, 0.5667],
  1804. [0.6125, 0.4035, 0.7825, 0.3100, 0.3463, 0.4900, 0.5832, 0.5637],
  1805. [ nan, nan, 0.8525, 0.2217, 0.5413, 0.2367, 0.7367, 0.5482]],
  1806. device='cuda:0', dtype=torch.float64)
  1807. predictions are: tensor([[ 0.4981, -0.5002, 1.8641, 0.2520, -0.4876, 0.1037, 0.5152, 0.1010],
  1808. [ 0.4992, -0.4909, 1.9148, 0.4428, -0.4518, 0.3332, 0.5102, 0.1114],
  1809. [ 0.5246, -0.4915, 1.7953, -0.2241, -0.4603, -0.4731, 0.5106, 0.1482],
  1810. [ 0.3892, -0.4989, 1.4743, -0.3986, -0.5017, -0.3789, 0.4380, 0.1431],
  1811. [ 0.3725, -0.5422, 1.3167, -0.9795, -0.4368, -1.2552, 0.3796, 0.1484],
  1812. [ 0.3573, -0.5144, 1.2350, -0.8814, -0.4860, -0.9428, 0.3407, 0.1568],
  1813. [ 0.3831, -0.4949, 1.4455, -0.4288, -0.5095, -0.3071, 0.3988, 0.1604],
  1814. [ 0.2896, -0.5915, 1.6280, -0.6673, -0.2973, -1.0396, 0.5594, 0.1253]],
  1815. device='cuda:0', grad_fn=<AddmmBackward>)
  1816. landmarks are: tensor([[[ 6.5036e-01, -3.7434e-01, 1.6171e+00, 4.3156e-01, -5.4226e-01,
  1817. 4.6189e-04, 7.1085e-01, 1.6077e-01],
  1818. [ 5.7829e-01, -4.3634e-01, 1.7724e+00, 4.3211e-02, -5.1531e-01,
  1819. 8.7136e-02, 4.8401e-01, 6.6263e-02],
  1820. [ 5.7685e-01, -3.8568e-01, 1.5305e+00, -7.6936e-01, -6.4619e-01,
  1821. -6.3079e-01, 3.9885e-01, 3.3149e-01],
  1822. [ 5.4440e-01, -3.8522e-01, 1.3786e+00, -5.4087e-01, -6.9238e-01,
  1823. -4.2294e-01, 1.7915e-01, 2.3412e-01],
  1824. [ 5.1085e-01, -4.3164e-01, 1.1436e+00, -1.3467e+00, -3.8637e-01,
  1825. -1.4160e+00, 1.2393e-01, -5.8033e-02],
  1826. [ 5.4099e-01, -4.3210e-01, 8.8383e-01, -9.8491e-01, -5.7691e-01,
  1827. -1.0003e+00, 2.6028e-01, 3.3149e-01],
  1828. [ 5.4319e-01, -4.2240e-01, 1.3284e+00, -8.5404e-01, -6.8661e-01,
  1829. -2.2633e-02, 4.0770e-01, 3.1769e-01],
  1830. [-2.2859e+00, -2.2859e+00, 1.6517e+00, -1.2620e+00, 2.1409e-01,
  1831. -1.1928e+00, 1.1166e+00, 2.4627e-01]]], device='cuda:0')
  1832. loss_train_step before backward: tensor(0.1953, device='cuda:0', grad_fn=<MseLossBackward>)
  1833. loss_train_step after backward: tensor(0.1953, device='cuda:0', grad_fn=<MseLossBackward>)
  1834. loss_train: 5.294639840722084
  1835. step: 32
  1836. running loss: 0.16545749502256513
  1837.  
  1838. Train Steps: 32/90 Loss: 0.1655 torch.Size([8, 600, 800])
  1839. torch.Size([8, 8])
  1840. tensor([[0.6221, 0.4107, 0.7788, 0.3033, 0.3950, 0.2817, 0.6075, 0.5517],
  1841. [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283],
  1842. [0.6151, 0.4125, 0.8738, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483],
  1843. [0.6136, 0.4085, 0.6688, 0.2317, 0.3862, 0.2367, 0.5517, 0.5783],
  1844. [0.6263, 0.4030, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771],
  1845. [ nan, nan, 0.7192, 0.2346, 0.4037, 0.2050, 0.5138, 0.5650],
  1846. [0.6099, 0.4030, 0.8638, 0.5117, 0.4983, 0.4965, 0.5086, 0.5388],
  1847. [0.6098, 0.3991, 0.8638, 0.4717, 0.4263, 0.4967, 0.5212, 0.5650]],
  1848. device='cuda:0', dtype=torch.float64)
  1849. predictions are: tensor([[ 0.3085, -0.5768, 1.4610, -0.7784, -0.3601, -1.1018, 0.4599, 0.1369],
  1850. [ 0.4371, -0.5274, 1.7047, -0.0446, -0.4892, -0.1888, 0.5027, 0.0903],
  1851. [ 0.4908, -0.4720, 1.7375, -0.0467, -0.5059, -0.2004, 0.4971, 0.1306],
  1852. [ 0.3127, -0.5506, 1.1132, -1.1260, -0.4676, -1.1954, 0.3508, 0.1386],
  1853. [ 0.4899, -0.4851, 1.8631, 0.2737, -0.4897, 0.2595, 0.5035, 0.1027],
  1854. [ 0.2675, -0.5867, 1.1643, -1.0927, -0.4417, -1.2381, 0.3678, 0.1496],
  1855. [ 0.3908, -0.5745, 1.7423, -0.0200, -0.3634, -0.2191, 0.4888, 0.1441],
  1856. [ 0.3999, -0.5149, 1.7153, 0.0344, -0.4431, 0.0088, 0.4822, 0.1426]],
  1857. device='cuda:0', grad_fn=<AddmmBackward>)
  1858. landmarks are: tensor([[[ 0.5875, -0.3888, 1.3111, -0.8848, -0.4614, -0.9849, 0.5201,
  1859. 0.2622],
  1860. [ 0.5900, -0.4377, 1.6113, 0.3623, -0.5018, -0.3229, 0.3700,
  1861. 0.1544],
  1862. [ 0.5551, -0.3807, 1.7499, -0.2459, -0.6346, -0.5384, 0.0871,
  1863. 0.2468],
  1864. [ 0.5483, -0.3991, 0.8030, -1.2159, -0.5018, -1.1928, 0.2624,
  1865. 0.3852],
  1866. [ 0.6072, -0.4247, 1.8711, -0.0842, -0.5307, 0.1005, 0.6771,
  1867. -0.0821],
  1868. [-2.2859, -2.2859, 1.0361, -1.2021, -0.4210, -1.3390, 0.0871,
  1869. 0.3238],
  1870. [ 0.5309, -0.4246, 1.7037, 0.0774, 0.0158, 0.0075, 0.0635,
  1871. 0.2026],
  1872. [ 0.5308, -0.4425, 1.7037, -0.1073, -0.3171, 0.0082, 0.1217,
  1873. 0.3238]]], device='cuda:0')
  1874. loss_train_step before backward: tensor(0.1797, device='cuda:0', grad_fn=<MseLossBackward>)
  1875. loss_train_step after backward: tensor(0.1797, device='cuda:0', grad_fn=<MseLossBackward>)
  1876. loss_train: 5.4743141531944275
  1877. step: 33
  1878. running loss: 0.1658883076725584
  1879. Train Steps: 33/90 Loss: 0.1659 torch.Size([8, 600, 800])
  1880. torch.Size([8, 8])
  1881. tensor([[0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672],
  1882. [0.6120, 0.4014, 0.6863, 0.2817, 0.3700, 0.2783, 0.5513, 0.5667],
  1883. [0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901],
  1884. [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583],
  1885. [0.6346, 0.4086, 0.7938, 0.5500, 0.3962, 0.4867, 0.7343, 0.5702],
  1886. [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749],
  1887. [0.6148, 0.3996, 0.8488, 0.3867, 0.3488, 0.4067, 0.5863, 0.5000],
  1888. [0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540]],
  1889. device='cuda:0', dtype=torch.float64)
  1890. predictions are: tensor([[ 0.4537, -0.5209, 1.6994, -0.1240, -0.4797, -0.3147, 0.4962, 0.1137],
  1891. [ 0.2515, -0.5921, 1.1194, -1.0232, -0.4625, -1.1073, 0.3233, 0.1711],
  1892. [ 0.4298, -0.5692, 1.8640, 0.1525, -0.3792, -0.1307, 0.4597, 0.1279],
  1893. [ 0.3892, -0.5137, 1.5070, -0.3672, -0.4984, -0.3483, 0.4242, 0.1417],
  1894. [ 0.3929, -0.5408, 1.5922, -0.0295, -0.4774, -0.1083, 0.4134, 0.1505],
  1895. [ 0.3016, -0.5772, 1.1988, -0.8329, -0.4858, -0.8551, 0.3505, 0.1630],
  1896. [ 0.4089, -0.5140, 1.5588, -0.3554, -0.5156, -0.3883, 0.4443, 0.1207],
  1897. [ 0.2977, -0.6028, 1.8114, -0.2007, -0.3145, -0.6246, 0.5291, 0.1498]],
  1898. device='cuda:0', grad_fn=<AddmmBackward>)
  1899. landmarks are: tensor([[[ 0.6132, -0.4324, 1.8192, -0.0842, -0.6231, -0.6385, 0.5537,
  1900. -0.1278],
  1901. [ 0.5410, -0.4321, 0.8838, -0.9849, -0.5769, -1.0003, 0.2603,
  1902. 0.3315],
  1903. [ 0.5663, -0.3955, 1.7788, -0.2382, -0.4037, -0.2690, 0.0828,
  1904. -0.0220],
  1905. [ 0.5761, -0.3966, 1.6171, -0.4845, -0.6346, -0.4691, 0.4739,
  1906. 0.2930],
  1907. [ 0.6454, -0.3984, 1.3804, 0.2545, -0.4557, -0.0380, 1.1057,
  1908. 0.3478],
  1909. [ 0.5419, -0.4160, 1.1810, -0.8939, -0.6808, -0.7463, 0.2485,
  1910. 0.3694],
  1911. [ 0.5538, -0.4400, 1.6344, -0.4999, -0.6751, -0.4075, 0.4219,
  1912. 0.0236],
  1913. [ 0.6240, -0.3912, 1.9115, -0.2382, -0.3979, -0.8694, 0.8644,
  1914. 0.2730]]], device='cuda:0')
  1915. loss_train_step before backward: tensor(0.0368, device='cuda:0', grad_fn=<MseLossBackward>)
  1916. loss_train_step after backward: tensor(0.0368, device='cuda:0', grad_fn=<MseLossBackward>)
  1917. loss_train: 5.511073064059019
  1918. step: 34
  1919. running loss: 0.16209038423702998
  1920. Train Steps: 34/90 Loss: 0.1621 torch.Size([8, 600, 800])
  1921. torch.Size([8, 8])
  1922. tensor([[0.6249, 0.4142, 0.8350, 0.3283, 0.3613, 0.3700, 0.6188, 0.5400],
  1923. [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170],
  1924. [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5687, 0.5567],
  1925. [0.6228, 0.4119, 0.7938, 0.2233, 0.4674, 0.1773, 0.6188, 0.5433],
  1926. [0.6214, 0.4112, 0.7838, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083],
  1927. [ nan, nan, 0.7525, 0.2291, 0.3838, 0.3017, 0.6050, 0.5667],
  1928. [0.6250, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6088, 0.5183],
  1929. [0.6189, 0.4033, 0.8650, 0.5267, 0.4487, 0.5150, 0.5925, 0.5050]],
  1930. device='cuda:0', dtype=torch.float64)
  1931. predictions are: tensor([[ 0.3724, -0.5650, 1.5811, -0.3080, -0.4808, -0.4925, 0.4385, 0.1313],
  1932. [ 0.2653, -0.6318, 1.3712, -0.6937, -0.3957, -0.8542, 0.4976, 0.1337],
  1933. [ 0.1592, -0.6391, 1.2529, -0.8141, -0.3916, -0.9404, 0.3738, 0.2095],
  1934. [ 0.1828, -0.6339, 1.3143, -0.7811, -0.3769, -1.0112, 0.4253, 0.1762],
  1935. [ 0.4055, -0.5633, 1.4417, -0.5345, -0.5245, -0.6789, 0.4411, 0.1603],
  1936. [ 0.2255, -0.6256, 1.2832, -0.7301, -0.4409, -0.8754, 0.3790, 0.1613],
  1937. [ 0.5063, -0.4833, 1.8031, 0.4133, -0.4401, 0.4369, 0.4695, 0.1416],
  1938. [ 0.5404, -0.4651, 1.7917, 0.4030, -0.4918, 0.3713, 0.4484, 0.1168]],
  1939. device='cuda:0', grad_fn=<AddmmBackward>)
  1940. landmarks are: tensor([[[ 0.6006, -0.3728, 1.5709, -0.7694, -0.6173, -0.5769, 0.5721,
  1941. 0.2083],
  1942. [ 0.6471, -0.3840, 1.5767, -1.0311, -0.0457, -1.5007, 0.6889,
  1943. 0.1020],
  1944. [ 0.5824, -0.3625, 1.0405, -0.9325, -0.4268, -1.1389, 0.3411,
  1945. 0.2853],
  1946. [ 0.5908, -0.3832, 1.3804, -1.2543, -0.1270, -1.4671, 0.5721,
  1947. 0.2237],
  1948. [ 0.5845, -0.3864, 1.3342, -1.3082, -0.6000, -0.8386, 0.3353,
  1949. 0.0620],
  1950. [-2.2859, -2.2859, 1.1898, -1.2278, -0.5134, -0.8925, 0.5085,
  1951. 0.3315],
  1952. [ 0.6012, -0.3846, 1.7326, -0.0457, -0.1381, 0.2853, 0.5259,
  1953. 0.1082],
  1954. [ 0.5726, -0.4230, 1.7095, 0.1467, -0.2132, 0.0928, 0.4508,
  1955. 0.0467]]], device='cuda:0')
  1956. loss_train_step before backward: tensor(0.2046, device='cuda:0', grad_fn=<MseLossBackward>)
  1957. loss_train_step after backward: tensor(0.2046, device='cuda:0', grad_fn=<MseLossBackward>)
  1958. loss_train: 5.71563608571887
  1959. step: 35
  1960. running loss: 0.16330388816339628
  1961. Train Steps: 35/90 Loss: 0.1633 torch.Size([8, 600, 800])
  1962. torch.Size([8, 8])
  1963. tensor([[0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395],
  1964. [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183],
  1965. [0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533],
  1966. [0.6231, 0.3973, 0.8650, 0.3950, 0.3625, 0.3183, 0.5837, 0.5167],
  1967. [0.6236, 0.4084, 0.7738, 0.2133, 0.3663, 0.3233, 0.5813, 0.5567],
  1968. [0.6314, 0.4107, 0.8750, 0.5100, 0.3788, 0.4900, 0.7121, 0.5864],
  1969. [0.6124, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485],
  1970. [0.6289, 0.4032, 0.8419, 0.5446, 0.4075, 0.5017, 0.6312, 0.5117]],
  1971. device='cuda:0', dtype=torch.float64)
  1972. predictions are: tensor([[ 0.1656, -0.6813, 1.2116, -0.9808, -0.3851, -1.2191, 0.4084, 0.2121],
  1973. [ 0.5132, -0.4670, 1.8552, 0.3621, -0.4981, 0.4013, 0.4923, 0.1279],
  1974. [ 0.2068, -0.6265, 1.0946, -0.9656, -0.4810, -1.0109, 0.3559, 0.1965],
  1975. [ 0.4097, -0.5306, 1.4900, -0.4212, -0.4880, -0.4898, 0.4622, 0.1524],
  1976. [ 0.1992, -0.6792, 1.3864, -0.7662, -0.3886, -1.0255, 0.4300, 0.2195],
  1977. [ 0.5568, -0.4758, 1.8857, 0.3602, -0.4827, 0.2015, 0.5142, 0.1116],
  1978. [ 0.2005, -0.6281, 1.0860, -1.0023, -0.4493, -1.0749, 0.3525, 0.2077],
  1979. [ 0.5050, -0.4917, 1.7101, 0.1303, -0.4779, 0.0353, 0.4208, 0.1442]],
  1980. device='cuda:0', grad_fn=<AddmmBackward>)
  1981. landmarks are: tensor([[[ 0.5713, -0.4538, 1.1028, -1.3659, -0.3831, -1.1273, 0.5340,
  1982. 0.2058],
  1983. [ 0.5934, -0.4276, 1.8192, -0.1458, -0.4499, 0.3777, 0.6702,
  1984. 0.1082],
  1985. [ 0.5878, -0.4052, 1.0229, -1.2855, -0.5596, -0.8232, 0.5316,
  1986. 0.2699],
  1987. [ 0.5924, -0.4507, 1.7095, -0.4614, -0.6115, -0.8156, 0.4104,
  1988. 0.1005],
  1989. [ 0.5946, -0.3995, 1.2880, -1.3005, -0.5942, -0.7925, 0.3988,
  1990. 0.2853],
  1991. [ 0.6306, -0.3888, 1.7557, 0.0697, -0.5365, -0.0226, 1.0033,
  1992. 0.4226],
  1993. [ 0.5425, -0.3998, 0.9263, -0.8683, -0.6000, -1.0157, 0.0990,
  1994. 0.2476],
  1995. [ 0.6189, -0.4238, 1.6026, 0.2295, -0.4037, 0.0313, 0.6298,
  1996. 0.0774]]], device='cuda:0')
  1997. loss_train_step before backward: tensor(0.0442, device='cuda:0', grad_fn=<MseLossBackward>)
  1998. loss_train_step after backward: tensor(0.0442, device='cuda:0', grad_fn=<MseLossBackward>)
  1999. loss_train: 5.759868372231722
  2000. step: 36
  2001. running loss: 0.1599963436731034
  2002.  
  2003. Train Steps: 36/90 Loss: 0.1600 torch.Size([8, 600, 800])
  2004. torch.Size([8, 8])
  2005. tensor([[0.6203, 0.4073, 0.8189, 0.2398, 0.4400, 0.2054, 0.5929, 0.5501],
  2006. [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690],
  2007. [0.6179, 0.3993, 0.8925, 0.4789, 0.3879, 0.4900, 0.6041, 0.5279],
  2008. [0.6212, 0.4171, 0.7875, 0.3633, 0.3813, 0.2933, 0.5675, 0.5700],
  2009. [0.6203, 0.4078, 0.8800, 0.5083, 0.3900, 0.5000, 0.6100, 0.5583],
  2010. [0.6185, 0.4067, 0.8838, 0.4450, 0.4037, 0.4733, 0.5213, 0.5142],
  2011. [0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633],
  2012. [0.6266, 0.4070, 0.8712, 0.5600, 0.3713, 0.4783, 0.5775, 0.6100]],
  2013. device='cuda:0', dtype=torch.float64)
  2014. predictions are: tensor([[ 0.0306, -0.7482, 1.1684, -1.1746, -0.3463, -1.4168, 0.4507, 0.2861],
  2015. [ 0.4355, -0.4647, 1.4831, -0.1241, -0.5088, -0.0437, 0.4310, 0.1761],
  2016. [ 0.5143, -0.4910, 1.6547, -0.1234, -0.5125, -0.2014, 0.4634, 0.1301],
  2017. [ 0.2105, -0.5922, 1.0609, -0.9738, -0.4900, -0.8628, 0.3837, 0.2377],
  2018. [ 0.5715, -0.4519, 1.6900, 0.0187, -0.5369, -0.0197, 0.4772, 0.1420],
  2019. [ 0.4221, -0.5485, 1.5904, -0.2299, -0.4761, -0.3609, 0.4290, 0.1838],
  2020. [ 0.0825, -0.7490, 1.3565, -1.0026, -0.3237, -1.2647, 0.5705, 0.2463],
  2021. [ 0.5655, -0.4344, 1.5807, 0.0510, -0.5576, 0.0759, 0.3991, 0.1501]],
  2022. device='cuda:0', grad_fn=<AddmmBackward>)
  2023. landmarks are: tensor([[[ 0.5792, -0.4048, 1.4965, -1.1781, -0.2534, -1.3371, 0.4528,
  2024. 0.2549],
  2025. [ 0.6306, -0.4153, 1.5141, 0.2224, -0.3691, 0.2622, 1.0033,
  2026. 0.3425],
  2027. [ 0.5680, -0.4417, 1.8365, -0.0740, -0.4941, -0.0227, 0.5045,
  2028. 0.1525],
  2029. [ 0.5832, -0.3593, 1.3515, -0.6077, -0.5249, -0.9310, 0.3353,
  2030. 0.3469],
  2031. [ 0.5794, -0.4023, 1.7788, 0.0620, -0.4845, 0.0236, 0.5316,
  2032. 0.2930],
  2033. [ 0.5708, -0.4075, 1.7961, -0.2305, -0.4210, -0.0996, 0.1219,
  2034. 0.0893],
  2035. [ 0.6042, -0.4105, 1.5478, -1.2082, -0.1208, -1.0927, 0.9704,
  2036. 0.3157],
  2037. [ 0.6081, -0.4059, 1.7383, 0.3007, -0.5711, -0.0765, 0.3815,
  2038. 0.5316]]], device='cuda:0')
  2039. loss_train_step before backward: tensor(0.0460, device='cuda:0', grad_fn=<MseLossBackward>)
  2040. loss_train_step after backward: tensor(0.0460, device='cuda:0', grad_fn=<MseLossBackward>)
  2041. loss_train: 5.805826131254435
  2042. step: 37
  2043. running loss: 0.15691421976363337
  2044. Train Steps: 37/90 Loss: 0.1569 torch.Size([8, 600, 800])
  2045. torch.Size([8, 8])
  2046. tensor([[0.6188, 0.4099, 0.7400, 0.2433, 0.3962, 0.2750, 0.6162, 0.5467],
  2047. [0.6125, 0.4035, 0.7825, 0.3100, 0.3463, 0.4900, 0.5832, 0.5637],
  2048. [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133],
  2049. [0.6131, 0.4037, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680],
  2050. [0.6300, 0.4133, 0.8538, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413],
  2051. [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810],
  2052. [0.6167, 0.4048, 0.6831, 0.3639, 0.3763, 0.3017, 0.5700, 0.5883],
  2053. [0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317]],
  2054. device='cuda:0', dtype=torch.float64)
  2055. predictions are: tensor([[ 0.3484, -0.5611, 1.1903, -0.8528, -0.5253, -0.8242, 0.4086, 0.2321],
  2056. [ 0.5839, -0.4033, 1.4489, -0.3269, -0.5890, -0.0762, 0.4293, 0.1951],
  2057. [ 0.6881, -0.3780, 1.8892, 0.4334, -0.4735, 0.3656, 0.5026, 0.1625],
  2058. [ 0.3486, -0.5398, 1.1285, -0.8707, -0.5217, -0.8116, 0.3905, 0.2235],
  2059. [ 0.1830, -0.6934, 1.5746, -0.6413, -0.2942, -0.8391, 0.6271, 0.2453],
  2060. [ 0.4457, -0.5310, 1.4380, -0.6530, -0.4888, -0.6945, 0.5348, 0.1828],
  2061. [ 0.3547, -0.5630, 1.3450, -0.6403, -0.4638, -0.7016, 0.4201, 0.2484],
  2062. [ 0.1226, -0.6902, 1.2788, -0.9352, -0.3668, -1.0977, 0.4852, 0.2776]],
  2063. device='cuda:0', grad_fn=<AddmmBackward>)
  2064. landmarks are: tensor([[[ 0.5725, -0.3926, 1.1321, -1.1620, -0.4557, -1.0157, 0.5605,
  2065. 0.2391],
  2066. [ 0.5432, -0.4224, 1.3284, -0.8540, -0.6866, -0.0226, 0.4077,
  2067. 0.3177],
  2068. [ 0.5786, -0.4141, 1.7037, 0.1544, -0.1862, 0.0736, 0.4393,
  2069. 0.0851],
  2070. [ 0.5459, -0.4215, 0.9043, -0.9838, -0.5827, -1.0388, 0.1236,
  2071. 0.3378],
  2072. [ 0.6240, -0.3768, 1.6575, -1.2851, 0.2949, -1.2467, 1.1276,
  2073. 0.2142],
  2074. [ 0.6064, -0.3990, 1.6113, -0.8309, -0.4268, -1.0696, 0.6421,
  2075. -0.0640],
  2076. [ 0.5626, -0.4162, 0.8692, -0.6051, -0.5480, -0.8925, 0.3469,
  2077. 0.4316],
  2078. [ 0.5849, -0.3836, 1.2649, -1.2236, -0.3460, -1.2313, 0.4508,
  2079. 0.1698]]], device='cuda:0')
  2080. loss_train_step before backward: tensor(0.0582, device='cuda:0', grad_fn=<MseLossBackward>)
  2081. loss_train_step after backward: tensor(0.0582, device='cuda:0', grad_fn=<MseLossBackward>)
  2082. loss_train: 5.863999526947737
  2083. step: 38
  2084. running loss: 0.15431577702494045
  2085. Train Steps: 38/90 Loss: 0.1543 torch.Size([8, 600, 800])
  2086. torch.Size([8, 8])
  2087. tensor([[0.6113, 0.4006, 0.8700, 0.5350, 0.3638, 0.3767, 0.5097, 0.4882],
  2088. [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395],
  2089. [0.6286, 0.4060, 0.9188, 0.4333, 0.3675, 0.4167, 0.7034, 0.5528],
  2090. [0.6236, 0.4084, 0.7738, 0.2133, 0.3663, 0.3233, 0.5813, 0.5567],
  2091. [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395],
  2092. [0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400],
  2093. [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200],
  2094. [0.6346, 0.4086, 0.7938, 0.5500, 0.3962, 0.4867, 0.7343, 0.5702]],
  2095. device='cuda:0', dtype=torch.float64)
  2096. predictions are: tensor([[ 0.6287, -0.4028, 1.6203, -0.1966, -0.4968, -0.2871, 0.5465, 0.1927],
  2097. [ 0.5755, -0.4245, 1.5422, -0.3178, -0.4948, -0.3611, 0.5096, 0.1966],
  2098. [ 0.5557, -0.4868, 1.7148, -0.1531, -0.4118, -0.2448, 0.5633, 0.2358],
  2099. [ 0.2434, -0.6120, 1.1403, -1.0868, -0.4222, -1.0653, 0.4561, 0.3103],
  2100. [ 0.2002, -0.6206, 0.9828, -1.2643, -0.4166, -1.2470, 0.4367, 0.3031],
  2101. [ 0.0895, -0.6730, 1.1629, -1.1852, -0.3008, -1.2543, 0.5230, 0.3339],
  2102. [ 0.6912, -0.3737, 1.6513, -0.1489, -0.5177, -0.0689, 0.4943, 0.1897],
  2103. [ 0.5583, -0.4039, 1.4031, -0.2909, -0.5239, -0.1276, 0.4636, 0.2189]],
  2104. device='cuda:0', grad_fn=<AddmmBackward>)
  2105. landmarks are: tensor([[[ 0.5379, -0.4358, 1.7326, 0.1852, -0.6058, -0.5461, 0.0684,
  2106. -0.0310],
  2107. [ 0.5697, -0.4514, 1.7420, 0.2672, -0.6055, -0.6312, 0.3449,
  2108. 0.2058],
  2109. [ 0.6174, -0.4105, 1.9577, -0.2844, -0.5885, -0.3614, 0.9631,
  2110. 0.2676],
  2111. [ 0.5946, -0.3995, 1.2880, -1.3005, -0.5942, -0.7925, 0.3988,
  2112. 0.2853],
  2113. [ 0.5713, -0.4538, 1.1028, -1.3659, -0.3831, -1.1273, 0.5340,
  2114. 0.2058],
  2115. [ 0.5958, -0.3818, 1.5536, -1.1081, -0.2074, -1.3390, 0.5663,
  2116. 0.2083],
  2117. [ 0.5865, -0.3973, 1.8423, -0.0688, -0.5192, -0.2305, 0.4162,
  2118. 0.1159],
  2119. [ 0.6454, -0.3984, 1.3804, 0.2545, -0.4557, -0.0380, 1.1057,
  2120. 0.3478]]], device='cuda:0')
  2121. loss_train_step before backward: tensor(0.0508, device='cuda:0', grad_fn=<MseLossBackward>)
  2122. loss_train_step after backward: tensor(0.0508, device='cuda:0', grad_fn=<MseLossBackward>)
  2123. loss_train: 5.914830353111029
  2124. step: 39
  2125. running loss: 0.15166231674643663
  2126. Train Steps: 39/90 Loss: 0.1517 torch.Size([8, 600, 800])
  2127. torch.Size([8, 8])
  2128. tensor([[0.6254, 0.3993, 0.8988, 0.4767, 0.3987, 0.5517, 0.6955, 0.5285],
  2129. [0.6222, 0.3957, 0.8838, 0.5017, 0.3937, 0.4600, 0.5900, 0.5017],
  2130. [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083],
  2131. [0.6182, 0.4058, 0.8738, 0.4350, 0.3563, 0.3400, 0.5290, 0.5822],
  2132. [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261],
  2133. [0.6357, 0.4118, 0.8400, 0.2500, 0.5413, 0.1633, 0.6725, 0.5586],
  2134. [0.6100, 0.4016, 0.8600, 0.5067, 0.4612, 0.5233, 0.5086, 0.5519],
  2135. [0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400]],
  2136. device='cuda:0', dtype=torch.float64)
  2137. predictions are: tensor([[ 0.6486, -0.3410, 1.5774, -0.2222, -0.4965, -0.0596, 0.5604, 0.2172],
  2138. [ 0.6847, -0.3742, 1.5988, -0.2892, -0.5014, -0.3354, 0.5350, 0.2231],
  2139. [ 0.6656, -0.3219, 1.4963, -0.2681, -0.5249, -0.1370, 0.4741, 0.2231],
  2140. [ 0.5834, -0.4588, 1.4839, -0.5551, -0.5066, -0.6224, 0.4980, 0.2645],
  2141. [ 0.6183, -0.3602, 1.5010, -0.4473, -0.4864, -0.3845, 0.5184, 0.2457],
  2142. [ 0.1681, -0.6630, 1.1836, -1.3643, -0.2738, -1.4343, 0.6444, 0.3350],
  2143. [ 0.5656, -0.3962, 1.4756, -0.3737, -0.4396, -0.3481, 0.4812, 0.2817],
  2144. [ 0.1295, -0.6485, 1.0784, -1.3811, -0.3357, -1.4434, 0.5232, 0.3749]],
  2145. device='cuda:0', grad_fn=<AddmmBackward>)
  2146. landmarks are: tensor([[[ 0.6026, -0.4417, 1.8654, -0.0842, -0.4441, 0.2622, 0.9265,
  2147. 0.1554],
  2148. [ 0.5879, -0.4583, 1.7961, 0.0313, -0.4672, -0.1612, 0.4393,
  2149. 0.0313],
  2150. [ 0.5713, -0.4105, 1.7557, 0.0467, -0.6520, -0.2690, 0.3988,
  2151. 0.5239],
  2152. [ 0.5697, -0.4113, 1.7499, -0.2767, -0.6404, -0.7155, 0.1576,
  2153. 0.4032],
  2154. [ 0.5746, -0.3623, 1.7961, -0.1150, -0.3691, -0.0380, 0.2208,
  2155. 0.1439],
  2156. [ 0.6504, -0.3840, 1.5940, -1.1312, 0.2141, -1.5315, 0.8205,
  2157. 0.2944],
  2158. [ 0.5318, -0.4310, 1.6864, 0.0543, -0.1554, 0.1313, 0.0635,
  2159. 0.2634],
  2160. [ 0.5911, -0.3888, 1.4727, -0.9541, -0.0919, -1.4930, 0.3988,
  2161. 0.2083]]], device='cuda:0')
  2162. loss_train_step before backward: tensor(0.0610, device='cuda:0', grad_fn=<MseLossBackward>)
  2163. loss_train_step after backward: tensor(0.0610, device='cuda:0', grad_fn=<MseLossBackward>)
  2164. loss_train: 5.975822579115629
  2165. step: 40
  2166. running loss: 0.14939556447789074
  2167.  
  2168. Train Steps: 40/90 Loss: 0.1494 torch.Size([8, 600, 800])
  2169. torch.Size([8, 8])
  2170. tensor([[0.6205, 0.4081, 0.8950, 0.4017, 0.3788, 0.4700, 0.5963, 0.5667],
  2171. [0.6200, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913],
  2172. [0.6114, 0.4018, 0.7213, 0.1967, 0.3763, 0.2700, 0.5875, 0.5533],
  2173. [0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019],
  2174. [0.6198, 0.4164, 0.8700, 0.5067, 0.4625, 0.5650, 0.5464, 0.5197],
  2175. [0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317],
  2176. [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220],
  2177. [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749]],
  2178. device='cuda:0', dtype=torch.float64)
  2179. predictions are: tensor([[ 0.7940, -0.2830, 1.7891, -0.1529, -0.4726, -0.1025, 0.5921, 0.2275],
  2180. [ 0.7956, -0.3095, 1.8062, 0.0619, -0.4303, 0.0046, 0.5869, 0.1973],
  2181. [ 0.4080, -0.4914, 1.0629, -1.2522, -0.4499, -1.1842, 0.4631, 0.3219],
  2182. [ 0.8304, -0.2782, 1.7126, -0.1263, -0.4942, -0.1525, 0.5709, 0.2095],
  2183. [ 0.6740, -0.3268, 1.7111, -0.1091, -0.3630, -0.1135, 0.5836, 0.2540],
  2184. [ 0.2240, -0.6083, 1.1813, -1.3228, -0.2818, -1.4002, 0.5638, 0.3649],
  2185. [ 0.2937, -0.5621, 1.0164, -1.4729, -0.3887, -1.4923, 0.4779, 0.3411],
  2186. [ 0.5184, -0.4149, 1.1081, -1.0105, -0.5213, -0.8176, 0.4427, 0.3183]],
  2187. device='cuda:0', grad_fn=<AddmmBackward>)
  2188. landmarks are: tensor([[[ 0.5799, -0.4012, 1.8480, -0.4306, -0.5365, -0.1150, 0.4681,
  2189. 0.3315],
  2190. [ 0.5776, -0.4784, 1.7730, 0.1236, -0.4037, 0.0697, 0.5132,
  2191. -0.0168],
  2192. [ 0.5380, -0.4302, 1.0455, -1.3775, -0.5480, -1.0388, 0.4277,
  2193. 0.2699],
  2194. [ 0.6372, -0.4129, 1.8942, -0.0765, -0.6173, -0.4768, 0.6999,
  2195. 0.0325],
  2196. [ 0.5770, -0.3624, 1.7326, 0.0543, -0.1497, 0.3238, 0.2378,
  2197. 0.1146],
  2198. [ 0.5849, -0.3836, 1.2649, -1.2236, -0.3460, -1.2313, 0.4508,
  2199. 0.1698],
  2200. [ 0.5566, -0.4160, 0.9935, -1.3313, -0.2824, -1.5161, 0.2144,
  2201. 0.1253],
  2202. [ 0.5419, -0.4160, 1.1810, -0.8939, -0.6808, -0.7463, 0.2485,
  2203. 0.3694]]], device='cuda:0')
  2204. loss_train_step before backward: tensor(0.0260, device='cuda:0', grad_fn=<MseLossBackward>)
  2205. loss_train_step after backward: tensor(0.0260, device='cuda:0', grad_fn=<MseLossBackward>)
  2206. loss_train: 6.001800514757633
  2207. step: 41
  2208. running loss: 0.14638537840872276
  2209. Train Steps: 41/90 Loss: 0.1464 torch.Size([8, 600, 800])
  2210. torch.Size([8, 8])
  2211. tensor([[0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5787, 0.5117],
  2212. [0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901],
  2213. [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350],
  2214. [0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5988, 0.5700],
  2215. [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617],
  2216. [0.6212, 0.4033, 0.8938, 0.4167, 0.3813, 0.4267, 0.5613, 0.5583],
  2217. [0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507],
  2218. [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6038, 0.4833]],
  2219. device='cuda:0', dtype=torch.float64)
  2220. predictions are: tensor([[ 0.4199, -0.4943, 1.0790, -1.4109, -0.3797, -1.4113, 0.5004, 0.3263],
  2221. [ 0.7100, -0.3301, 1.7054, -0.2859, -0.3876, -0.3165, 0.5298, 0.2141],
  2222. [ 0.6441, -0.3112, 1.6181, -0.4765, -0.3953, -0.5129, 0.5334, 0.2616],
  2223. [ 0.4550, -0.4527, 1.3070, -1.0951, -0.3657, -1.0771, 0.5272, 0.3358],
  2224. [ 0.7904, -0.2419, 1.7333, -0.0269, -0.4500, -0.0779, 0.5352, 0.2389],
  2225. [ 0.7278, -0.3011, 1.6891, -0.2632, -0.4183, -0.2551, 0.5382, 0.2533],
  2226. [ 0.3967, -0.4765, 0.9523, -1.4774, -0.4310, -1.3825, 0.4442, 0.3253],
  2227. [ 0.7859, -0.2656, 1.7115, -0.2033, -0.4576, -0.1586, 0.5810, 0.1942]],
  2228. device='cuda:0', grad_fn=<AddmmBackward>)
  2229. landmarks are: tensor([[[ 0.5598, -0.4201, 1.1898, -1.3005, -0.3806, -1.3313, 0.3873,
  2230. 0.0774],
  2231. [ 0.5663, -0.3955, 1.7788, -0.2382, -0.4037, -0.2690, 0.0828,
  2232. -0.0220],
  2233. [ 0.6072, -0.3239, 1.8423, -0.3537, -0.4961, -0.3921, 0.2083,
  2234. 0.1852],
  2235. [ 0.5960, -0.3888, 1.4840, -1.0095, -0.6115, -0.6231, 0.4797,
  2236. 0.3469],
  2237. [ 0.5897, -0.3527, 1.8018, 0.2545, -0.5307, -0.3229, 0.3122,
  2238. 0.3084],
  2239. [ 0.5832, -0.4231, 1.8423, -0.3614, -0.5249, -0.3152, 0.3065,
  2240. 0.2930],
  2241. [ 0.5185, -0.4252, 0.9647, -1.2928, -0.4788, -1.2390, 0.2617,
  2242. 0.2576],
  2243. [ 0.5726, -0.4795, 1.7788, -0.0149, -0.5711, -0.0688, 0.5028,
  2244. -0.0534]]], device='cuda:0')
  2245. loss_train_step before backward: tensor(0.0263, device='cuda:0', grad_fn=<MseLossBackward>)
  2246. loss_train_step after backward: tensor(0.0263, device='cuda:0', grad_fn=<MseLossBackward>)
  2247. loss_train: 6.028124667704105
  2248. step: 42
  2249. running loss: 0.1435267778024787
  2250. Train Steps: 42/90 Loss: 0.1435 torch.Size([8, 600, 800])
  2251. torch.Size([8, 8])
  2252. tensor([[0.6102, 0.4005, 0.8688, 0.5100, 0.4813, 0.5400, 0.5404, 0.5064],
  2253. [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150],
  2254. [ nan, nan, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609],
  2255. [0.6218, 0.4098, 0.7238, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350],
  2256. [0.6346, 0.4086, 0.7938, 0.5500, 0.3962, 0.4867, 0.7343, 0.5702],
  2257. [0.6091, 0.3997, 0.8314, 0.4334, 0.3788, 0.4550, 0.5213, 0.5656],
  2258. [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383],
  2259. [0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950]],
  2260. device='cuda:0', dtype=torch.float64)
  2261. predictions are: tensor([[ 0.6704, -0.3256, 1.6446, -0.3800, -0.3220, -0.4287, 0.5034, 0.2098],
  2262. [ 0.7600, -0.2821, 1.6958, -0.3735, -0.4332, -0.5067, 0.5104, 0.2388],
  2263. [ 0.4023, -0.5221, 1.4291, -1.2232, -0.2208, -1.2984, 0.6148, 0.2990],
  2264. [ 0.4486, -0.4718, 1.0584, -1.5126, -0.3953, -1.4553, 0.4707, 0.3244],
  2265. [ 0.7212, -0.2695, 1.4876, -0.3751, -0.4994, -0.2967, 0.4663, 0.2486],
  2266. [ 0.7195, -0.2523, 1.4674, -0.6101, -0.5033, -0.4567, 0.4606, 0.2342],
  2267. [ 0.7293, -0.2780, 1.6936, -0.2997, -0.3912, -0.3453, 0.5069, 0.1911],
  2268. [ 0.7337, -0.2971, 1.6755, -0.2719, -0.3963, -0.3425, 0.4923, 0.2010]],
  2269. device='cuda:0', grad_fn=<AddmmBackward>)
  2270. landmarks are: tensor([[[ 5.3279e-01, -4.3610e-01, 1.7268e+00, 6.9746e-02, -6.3048e-02,
  2271. 2.0831e-01, 2.1029e-01, 5.3181e-02],
  2272. [ 6.0964e-01, -4.0462e-01, 1.8249e+00, -7.2363e-03, -6.1155e-01,
  2273. -3.5366e-01, 6.1824e-01, 9.2841e-02],
  2274. [-2.2859e+00, -2.2859e+00, 1.8249e+00, -8.0015e-01, 4.0878e-02,
  2275. -1.2543e+00, 8.0590e-01, 3.0505e-01],
  2276. [ 5.8614e-01, -3.9292e-01, 1.0570e+00, -1.4314e+00, -3.2864e-01,
  2277. -1.1235e+00, 6.1824e-01, 1.8522e-01],
  2278. [ 6.4542e-01, -3.9842e-01, 1.3804e+00, 2.5450e-01, -4.5566e-01,
  2279. -3.8029e-02, 1.1057e+00, 3.4780e-01],
  2280. [ 5.2748e-01, -4.3957e-01, 1.5543e+00, -2.8408e-01, -5.3649e-01,
  2281. -1.8430e-01, 1.2208e-01, 3.2654e-01],
  2282. [ 5.3603e-01, -4.6490e-01, 1.6517e+00, 4.6189e-04, -2.8245e-01,
  2283. -6.8822e-02, 2.3086e-01, 2.0046e-01],
  2284. [ 5.8199e-01, -4.7544e-01, 1.7095e+00, 1.9292e-01, -2.5358e-01,
  2285. 5.4350e-02, 4.9700e-01, 4.6189e-04]]], device='cuda:0')
  2286. loss_train_step before backward: tensor(0.2217, device='cuda:0', grad_fn=<MseLossBackward>)
  2287. loss_train_step after backward: tensor(0.2217, device='cuda:0', grad_fn=<MseLossBackward>)
  2288. loss_train: 6.249815843999386
  2289. step: 43
  2290. running loss: 0.14534455451161363
  2291. Train Steps: 43/90 Loss: 0.1453 torch.Size([8, 600, 800])
  2292. torch.Size([8, 8])
  2293. tensor([[0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6262, 0.5167],
  2294. [0.6200, 0.4112, 0.8862, 0.4100, 0.3638, 0.4917, 0.6088, 0.6050],
  2295. [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200],
  2296. [ nan, nan, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819],
  2297. [0.6189, 0.4029, 0.8375, 0.5767, 0.4745, 0.4829, 0.5551, 0.5598],
  2298. [ nan, nan, 0.6793, 0.2110, 0.4012, 0.2167, 0.5112, 0.5583],
  2299. [ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600],
  2300. [0.6199, 0.4102, 0.8950, 0.4417, 0.4012, 0.5367, 0.6112, 0.5967]],
  2301. device='cuda:0', dtype=torch.float64)
  2302. predictions are: tensor([[ 0.7581, -0.3525, 1.8243, -0.3983, -0.3701, -0.6395, 0.5296, 0.1889],
  2303. [ 0.8480, -0.2333, 1.9051, -0.0124, -0.4633, -0.0062, 0.5412, 0.1463],
  2304. [ 0.8636, -0.2277, 1.8561, -0.0840, -0.4672, -0.1841, 0.4694, 0.1467],
  2305. [ 0.4291, -0.4614, 1.0137, -1.4131, -0.4086, -1.3380, 0.3718, 0.2872],
  2306. [ 0.6720, -0.2897, 1.6456, -0.1900, -0.3317, -0.2876, 0.4857, 0.1960],
  2307. [ 0.3500, -0.5126, 0.9928, -1.5114, -0.3518, -1.4417, 0.4003, 0.3029],
  2308. [ 0.3663, -0.4884, 1.2399, -1.2750, -0.2920, -1.2833, 0.4599, 0.3037],
  2309. [ 0.7835, -0.2698, 1.8654, 0.0564, -0.3705, 0.0656, 0.5092, 0.1624]],
  2310. device='cuda:0', grad_fn=<AddmmBackward>)
  2311. landmarks are: tensor([[[ 0.6094, -0.3947, 1.8885, -0.2998, -0.5769, -0.6770, 0.6067,
  2312. 0.1005],
  2313. [ 0.5777, -0.3868, 1.8076, -0.3921, -0.6058, -0.0149, 0.5259,
  2314. 0.5085],
  2315. [ 0.5865, -0.3973, 1.8423, -0.0688, -0.5192, -0.2305, 0.4162,
  2316. 0.1159],
  2317. [-2.2859, -2.2859, 0.9438, -0.9967, -0.4614, -1.1851, 0.2468,
  2318. 0.4019],
  2319. [ 0.5726, -0.4249, 1.5824, 0.3777, -0.0942, -0.0556, 0.2782,
  2320. 0.2997],
  2321. [-2.2859, -2.2859, 0.8516, -1.3112, -0.4326, -1.2851, 0.0755,
  2322. 0.2930],
  2323. [-2.2859, -2.2859, 1.4266, -1.1389, -0.4499, -1.1235, 0.2891,
  2324. 0.3007],
  2325. [ 0.5772, -0.3913, 1.8480, -0.2459, -0.4326, 0.1929, 0.5374,
  2326. 0.4701]]], device='cuda:0')
  2327. loss_train_step before backward: tensor(0.5159, device='cuda:0', grad_fn=<MseLossBackward>)
  2328. loss_train_step after backward: tensor(0.5159, device='cuda:0', grad_fn=<MseLossBackward>)
  2329.  
  2330. loss_train: 6.765694998204708
  2331. step: 44
  2332. running loss: 0.15376579541374336
  2333. Train Steps: 44/90 Loss: 0.1538 torch.Size([8, 600, 800])
  2334. torch.Size([8, 8])
  2335. tensor([[0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748],
  2336. [0.6273, 0.4143, 0.8750, 0.5700, 0.3987, 0.4717, 0.6013, 0.5467],
  2337. [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650],
  2338. [0.6200, 0.3999, 0.8653, 0.5207, 0.4100, 0.5125, 0.5975, 0.5103],
  2339. [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333],
  2340. [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389],
  2341. [0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222],
  2342. [0.6111, 0.4033, 0.8300, 0.3267, 0.3588, 0.3333, 0.5444, 0.5637]],
  2343. device='cuda:0', dtype=torch.float64)
  2344. predictions are: tensor([[ 0.5681, -0.3922, 1.6770, -0.3190, -0.3526, -0.5051, 0.4730, 0.1606],
  2345. [ 0.6769, -0.3203, 1.6853, -0.1504, -0.4270, -0.2334, 0.4160, 0.1653],
  2346. [ 0.4975, -0.4291, 1.5476, -0.6745, -0.3673, -0.7478, 0.4353, 0.2326],
  2347. [ 0.7256, -0.3172, 1.7803, -0.0423, -0.4017, -0.0852, 0.4411, 0.1329],
  2348. [ 0.7370, -0.3005, 1.8452, -0.0926, -0.4353, -0.1475, 0.4734, 0.1246],
  2349. [ 0.3213, -0.5524, 1.2729, -1.2532, -0.2965, -1.3091, 0.4605, 0.2714],
  2350. [ 0.2930, -0.5338, 1.3079, -1.0069, -0.2362, -0.9552, 0.4142, 0.2860],
  2351. [ 0.5536, -0.4193, 1.4912, -0.7955, -0.4314, -0.8450, 0.4179, 0.2130]],
  2352. device='cuda:0', grad_fn=<AddmmBackward>)
  2353. landmarks are: tensor([[[ 0.6421, -0.3647, 1.5940, 0.3084, -0.5192, -0.3691, 1.1057,
  2354. 0.3692],
  2355. [ 0.6115, -0.3724, 1.7557, 0.3469, -0.4441, -0.1073, 0.4912,
  2356. 0.2391],
  2357. [ 0.5894, -0.3503, 1.6402, -0.3614, -0.5827, -0.7925, 0.3238,
  2358. 0.3238],
  2359. [ 0.5778, -0.4389, 1.7107, 0.1192, -0.3921, 0.0815, 0.4741,
  2360. 0.0711],
  2361. [ 0.6003, -0.3747, 1.8885, -0.1997, -0.5018, -0.0149, 0.5490,
  2362. 0.1775],
  2363. [ 0.5697, -0.4466, 1.1973, -1.1871, -0.4571, -0.9965, 0.5219,
  2364. 0.2032],
  2365. [ 0.6237, -0.2983, 1.3919, -0.4691, -0.0457, -1.2313, 0.2453,
  2366. 0.5882],
  2367. [ 0.5366, -0.4232, 1.5478, -0.7771, -0.6289, -0.7463, 0.2288,
  2368. 0.3177]]], device='cuda:0')
  2369. loss_train_step before backward: tensor(0.0398, device='cuda:0', grad_fn=<MseLossBackward>)
  2370. loss_train_step after backward: tensor(0.0398, device='cuda:0', grad_fn=<MseLossBackward>)
  2371. loss_train: 6.80553362518549
  2372. step: 45
  2373. running loss: 0.15123408055967755
  2374. Train Steps: 45/90 Loss: 0.1512 torch.Size([8, 600, 800])
  2375. torch.Size([8, 8])
  2376. tensor([[0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209],
  2377. [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5413, 0.5717],
  2378. [0.6200, 0.4118, 0.8287, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799],
  2379. [0.6259, 0.4156, 0.8812, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960],
  2380. [0.6142, 0.3982, 0.8650, 0.4883, 0.3912, 0.4317, 0.5315, 0.5350],
  2381. [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133],
  2382. [0.6145, 0.4008, 0.8750, 0.5383, 0.3975, 0.4650, 0.5563, 0.5533],
  2383. [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646]],
  2384. device='cuda:0', dtype=torch.float64)
  2385. predictions are: tensor([[ 0.3271, -0.5318, 1.5833, -0.6055, -0.2818, -0.7147, 0.4776, 0.2189],
  2386. [ 0.6006, -0.3905, 1.7411, -0.1581, -0.3956, -0.1973, 0.4302, 0.1447],
  2387. [ 0.3604, -0.5165, 1.4272, -0.8646, -0.3821, -0.9327, 0.3752, 0.2403],
  2388. [ 0.1835, -0.6378, 1.4985, -1.1170, -0.1787, -1.2397, 0.5342, 0.2482],
  2389. [ 0.6568, -0.3744, 1.8330, -0.0872, -0.4559, -0.2435, 0.4343, 0.1183],
  2390. [ 0.5622, -0.3841, 1.6823, -0.1538, -0.3768, -0.2230, 0.3884, 0.1781],
  2391. [ 0.6354, -0.3549, 1.7240, -0.0143, -0.4266, -0.0859, 0.4077, 0.1498],
  2392. [ 0.3661, -0.5067, 1.3478, -0.9482, -0.4265, -0.9791, 0.3621, 0.2470]],
  2393. device='cuda:0', grad_fn=<AddmmBackward>)
  2394. landmarks are: tensor([[[ 0.6174, -0.3118, 1.6402, -0.2074, -0.1958, -1.0927, 0.2267,
  2395. 0.5822],
  2396. [ 0.5432, -0.4462, 1.7557, -0.0380, -0.3113, -0.0765, 0.2141,
  2397. 0.3546],
  2398. [ 0.5779, -0.3840, 1.5420, -0.4306, -0.5423, -0.9772, 0.2041,
  2399. 0.3928],
  2400. [ 0.6052, -0.3663, 1.7845, -0.8156, -0.0804, -1.4237, 0.5866,
  2401. 0.0051],
  2402. [ 0.5513, -0.4467, 1.7095, -0.0303, -0.4788, -0.2921, 0.1692,
  2403. 0.1854],
  2404. [ 0.5786, -0.4141, 1.7037, 0.1544, -0.1862, 0.0736, 0.4393,
  2405. 0.0851],
  2406. [ 0.5526, -0.4347, 1.7557, 0.2006, -0.4499, -0.1381, 0.2834,
  2407. 0.2699],
  2408. [ 0.5483, -0.4105, 1.4208, -0.8002, -0.6000, -0.9002, 0.0511,
  2409. 0.3220]]], device='cuda:0')
  2410. loss_train_step before backward: tensor(0.0359, device='cuda:0', grad_fn=<MseLossBackward>)
  2411. loss_train_step after backward: tensor(0.0359, device='cuda:0', grad_fn=<MseLossBackward>)
  2412. loss_train: 6.841406472027302
  2413. step: 46
  2414. running loss: 0.14872622765276744
  2415. Train Steps: 46/90 Loss: 0.1487 torch.Size([8, 600, 800])
  2416. torch.Size([8, 8])
  2417. tensor([[0.6336, 0.4191, 0.8938, 0.5167, 0.3937, 0.3517, 0.7343, 0.5748],
  2418. [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617],
  2419. [0.6111, 0.4033, 0.8300, 0.3267, 0.3588, 0.3333, 0.5444, 0.5637],
  2420. [ nan, nan, 0.7612, 0.3250, 0.4037, 0.2533, 0.5438, 0.5767],
  2421. [0.6126, 0.3954, 0.8538, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350],
  2422. [0.6097, 0.4000, 0.7325, 0.2667, 0.3450, 0.3517, 0.5284, 0.5045],
  2423. [0.6250, 0.4013, 0.8525, 0.5417, 0.4037, 0.5117, 0.6325, 0.5017],
  2424. [0.6329, 0.4196, 0.9238, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748]],
  2425. device='cuda:0', dtype=torch.float64)
  2426. predictions are: tensor([[ 0.3840, -0.5315, 1.8182, -0.2895, -0.2842, -0.6059, 0.4778, 0.1746],
  2427. [ 0.5873, -0.3852, 1.7610, 0.0130, -0.4095, -0.0445, 0.3412, 0.1994],
  2428. [ 0.4079, -0.5127, 1.5575, -0.6584, -0.4286, -0.7907, 0.3493, 0.2210],
  2429. [ 0.1248, -0.6520, 1.2992, -1.0770, -0.2927, -1.1115, 0.3463, 0.2811],
  2430. [ 0.5570, -0.4181, 1.8028, -0.0158, -0.3737, -0.1759, 0.3713, 0.1611],
  2431. [ 0.3128, -0.5459, 1.2268, -0.9812, -0.4740, -0.9341, 0.2947, 0.2475],
  2432. [ 0.6236, -0.3824, 1.7676, 0.0590, -0.4433, -0.0679, 0.3503, 0.1704],
  2433. [ 0.3081, -0.5636, 1.7776, -0.4745, -0.2314, -0.7536, 0.5243, 0.2039]],
  2434. device='cuda:0', grad_fn=<AddmmBackward>)
  2435. landmarks are: tensor([[[ 0.6405, -0.3503, 1.8423, 0.1005, -0.4672, -0.6616, 1.1057,
  2436. 0.3692],
  2437. [ 0.5637, -0.4143, 1.7519, -0.0787, -0.3055, -0.0149, 0.3758,
  2438. 0.3084],
  2439. [ 0.5366, -0.4232, 1.5478, -0.7771, -0.6289, -0.7463, 0.2288,
  2440. 0.3177],
  2441. [-2.2859, -2.2859, 1.2303, -0.7848, -0.4210, -1.1158, 0.2256,
  2442. 0.3777],
  2443. [ 0.5435, -0.4597, 1.6575, 0.0159, -0.3229, -0.1150, 0.1877,
  2444. 0.1854],
  2445. [ 0.5303, -0.4384, 1.0975, -1.0542, -0.6924, -0.6616, 0.1548,
  2446. 0.0442],
  2447. [ 0.6010, -0.4321, 1.6517, 0.2160, -0.4210, 0.0774, 0.6356,
  2448. 0.0313],
  2449. [ 0.6372, -0.3479, 1.9808, -0.1689, -0.3171, -0.9310, 1.1057,
  2450. 0.3692]]], device='cuda:0')
  2451. loss_train_step before backward: tensor(0.1665, device='cuda:0', grad_fn=<MseLossBackward>)
  2452. loss_train_step after backward: tensor(0.1665, device='cuda:0', grad_fn=<MseLossBackward>)
  2453. loss_train: 7.0079174265265465
  2454. step: 47
  2455. running loss: 0.1491046260963095
  2456. Train Steps: 47/90 Loss: 0.1491 torch.Size([8, 600, 800])
  2457. torch.Size([8, 8])
  2458. tensor([[ nan, nan, 0.7512, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617],
  2459. [0.6200, 0.4098, 0.8237, 0.2917, 0.4012, 0.2967, 0.6000, 0.5683],
  2460. [0.6250, 0.4236, 0.8638, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799],
  2461. [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391],
  2462. [ nan, nan, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650],
  2463. [0.6273, 0.4105, 0.8988, 0.4517, 0.3912, 0.2550, 0.5894, 0.4811],
  2464. [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320],
  2465. [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5963, 0.6217]],
  2466. device='cuda:0', dtype=torch.float64)
  2467. predictions are: tensor([[ 0.0535, -0.7069, 1.2344, -0.9993, -0.3315, -1.0784, 0.2902, 0.2823],
  2468. [ 0.2657, -0.5997, 1.6602, -0.5471, -0.2962, -0.7206, 0.4205, 0.2333],
  2469. [ 0.3856, -0.5336, 1.8462, -0.1078, -0.3412, -0.3655, 0.4130, 0.2259],
  2470. [ 0.3312, -0.5656, 1.4524, -0.7236, -0.4373, -0.8358, 0.3779, 0.2153],
  2471. [-0.0348, -0.7601, 1.2570, -1.0044, -0.2545, -1.0930, 0.3087, 0.3088],
  2472. [ 0.5419, -0.4685, 2.0130, 0.1167, -0.3741, -0.3086, 0.4475, 0.1471],
  2473. [ 0.8008, -0.2977, 1.9959, 0.6074, -0.5156, 0.4412, 0.3769, 0.1513],
  2474. [ 0.2861, -0.5947, 1.3415, -0.6672, -0.4133, -0.7388, 0.2971, 0.2424]],
  2475. device='cuda:0', grad_fn=<AddmmBackward>)
  2476. landmarks are: tensor([[[-2.2859, -2.2859, 1.1841, -1.3082, -0.3055, -1.3621, 0.3007,
  2477. 0.3084],
  2478. [ 0.5778, -0.3930, 1.5189, -0.9387, -0.4326, -0.9156, 0.4855,
  2479. 0.3392],
  2480. [ 0.6008, -0.3293, 1.7037, -0.5461, -0.4152, -0.8309, 0.3234,
  2481. 0.3928],
  2482. [ 0.5785, -0.4259, 1.4228, -1.0261, -0.4190, -1.2189, 0.4763,
  2483. 0.2043],
  2484. [-2.2859, -2.2859, 0.9175, -1.3947, -0.3691, -1.2467, 0.2314,
  2485. 0.3238],
  2486. [ 0.6116, -0.3898, 1.8654, -0.1997, -0.4788, -1.1081, 0.4367,
  2487. -0.0637],
  2488. [ 0.6273, -0.4393, 1.6402, 0.1313, -0.5076, 0.0467, 1.1532,
  2489. 0.1715],
  2490. [ 0.5687, -0.4088, 1.0397, -1.1466, -0.3113, -1.1928, 0.4681,
  2491. 0.5855]]], device='cuda:0')
  2492. loss_train_step before backward: tensor(0.3112, device='cuda:0', grad_fn=<MseLossBackward>)
  2493.  
  2494. loss_train_step after backward: tensor(0.3112, device='cuda:0', grad_fn=<MseLossBackward>)
  2495. loss_train: 7.319082103669643
  2496. step: 48
  2497. running loss: 0.15248087715978423
  2498. Train Steps: 48/90 Loss: 0.1525 torch.Size([8, 600, 800])
  2499. torch.Size([8, 8])
  2500. tensor([[0.6075, 0.4007, 0.8275, 0.4917, 0.4050, 0.5100, 0.5167, 0.5280],
  2501. [0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166],
  2502. [0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742],
  2503. [0.6371, 0.4092, 0.8337, 0.5850, 0.3950, 0.5117, 0.6559, 0.5262],
  2504. [0.6109, 0.4015, 0.7668, 0.3639, 0.3513, 0.3667, 0.5200, 0.5641],
  2505. [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012],
  2506. [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533],
  2507. [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250]],
  2508. device='cuda:0', dtype=torch.float64)
  2509. predictions are: tensor([[ 0.2065, -0.6275, 1.5711, -0.3425, -0.3605, -0.5036, 0.3276, 0.2431],
  2510. [ 0.3137, -0.5515, 1.6447, -0.2573, -0.4272, -0.3420, 0.3673, 0.2457],
  2511. [ 0.3199, -0.6279, 1.8709, -0.2268, -0.3714, -0.7510, 0.4436, 0.2008],
  2512. [ 0.3236, -0.5680, 1.6676, -0.1668, -0.4192, -0.4400, 0.3268, 0.2124],
  2513. [ 0.1441, -0.6837, 1.2961, -0.8674, -0.4645, -0.9738, 0.2507, 0.2833],
  2514. [ 0.4580, -0.5463, 1.8016, -0.0027, -0.4537, -0.3965, 0.3903, 0.1912],
  2515. [ 0.2008, -0.6386, 1.6768, -0.3726, -0.3226, -0.5094, 0.3750, 0.2283],
  2516. [ 0.3267, -0.5803, 1.8235, -0.1224, -0.3388, -0.3754, 0.4049, 0.2350]],
  2517. device='cuda:0', grad_fn=<AddmmBackward>)
  2518. landmarks are: tensor([[[ 0.5200, -0.4353, 1.5363, -0.0149, -0.4152, 0.0697, 0.1005,
  2519. 0.1530],
  2520. [ 0.5944, -0.4490, 1.8643, -0.0659, -0.5147, 0.1235, 0.7684,
  2521. 0.1004],
  2522. [ 0.6113, -0.4057, 1.7499, 0.3007, -0.5885, -0.5384, 0.4513,
  2523. -0.0957],
  2524. [ 0.6569, -0.3960, 1.5651, 0.4162, -0.4614, 0.0774, 0.7438,
  2525. 0.1447],
  2526. [ 0.5357, -0.4315, 1.2557, -0.6051, -0.6635, -0.5923, 0.1159,
  2527. 0.3198],
  2528. [ 0.6249, -0.4352, 1.8018, 0.2545, -0.6173, -0.1997, 0.6401,
  2529. 0.0291],
  2530. [ 0.5499, -0.4225, 1.8018, -0.2921, -0.3055, 0.0543, 0.4046,
  2531. 0.2699],
  2532. [ 0.5991, -0.3803, 1.8018, -0.0534, -0.3460, 0.1852, 0.5374,
  2533. 0.1390]]], device='cuda:0')
  2534. loss_train_step before backward: tensor(0.0701, device='cuda:0', grad_fn=<MseLossBackward>)
  2535. loss_train_step after backward: tensor(0.0701, device='cuda:0', grad_fn=<MseLossBackward>)
  2536. loss_train: 7.389189012348652
  2537. step: 49
  2538. running loss: 0.1507997757622174
  2539. Train Steps: 49/90 Loss: 0.1508 torch.Size([8, 600, 800])
  2540. torch.Size([8, 8])
  2541. tensor([[0.6263, 0.4039, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804],
  2542. [0.6266, 0.4067, 0.8588, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267],
  2543. [0.6193, 0.4034, 0.7757, 0.2347, 0.3733, 0.2919, 0.5930, 0.4926],
  2544. [0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967],
  2545. [0.6185, 0.4098, 0.8838, 0.4900, 0.4537, 0.5800, 0.6288, 0.5400],
  2546. [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131],
  2547. [0.6076, 0.3953, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954],
  2548. [0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200]],
  2549. device='cuda:0', dtype=torch.float64)
  2550. predictions are: tensor([[ 0.5294, -0.4750, 1.7715, 0.1190, -0.5351, -0.1170, 0.4085, 0.2083],
  2551. [-0.0845, -0.8747, 1.6868, -0.7018, -0.2010, -1.1213, 0.4768, 0.2605],
  2552. [ 0.0189, -0.7988, 1.2685, -0.9018, -0.4537, -1.1003, 0.2813, 0.2620],
  2553. [ 0.4374, -0.4821, 1.6311, -0.0043, -0.5430, -0.1109, 0.3076, 0.2101],
  2554. [ 0.3483, -0.5449, 1.8120, 0.2408, -0.3022, 0.0264, 0.3682, 0.2554],
  2555. [ 0.4092, -0.5825, 1.6334, -0.1759, -0.5376, -0.3951, 0.3223, 0.2299],
  2556. [ 0.3080, -0.5890, 1.4904, -0.3696, -0.5447, -0.5251, 0.2846, 0.2454],
  2557. [-0.0895, -0.8739, 1.7312, -0.6509, -0.1901, -1.0867, 0.4873, 0.2669]],
  2558. device='cuda:0', grad_fn=<AddmmBackward>)
  2559. landmarks are: tensor([[[ 0.6072, -0.4206, 1.8711, -0.2536, -0.6115, -0.1304, 0.6812,
  2560. -0.0670],
  2561. [ 0.6084, -0.4076, 1.6806, -0.9618, -0.2998, -0.9695, 0.6356,
  2562. 0.1467],
  2563. [ 0.5746, -0.4225, 1.2968, -1.2019, -0.5615, -0.9374, 0.4533,
  2564. -0.0108],
  2565. [ 0.4974, -0.4482, 1.6633, -0.3306, -0.6173, 0.1313, 0.2925,
  2566. 0.0081],
  2567. [ 0.5709, -0.3933, 1.7961, -0.0226, -0.1901, 0.3931, 0.6182,
  2568. 0.2083],
  2569. [ 0.5783, -0.4216, 1.6847, -0.5078, -0.6732, -0.5377, 0.4752,
  2570. 0.0839],
  2571. [ 0.5206, -0.4603, 1.4670, -0.4768, -0.6577, -0.3998, 0.1836,
  2572. 0.0021],
  2573. [ 0.6089, -0.3948, 1.7383, -0.8617, -0.2536, -1.2390, 0.6009,
  2574. 0.1159]]], device='cuda:0')
  2575. loss_train_step before backward: tensor(0.0571, device='cuda:0', grad_fn=<MseLossBackward>)
  2576. loss_train_step after backward: tensor(0.0571, device='cuda:0', grad_fn=<MseLossBackward>)
  2577. loss_train: 7.446262441575527
  2578. step: 50
  2579. running loss: 0.14892524883151054
  2580. Train Steps: 50/90 Loss: 0.1489 torch.Size([8, 600, 800])
  2581. torch.Size([8, 8])
  2582. tensor([[0.6113, 0.4088, 0.6859, 0.2208, 0.4363, 0.1700, 0.5188, 0.5533],
  2583. [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167],
  2584. [0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283],
  2585. [0.6268, 0.4102, 0.8938, 0.3667, 0.4025, 0.2833, 0.6275, 0.5183],
  2586. [0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317],
  2587. [0.6361, 0.4076, 0.8862, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297],
  2588. [0.6304, 0.4024, 0.8925, 0.4800, 0.3937, 0.4817, 0.7485, 0.5297],
  2589. [0.6132, 0.4118, 0.8200, 0.3633, 0.3563, 0.5400, 0.5787, 0.5136]],
  2590. device='cuda:0', dtype=torch.float64)
  2591. predictions are: tensor([[-0.1917, -0.8912, 1.1326, -1.0317, -0.4090, -1.2113, 0.2342, 0.2647],
  2592. [ 0.1447, -0.7247, 1.6322, -0.4428, -0.4324, -0.7945, 0.3599, 0.2245],
  2593. [ 0.3594, -0.6005, 1.9348, -0.0460, -0.4138, -0.4227, 0.5189, 0.1961],
  2594. [ 0.1087, -0.7685, 1.8399, -0.3096, -0.2694, -0.7375, 0.5243, 0.2236],
  2595. [-0.2596, -0.9368, 1.3766, -0.8925, -0.2801, -1.1774, 0.3548, 0.2806],
  2596. [ 0.7107, -0.3776, 1.8623, 0.5153, -0.5687, 0.2528, 0.3826, 0.1827],
  2597. [ 0.5197, -0.4508, 1.6900, 0.1712, -0.5216, 0.0608, 0.3766, 0.1881],
  2598. [ 0.3376, -0.5243, 1.4465, -0.2484, -0.5582, -0.1278, 0.3285, 0.2552]],
  2599. device='cuda:0', grad_fn=<AddmmBackward>)
  2600. landmarks are: tensor([[[ 0.5377, -0.3978, 0.8824, -1.2663, -0.2709, -1.5007, 0.1102,
  2601. 0.2699],
  2602. [ 0.5483, -0.4249, 1.5305, -0.7386, -0.6115, -0.8694, 0.3353,
  2603. 0.1005],
  2604. [ 0.6073, -0.4058, 1.8885, -0.4999, -0.5942, -0.4768, 0.6413,
  2605. 0.1544],
  2606. [ 0.6092, -0.3913, 1.8423, -0.5923, -0.4268, -0.9772, 0.6125,
  2607. 0.1082],
  2608. [ 0.5849, -0.3836, 1.2649, -1.2236, -0.3460, -1.2313, 0.4508,
  2609. 0.1698],
  2610. [ 0.6520, -0.4032, 1.8076, 0.1852, -0.5711, -0.1381, 0.7876,
  2611. 0.1608],
  2612. [ 0.6257, -0.4273, 1.8365, -0.0688, -0.4672, -0.0611, 1.1715,
  2613. 0.1608],
  2614. [ 0.5466, -0.3840, 1.5016, -0.6077, -0.6404, 0.2083, 0.3871,
  2615. 0.0862]]], device='cuda:0')
  2616. loss_train_step before backward: tensor(0.0788, device='cuda:0', grad_fn=<MseLossBackward>)
  2617. loss_train_step after backward: tensor(0.0788, device='cuda:0', grad_fn=<MseLossBackward>)
  2618. loss_train: 7.525062687695026
  2619. step: 51
  2620. running loss: 0.14755024877833386
  2621. Train Steps: 51/90 Loss: 0.1476 torch.Size([8, 600, 800])
  2622. torch.Size([8, 8])
  2623. tensor([[0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033],
  2624. [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650],
  2625. [0.6266, 0.4070, 0.8712, 0.5600, 0.3713, 0.4783, 0.5775, 0.6100],
  2626. [0.6111, 0.4033, 0.8300, 0.3267, 0.3588, 0.3333, 0.5444, 0.5637],
  2627. [0.6109, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117],
  2628. [0.6203, 0.4073, 0.8189, 0.2398, 0.4400, 0.2054, 0.5929, 0.5501],
  2629. [0.6196, 0.4088, 0.8888, 0.4583, 0.4500, 0.5683, 0.6138, 0.5883],
  2630. [0.6176, 0.3911, 0.8738, 0.4217, 0.3488, 0.4033, 0.6025, 0.4817]],
  2631. device='cuda:0', dtype=torch.float64)
  2632. predictions are: tensor([[ 0.3924, -0.5213, 1.6944, 0.0514, -0.4727, -0.0790, 0.4196, 0.1729],
  2633. [ 0.0863, -0.7315, 1.6527, -0.4401, -0.4479, -0.7708, 0.4044, 0.2342],
  2634. [ 0.5413, -0.4434, 1.6979, 0.2409, -0.5767, 0.0799, 0.3926, 0.2002],
  2635. [ 0.1163, -0.7415, 1.5643, -0.5669, -0.5108, -0.8526, 0.3755, 0.2119],
  2636. [ 0.3633, -0.6097, 1.8357, -0.0659, -0.4941, -0.4124, 0.4238, 0.1571],
  2637. [-0.2876, -0.9784, 1.4261, -0.9724, -0.2718, -1.3018, 0.4072, 0.2629],
  2638. [ 0.2977, -0.5920, 1.6902, -0.0351, -0.3639, -0.1044, 0.4685, 0.2399],
  2639. [ 0.3709, -0.5620, 1.5785, -0.3152, -0.6058, -0.4485, 0.4069, 0.1557]],
  2640. device='cuda:0', grad_fn=<AddmmBackward>)
  2641. landmarks are: tensor([[[ 0.5668, -0.4306, 1.7730, 0.0697, -0.4037, 0.1390, 0.4450,
  2642. 0.0390],
  2643. [ 0.5894, -0.3503, 1.6402, -0.3614, -0.5827, -0.7925, 0.3238,
  2644. 0.3238],
  2645. [ 0.6081, -0.4059, 1.7383, 0.3007, -0.5711, -0.0765, 0.3815,
  2646. 0.5316],
  2647. [ 0.5366, -0.4232, 1.5478, -0.7771, -0.6289, -0.7463, 0.2288,
  2648. 0.3177],
  2649. [ 0.5356, -0.4343, 1.7441, -0.1150, -0.6462, -0.3614, 0.3238,
  2650. 0.0774],
  2651. [ 0.5792, -0.4048, 1.4965, -1.1781, -0.2534, -1.3371, 0.4528,
  2652. 0.2549],
  2653. [ 0.5759, -0.3978, 1.8192, -0.1689, -0.2074, 0.3392, 0.5490,
  2654. 0.4316],
  2655. [ 0.5669, -0.4794, 1.7499, -0.3383, -0.6751, -0.4229, 0.4970,
  2656. -0.0611]]], device='cuda:0')
  2657. loss_train_step before backward: tensor(0.0449, device='cuda:0', grad_fn=<MseLossBackward>)
  2658.  
  2659. loss_train_step after backward: tensor(0.0449, device='cuda:0', grad_fn=<MseLossBackward>)
  2660. loss_train: 7.569976683706045
  2661. step: 52
  2662. running loss: 0.14557647468665472
  2663. Train Steps: 52/90 Loss: 0.1456 torch.Size([8, 600, 800])
  2664. torch.Size([8, 8])
  2665. tensor([[0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274],
  2666. [0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117],
  2667. [ nan, nan, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650],
  2668. [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646],
  2669. [0.6157, 0.4102, 0.8513, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890],
  2670. [0.6268, 0.4052, 0.8175, 0.2250, 0.4688, 0.1917, 0.6375, 0.5267],
  2671. [0.6193, 0.4034, 0.7757, 0.2347, 0.3733, 0.2919, 0.5930, 0.4926],
  2672. [0.6114, 0.4018, 0.7213, 0.1967, 0.3763, 0.2700, 0.5875, 0.5533]],
  2673. device='cuda:0', dtype=torch.float64)
  2674. predictions are: tensor([[ 0.0673, -0.7542, 1.7669, -0.3813, -0.2630, -0.6866, 0.5843, 0.1976],
  2675. [ 0.8705, -0.3015, 2.0107, 0.5344, -0.6006, 0.4336, 0.6015, 0.1237],
  2676. [-0.2974, -0.9372, 1.2530, -0.8868, -0.3178, -1.0227, 0.3237, 0.2703],
  2677. [ 0.3326, -0.5938, 1.5830, -0.2542, -0.5651, -0.4246, 0.3844, 0.1999],
  2678. [ 0.6440, -0.4098, 1.8351, 0.1432, -0.6327, -0.0068, 0.4699, 0.1745],
  2679. [ 0.0210, -0.7968, 1.5529, -0.6347, -0.3618, -0.9225, 0.4992, 0.1978],
  2680. [ 0.2055, -0.6765, 1.3400, -0.6754, -0.5636, -0.7407, 0.3770, 0.1858],
  2681. [ 0.1917, -0.6800, 1.3017, -0.6446, -0.5512, -0.7384, 0.3369, 0.1998]],
  2682. device='cuda:0', grad_fn=<AddmmBackward>)
  2683. landmarks are: tensor([[[ 0.6537, -0.3719, 1.6979, -0.8617, -0.0169, -1.3621, 0.6926,
  2684. 0.1501],
  2685. [ 0.6262, -0.4461, 1.8480, -0.0534, -0.5827, -0.1227, 0.6587,
  2686. 0.0774],
  2687. [-2.2859, -2.2859, 0.9175, -1.3947, -0.3691, -1.2467, 0.2314,
  2688. 0.3238],
  2689. [ 0.5483, -0.4105, 1.4208, -0.8002, -0.6000, -0.9002, 0.0511,
  2690. 0.3220],
  2691. [ 0.5581, -0.3912, 1.6460, -0.5230, -0.6173, -0.5923, 0.0681,
  2692. 0.4348],
  2693. [ 0.6092, -0.4143, 1.4901, -1.2467, -0.1208, -1.4006, 0.6587,
  2694. 0.1467],
  2695. [ 0.5746, -0.4225, 1.2968, -1.2019, -0.5615, -0.9374, 0.4533,
  2696. -0.0108],
  2697. [ 0.5380, -0.4302, 1.0455, -1.3775, -0.5480, -1.0388, 0.4277,
  2698. 0.2699]]], device='cuda:0')
  2699. loss_train_step before backward: tensor(0.1982, device='cuda:0', grad_fn=<MseLossBackward>)
  2700. loss_train_step after backward: tensor(0.1982, device='cuda:0', grad_fn=<MseLossBackward>)
  2701. loss_train: 7.7682068310678005
  2702. step: 53
  2703. running loss: 0.14656994020882644
  2704. Train Steps: 53/90 Loss: 0.1466 torch.Size([8, 600, 800])
  2705. torch.Size([8, 8])
  2706. tensor([[0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400],
  2707. [ nan, nan, 0.8488, 0.2300, 0.5563, 0.2100, 0.7390, 0.5679],
  2708. [0.6266, 0.4070, 0.8712, 0.5600, 0.3713, 0.4783, 0.5775, 0.6100],
  2709. [0.6102, 0.4005, 0.8688, 0.5100, 0.4813, 0.5400, 0.5404, 0.5064],
  2710. [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672],
  2711. [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617],
  2712. [0.6278, 0.4253, 0.8875, 0.5017, 0.4113, 0.2750, 0.5413, 0.6196],
  2713. [0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117]],
  2714. device='cuda:0', dtype=torch.float64)
  2715. predictions are: tensor([[-0.2143, -0.9404, 1.4668, -1.0647, -0.3230, -1.3438, 0.4597, 0.2239],
  2716. [-0.2777, -0.9847, 1.5412, -1.0364, -0.2318, -1.3185, 0.5605, 0.2012],
  2717. [ 0.6630, -0.3745, 1.6541, 0.1284, -0.6696, 0.0945, 0.4320, 0.1933],
  2718. [ 0.3220, -0.5820, 1.6807, -0.1678, -0.4297, -0.2776, 0.4604, 0.1703],
  2719. [ 0.7033, -0.3861, 1.6634, 0.0126, -0.7016, -0.1221, 0.5019, 0.1436],
  2720. [ 0.5057, -0.4786, 1.6570, -0.0877, -0.5653, -0.0970, 0.4212, 0.2048],
  2721. [ 0.2947, -0.5987, 1.6497, -0.2465, -0.4861, -0.4751, 0.4378, 0.1929],
  2722. [ 0.5106, -0.4723, 1.6720, -0.1766, -0.6225, -0.1774, 0.5021, 0.1598]],
  2723. device='cuda:0', grad_fn=<AddmmBackward>)
  2724. landmarks are: tensor([[[ 0.5958, -0.3818, 1.5536, -1.1081, -0.2074, -1.3390, 0.5663,
  2725. 0.2083],
  2726. [-2.2859, -2.2859, 1.6344, -1.2236, 0.2834, -1.3159, 1.1276,
  2727. 0.3371],
  2728. [ 0.6081, -0.4059, 1.7383, 0.3007, -0.5711, -0.0765, 0.3815,
  2729. 0.5316],
  2730. [ 0.5328, -0.4361, 1.7268, 0.0697, -0.0630, 0.2083, 0.2103,
  2731. 0.0532],
  2732. [ 0.6128, -0.4375, 1.7788, 0.2699, -0.6346, -0.2536, 0.5463,
  2733. -0.1278],
  2734. [ 0.5637, -0.4143, 1.7519, -0.0787, -0.3055, -0.0149, 0.3758,
  2735. 0.3084],
  2736. [ 0.6139, -0.3216, 1.8134, 0.0313, -0.3864, -1.0157, 0.2144,
  2737. 0.5762],
  2738. [ 0.6262, -0.4461, 1.8480, -0.0534, -0.5827, -0.1227, 0.6587,
  2739. 0.0774]]], device='cuda:0')
  2740. loss_train_step before backward: tensor(0.1455, device='cuda:0', grad_fn=<MseLossBackward>)
  2741. loss_train_step after backward: tensor(0.1455, device='cuda:0', grad_fn=<MseLossBackward>)
  2742. loss_train: 7.913748513907194
  2743. step: 54
  2744. running loss: 0.14655089840568877
  2745. Train Steps: 54/90 Loss: 0.1466 torch.Size([8, 600, 800])
  2746. torch.Size([8, 8])
  2747. tensor([[0.6261, 0.3987, 0.8688, 0.4917, 0.4300, 0.5333, 0.7010, 0.5309],
  2748. [ nan, nan, 0.6412, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700],
  2749. [0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5787, 0.5600],
  2750. [0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567],
  2751. [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550],
  2752. [0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767],
  2753. [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967],
  2754. [0.6126, 0.3954, 0.8538, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350]],
  2755. device='cuda:0', dtype=torch.float64)
  2756. predictions are: tensor([[ 0.5348, -0.4461, 1.6849, -0.0344, -0.5152, 0.0100, 0.5367, 0.1942],
  2757. [-0.2356, -0.9501, 1.0968, -1.2913, -0.4973, -1.4036, 0.2616, 0.2211],
  2758. [ 0.3467, -0.5774, 1.6509, -0.2637, -0.4446, -0.3173, 0.4580, 0.2147],
  2759. [ 0.2504, -0.6009, 1.6104, -0.3901, -0.4216, -0.3745, 0.5047, 0.1924],
  2760. [ 0.2240, -0.6485, 1.7115, -0.4295, -0.3777, -0.5357, 0.5990, 0.1783],
  2761. [ 0.5265, -0.4631, 1.7272, -0.2212, -0.6034, -0.4052, 0.5230, 0.1763],
  2762. [ 0.4698, -0.5562, 1.8330, -0.1877, -0.5408, -0.6042, 0.5758, 0.1269],
  2763. [ 0.5352, -0.5074, 1.7070, -0.1354, -0.5545, -0.2338, 0.4854, 0.1544]],
  2764. device='cuda:0', grad_fn=<AddmmBackward>)
  2765. landmarks are: tensor([[[ 0.6059, -0.4442, 1.7268, -0.0149, -0.2998, 0.1775, 0.9521,
  2766. 0.1661],
  2767. [-2.2859, -2.2859, 0.6760, -1.4083, -0.3286, -1.4160, 0.2487,
  2768. 0.3469],
  2769. [ 0.5505, -0.4207, 1.7095, -0.0534, -0.0509, 0.1050, 0.3873,
  2770. 0.3007],
  2771. [ 0.5538, -0.4139, 1.7557, -0.1843, -0.0459, 0.1242, 0.4219,
  2772. 0.2853],
  2773. [ 0.5351, -0.4321, 1.7326, -0.1381, 0.0463, 0.1049, 0.4162,
  2774. 0.2776],
  2775. [ 0.6058, -0.3216, 1.8423, -0.2536, -0.5885, -0.6000, 0.3353,
  2776. 0.3777],
  2777. [ 0.6054, -0.3767, 1.7788, 0.0774, -0.5711, -0.7694, 0.5887,
  2778. 0.0081],
  2779. [ 0.5435, -0.4597, 1.6575, 0.0159, -0.3229, -0.1150, 0.1877,
  2780. 0.1854]]], device='cuda:0')
  2781. loss_train_step before backward: tensor(0.1391, device='cuda:0', grad_fn=<MseLossBackward>)
  2782. loss_train_step after backward: tensor(0.1391, device='cuda:0', grad_fn=<MseLossBackward>)
  2783. loss_train: 8.052847858518362
  2784. step: 55
  2785. running loss: 0.14641541560942475
  2786. Train Steps: 55/90 Loss: 0.1464 torch.Size([8, 600, 800])
  2787. torch.Size([8, 8])
  2788. tensor([[0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309],
  2789. [0.6218, 0.4185, 0.7338, 0.2650, 0.4625, 0.1950, 0.5687, 0.5800],
  2790. [0.6087, 0.3951, 0.8387, 0.5833, 0.4188, 0.4933, 0.5146, 0.4830],
  2791. [0.6182, 0.3930, 0.8841, 0.3892, 0.3556, 0.4967, 0.6222, 0.5279],
  2792. [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633],
  2793. [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586],
  2794. [0.6260, 0.4120, 0.8013, 0.2350, 0.4888, 0.1533, 0.6281, 0.4895],
  2795. [0.6248, 0.4185, 0.8500, 0.5767, 0.4463, 0.4550, 0.5613, 0.5917]],
  2796. device='cuda:0', dtype=torch.float64)
  2797. predictions are: tensor([[ 0.5108, -0.4988, 1.8392, -0.1524, -0.4782, -0.4393, 0.6054, 0.1707],
  2798. [-0.1989, -0.8882, 1.1078, -1.1526, -0.4256, -1.1660, 0.2669, 0.2613],
  2799. [ 0.6571, -0.3982, 1.7409, 0.0728, -0.5547, -0.0445, 0.5122, 0.1667],
  2800. [ 0.5398, -0.4333, 1.5151, -0.3352, -0.6108, -0.1219, 0.4888, 0.1776],
  2801. [ 0.8204, -0.2803, 1.7220, 0.1719, -0.6168, 0.2006, 0.5326, 0.1883],
  2802. [-0.0578, -0.8517, 1.5819, -0.9280, -0.2828, -1.1182, 0.6220, 0.1895],
  2803. [-0.1826, -0.9124, 1.3633, -1.1262, -0.3392, -1.3109, 0.4546, 0.1922],
  2804. [ 0.6652, -0.3629, 1.7669, 0.1831, -0.4866, 0.0789, 0.5693, 0.1919]],
  2805. device='cuda:0', grad_fn=<AddmmBackward>)
  2806. landmarks are: tensor([[[ 0.6520, -0.3623, 1.8885, 0.0313, -0.5538, -0.5384, 0.6926,
  2807. 0.1661],
  2808. [ 0.5862, -0.3530, 1.1032, -1.0619, -0.1497, -1.3852, 0.3411,
  2809. 0.3931],
  2810. [ 0.5258, -0.4610, 1.5882, 0.4085, -0.3517, -0.0072, 0.0910,
  2811. -0.0550],
  2812. [ 0.5697, -0.4706, 1.7976, -0.4884, -0.6433, 0.0081, 0.5878,
  2813. 0.1525],
  2814. [ 0.5879, -0.3591, 1.8018, 0.1236, -0.5596, -0.1612, 0.3469,
  2815. 0.3161],
  2816. [ 0.6075, -0.4129, 1.6055, -1.0080, -0.2420, -1.0080, 0.9704,
  2817. 0.2944],
  2818. [ 0.6053, -0.3830, 1.4150, -1.2005, -0.0284, -1.5777, 0.6154,
  2819. -0.0250],
  2820. [ 0.5999, -0.3530, 1.6402, 0.3777, -0.2247, -0.1843, 0.3065,
  2821. 0.4470]]], device='cuda:0')
  2822. loss_train_step before backward: tensor(0.0662, device='cuda:0', grad_fn=<MseLossBackward>)
  2823. loss_train_step after backward: tensor(0.0662, device='cuda:0', grad_fn=<MseLossBackward>)
  2824. loss_train: 8.11908669397235
  2825. step: 56
  2826. running loss: 0.14498369096379196
  2827.  
  2828. Train Steps: 56/90 Loss: 0.1450 torch.Size([8, 600, 800])
  2829. torch.Size([8, 8])
  2830. tensor([[0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433],
  2831. [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609],
  2832. [0.6128, 0.4118, 0.8638, 0.5333, 0.4625, 0.5267, 0.5193, 0.5475],
  2833. [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578],
  2834. [0.6055, 0.4015, 0.7425, 0.2033, 0.4113, 0.1883, 0.5217, 0.4823],
  2835. [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517],
  2836. [0.6179, 0.3993, 0.8925, 0.4789, 0.3879, 0.4900, 0.6041, 0.5279],
  2837. [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123]],
  2838. device='cuda:0', dtype=torch.float64)
  2839. predictions are: tensor([[ 0.4652, -0.4927, 1.4310, -0.5926, -0.5860, -0.4583, 0.4855, 0.1917],
  2840. [ 0.0168, -0.8016, 1.7405, -0.7322, -0.1639, -0.9592, 0.7691, 0.1852],
  2841. [ 0.9244, -0.2048, 1.9678, 0.4568, -0.4284, 0.4085, 0.6313, 0.1853],
  2842. [ 0.0860, -0.7330, 1.3060, -0.8993, -0.4453, -0.9782, 0.3544, 0.1966],
  2843. [ 0.0321, -0.7831, 1.3198, -0.9540, -0.4452, -1.0702, 0.3742, 0.1955],
  2844. [-0.0107, -0.7694, 1.3218, -0.8850, -0.4071, -0.9168, 0.3962, 0.2506],
  2845. [ 1.1011, -0.1720, 2.0850, 0.4680, -0.5672, 0.4022, 0.6618, 0.1479],
  2846. [ 0.2294, -0.6558, 1.3344, -0.7895, -0.5082, -0.8425, 0.4000, 0.1940]],
  2847. device='cuda:0', grad_fn=<AddmmBackward>)
  2848. landmarks are: tensor([[[ 0.5665, -0.3963, 1.3457, -0.9926, -0.6058, -0.6308, 0.5721,
  2849. 0.2237],
  2850. [ 0.6372, -0.3575, 1.6633, -1.2082, 0.2199, -1.2467, 1.1313,
  2851. 0.3050],
  2852. [ 0.5447, -0.3838, 1.7037, 0.1775, -0.1497, 0.1467, 0.1128,
  2853. 0.2431],
  2854. [ 0.5232, -0.4189, 0.9358, -1.4006, -0.3921, -1.3698, 0.2555,
  2855. 0.2906],
  2856. [ 0.5109, -0.4316, 1.1436, -1.3467, -0.3864, -1.4160, 0.1239,
  2857. -0.0580],
  2858. [ 0.5713, -0.3671, 0.8665, -1.0696, -0.3691, -1.2236, 0.3527,
  2859. 0.2622],
  2860. [ 0.5680, -0.4417, 1.8365, -0.0740, -0.4941, -0.0227, 0.5045,
  2861. 0.1525],
  2862. [ 0.5342, -0.4280, 1.0051, -1.0619, -0.5711, -1.0388, 0.2226,
  2863. 0.0802]]], device='cuda:0')
  2864. loss_train_step before backward: tensor(0.0934, device='cuda:0', grad_fn=<MseLossBackward>)
  2865. loss_train_step after backward: tensor(0.0934, device='cuda:0', grad_fn=<MseLossBackward>)
  2866. loss_train: 8.212458346039057
  2867. step: 57
  2868. running loss: 0.14407821659717643
  2869. Train Steps: 57/90 Loss: 0.1441 torch.Size([8, 600, 800])
  2870. torch.Size([8, 8])
  2871. tensor([[0.6277, 0.4057, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823],
  2872. [0.6170, 0.4102, 0.7468, 0.3695, 0.3463, 0.3767, 0.5238, 0.5823],
  2873. [0.6200, 0.4049, 0.8638, 0.5617, 0.4125, 0.5100, 0.6013, 0.5317],
  2874. [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285],
  2875. [0.6218, 0.4098, 0.7238, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350],
  2876. [0.6187, 0.4104, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683],
  2877. [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383],
  2878. [ nan, nan, 0.8850, 0.2817, 0.5112, 0.2183, 0.7184, 0.5436]],
  2879. device='cuda:0', dtype=torch.float64)
  2880. predictions are: tensor([[ 0.1649, -0.7242, 1.5017, -0.9635, -0.3496, -1.1003, 0.5266, 0.1752],
  2881. [ 0.5651, -0.4185, 1.4566, -0.3872, -0.5721, -0.2852, 0.4112, 0.2345],
  2882. [ 0.9952, -0.1645, 1.9069, 0.4831, -0.4560, 0.3968, 0.5863, 0.1960],
  2883. [ 0.6640, -0.3472, 1.5984, -0.2779, -0.5041, -0.0886, 0.5670, 0.2048],
  2884. [ 0.1283, -0.7124, 1.3394, -0.9232, -0.3711, -1.0119, 0.4405, 0.2160],
  2885. [ 0.1406, -0.7053, 1.3067, -0.9091, -0.4227, -0.9925, 0.3371, 0.2296],
  2886. [ 0.6080, -0.4196, 1.5368, -0.4087, -0.5550, -0.3284, 0.4938, 0.1815],
  2887. [ 0.0596, -0.7789, 1.7898, -0.7913, -0.1113, -1.0448, 0.7324, 0.1776]],
  2888. device='cuda:0', grad_fn=<AddmmBackward>)
  2889. landmarks are: tensor([[[ 0.6132, -0.4122, 1.5478, -1.0619, -0.2709, -1.4314, 0.5500,
  2890. -0.0583],
  2891. [ 0.5639, -0.3911, 1.1634, -0.5794, -0.6866, -0.5461, 0.1334,
  2892. 0.4036],
  2893. [ 0.5776, -0.4159, 1.7037, 0.3084, -0.3806, 0.0697, 0.4912,
  2894. 0.1698],
  2895. [ 0.6108, -0.4273, 1.8711, -0.6616, -0.5769, -0.1997, 0.9156,
  2896. 0.1554],
  2897. [ 0.5861, -0.3929, 1.0570, -1.4314, -0.3286, -1.1235, 0.6182,
  2898. 0.1852],
  2899. [ 0.5718, -0.3905, 1.0053, -1.3305, -0.4614, -1.1235, 0.4450,
  2900. 0.3392],
  2901. [ 0.5779, -0.4275, 1.5894, -0.8362, -0.6577, -0.5153, 0.5605,
  2902. 0.2006],
  2903. [-2.2859, -2.2859, 1.8018, -0.9849, 0.0755, -1.2774, 1.0326,
  2904. 0.2249]]], device='cuda:0')
  2905. loss_train_step before backward: tensor(0.1737, device='cuda:0', grad_fn=<MseLossBackward>)
  2906. loss_train_step after backward: tensor(0.1737, device='cuda:0', grad_fn=<MseLossBackward>)
  2907. loss_train: 8.38615109398961
  2908. step: 58
  2909. running loss: 0.14458881196533813
  2910. Train Steps: 58/90 Loss: 0.1446 torch.Size([8, 600, 800])
  2911. torch.Size([8, 8])
  2912. tensor([[0.6260, 0.4214, 0.8538, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983],
  2913. [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650],
  2914. [0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162],
  2915. [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333],
  2916. [0.6336, 0.4086, 0.8900, 0.3950, 0.3900, 0.2950, 0.6504, 0.5066],
  2917. [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478],
  2918. [0.6198, 0.4164, 0.8700, 0.5067, 0.4625, 0.5650, 0.5464, 0.5197],
  2919. [0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742]],
  2920. device='cuda:0', dtype=torch.float64)
  2921. predictions are: tensor([[ 0.7843, -0.2825, 1.5981, -0.0211, -0.4891, -0.0875, 0.5092, 0.2230],
  2922. [ 0.2781, -0.6108, 1.5591, -0.6938, -0.3728, -0.8706, 0.4629, 0.2388],
  2923. [ 0.5331, -0.3995, 1.5068, -0.4493, -0.4661, -0.2366, 0.5199, 0.2029],
  2924. [ 0.6395, -0.3974, 1.7493, -0.2473, -0.3899, -0.2536, 0.5395, 0.1864],
  2925. [ 0.3610, -0.5704, 1.4294, -0.7834, -0.4042, -0.8035, 0.5313, 0.1834],
  2926. [ 0.3174, -0.6453, 1.6677, -0.7387, -0.2951, -0.9430, 0.5775, 0.1927],
  2927. [ 0.4922, -0.4571, 1.6685, -0.2111, -0.2608, -0.2328, 0.5452, 0.2422],
  2928. [ 0.2434, -0.6341, 1.4036, -0.9700, -0.3803, -1.0104, 0.5442, 0.1745]],
  2929. device='cuda:0', grad_fn=<AddmmBackward>)
  2930. landmarks are: tensor([[[ 0.6055, -0.3393, 1.6575, 0.2545, -0.5942, -0.5461, 0.2949,
  2931. 0.4778],
  2932. [ 0.5894, -0.3503, 1.6402, -0.3614, -0.5827, -0.7925, 0.3238,
  2933. 0.3238],
  2934. [ 0.6059, -0.4442, 1.8921, -0.3422, -0.6230, -0.1461, 0.7489,
  2935. 0.0985],
  2936. [ 0.6003, -0.3747, 1.8885, -0.1997, -0.5018, -0.0149, 0.5490,
  2937. 0.1775],
  2938. [ 0.6405, -0.3984, 1.8249, -0.4614, -0.4845, -0.9233, 0.7182,
  2939. 0.0539],
  2940. [ 0.5792, -0.4052, 1.8214, -0.6587, -0.5384, -0.8924, 0.4381,
  2941. 0.2442],
  2942. [ 0.5770, -0.3624, 1.7326, 0.0543, -0.1497, 0.3238, 0.2378,
  2943. 0.1146],
  2944. [ 0.6134, -0.4218, 1.7268, -0.6154, -0.4730, -1.0850, 0.5463,
  2945. -0.0957]]], device='cuda:0')
  2946. loss_train_step before backward: tensor(0.0448, device='cuda:0', grad_fn=<MseLossBackward>)
  2947. loss_train_step after backward: tensor(0.0448, device='cuda:0', grad_fn=<MseLossBackward>)
  2948. loss_train: 8.43092280998826
  2949. step: 59
  2950. running loss: 0.14289699677946205
  2951. Train Steps: 59/90 Loss: 0.1429 torch.Size([8, 600, 800])
  2952. torch.Size([8, 8])
  2953. tensor([[0.6120, 0.4014, 0.6863, 0.2817, 0.3700, 0.2783, 0.5513, 0.5667],
  2954. [0.6229, 0.4066, 0.7612, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350],
  2955. [0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272],
  2956. [0.6200, 0.4049, 0.8638, 0.5617, 0.4125, 0.5100, 0.6013, 0.5317],
  2957. [ nan, nan, 0.7515, 0.2708, 0.3987, 0.2267, 0.5162, 0.5567],
  2958. [0.6249, 0.4142, 0.8350, 0.3283, 0.3613, 0.3700, 0.6188, 0.5400],
  2959. [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817],
  2960. [0.6310, 0.4017, 0.8563, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006]],
  2961. device='cuda:0', dtype=torch.float64)
  2962. predictions are: tensor([[ 0.2818, -0.5519, 1.1398, -0.9494, -0.4569, -0.8034, 0.3754, 0.2358],
  2963. [ 0.2219, -0.6087, 1.2525, -1.0545, -0.3426, -0.9762, 0.4758, 0.1897],
  2964. [ 0.3792, -0.5581, 1.6016, -0.6791, -0.3140, -0.8254, 0.5364, 0.1975],
  2965. [ 0.9969, -0.1383, 1.8140, 0.2493, -0.3933, 0.2813, 0.6364, 0.1949],
  2966. [ 0.0079, -0.7529, 1.2589, -1.0732, -0.2950, -1.1134, 0.4069, 0.2175],
  2967. [ 0.4674, -0.4826, 1.6665, -0.6039, -0.3580, -0.7430, 0.5824, 0.1887],
  2968. [-0.0475, -0.7370, 1.3782, -0.9749, -0.1228, -1.0500, 0.4923, 0.2548],
  2969. [ 1.1495, -0.1309, 1.9559, 0.3334, -0.4791, 0.2528, 0.6781, 0.1574]],
  2970. device='cuda:0', grad_fn=<AddmmBackward>)
  2971. landmarks are: tensor([[[ 0.5410, -0.4321, 0.8838, -0.9849, -0.5769, -1.0003, 0.2603,
  2972. 0.3315],
  2973. [ 0.5911, -0.4080, 1.2303, -0.9156, -0.3229, -1.2851, 0.4508,
  2974. 0.1852],
  2975. [ 0.5711, -0.4015, 1.6979, -0.6770, -0.5365, -1.0619, 0.1712,
  2976. 0.1494],
  2977. [ 0.5776, -0.4159, 1.7037, 0.3084, -0.3806, 0.0697, 0.4912,
  2978. 0.1698],
  2979. [-2.2859, -2.2859, 1.1854, -1.0352, -0.4441, -1.2390, 0.0986,
  2980. 0.2853],
  2981. [ 0.6006, -0.3728, 1.5709, -0.7694, -0.6173, -0.5769, 0.5721,
  2982. 0.2083],
  2983. [ 0.5964, -0.3380, 1.4785, -0.8386, -0.2420, -1.0619, 0.3238,
  2984. 0.4008],
  2985. [ 0.6286, -0.4303, 1.6691, 0.3931, -0.5827, -0.0919, 0.6359,
  2986. 0.0261]]], device='cuda:0')
  2987. loss_train_step before backward: tensor(0.1616, device='cuda:0', grad_fn=<MseLossBackward>)
  2988. loss_train_step after backward: tensor(0.1616, device='cuda:0', grad_fn=<MseLossBackward>)
  2989. loss_train: 8.592534977942705
  2990. step: 60
  2991. running loss: 0.14320891629904509
  2992.  
  2993. Train Steps: 60/90 Loss: 0.1432 torch.Size([8, 600, 800])
  2994. torch.Size([8, 8])
  2995. tensor([[0.6275, 0.4048, 0.8488, 0.2883, 0.4463, 0.2033, 0.6321, 0.5155],
  2996. [ nan, nan, 0.8938, 0.2850, 0.4662, 0.3117, 0.7406, 0.5528],
  2997. [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343],
  2998. [0.6270, 0.4267, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183],
  2999. [0.6224, 0.4179, 0.8700, 0.5683, 0.4037, 0.4683, 0.5650, 0.5600],
  3000. [0.6048, 0.3928, 0.8538, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719],
  3001. [0.6263, 0.4065, 0.9038, 0.4317, 0.3588, 0.4550, 0.6325, 0.5250],
  3002. [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6038, 0.4833]],
  3003. device='cuda:0', dtype=torch.float64)
  3004. predictions are: tensor([[ 0.2086, -0.6300, 1.3176, -1.0979, -0.3326, -1.1253, 0.5210, 0.1824],
  3005. [ 0.1186, -0.6808, 1.6586, -0.8910, -0.1042, -1.1124, 0.6969, 0.1947],
  3006. [ 0.0176, -0.7576, 1.3594, -1.1991, -0.1722, -1.2693, 0.6071, 0.1854],
  3007. [-0.0947, -0.7664, 1.0023, -1.2901, -0.3211, -1.2261, 0.3050, 0.2503],
  3008. [ 0.9525, -0.1494, 1.7418, 0.1433, -0.4784, 0.0591, 0.4824, 0.2212],
  3009. [ 0.8451, -0.1980, 1.6479, 0.0220, -0.4704, 0.0241, 0.4932, 0.2086],
  3010. [ 0.8089, -0.3003, 1.8164, -0.1682, -0.4304, -0.2600, 0.5217, 0.2167],
  3011. [ 0.7858, -0.2762, 1.6638, -0.1935, -0.4581, -0.1493, 0.5384, 0.1817]],
  3012. device='cuda:0', grad_fn=<AddmmBackward>)
  3013. landmarks are: tensor([[[ 0.6126, -0.4161, 1.6344, -0.9541, -0.2247, -1.3467, 0.6339,
  3014. 0.0953],
  3015. [-2.2859, -2.2859, 1.8423, -0.9695, -0.1323, -0.8463, 1.1349,
  3016. 0.2676],
  3017. [ 0.6125, -0.4321, 1.8018, -0.9772, 0.0120, -1.2543, 0.9741,
  3018. 0.1821],
  3019. [ 0.6101, -0.3152, 1.0166, -0.7540, -0.0226, -1.4468, 0.2823,
  3020. 0.5702],
  3021. [ 0.5891, -0.3557, 1.7326, 0.3392, -0.4210, -0.1227, 0.3238,
  3022. 0.3007],
  3023. [ 0.5079, -0.4714, 1.6575, 0.2237, -0.4961, 0.0774, 0.1465,
  3024. -0.1061],
  3025. [ 0.6070, -0.4085, 1.8885, -0.2921, -0.6289, -0.1843, 0.6356,
  3026. 0.1390],
  3027. [ 0.5726, -0.4795, 1.7788, -0.0149, -0.5711, -0.0688, 0.5028,
  3028. -0.0534]]], device='cuda:0')
  3029. loss_train_step before backward: tensor(0.1915, device='cuda:0', grad_fn=<MseLossBackward>)
  3030. loss_train_step after backward: tensor(0.1915, device='cuda:0', grad_fn=<MseLossBackward>)
  3031. loss_train: 8.784059438854456
  3032. step: 61
  3033. running loss: 0.1440009744074501
  3034. Train Steps: 61/90 Loss: 0.1440 torch.Size([8, 600, 800])
  3035. torch.Size([8, 8])
  3036. tensor([[0.6213, 0.4131, 0.8438, 0.3550, 0.3513, 0.4400, 0.5716, 0.5123],
  3037. [0.6233, 0.4091, 0.8100, 0.2950, 0.3563, 0.3883, 0.6013, 0.5200],
  3038. [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869],
  3039. [0.6308, 0.3990, 0.8688, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133],
  3040. [0.6239, 0.4206, 0.8750, 0.5400, 0.3688, 0.4850, 0.5737, 0.5700],
  3041. [0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343],
  3042. [0.6193, 0.3930, 0.8949, 0.4437, 0.3852, 0.5435, 0.6263, 0.5263],
  3043. [0.6339, 0.4102, 0.9088, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390]],
  3044. device='cuda:0', dtype=torch.float64)
  3045. predictions are: tensor([[ 0.4254, -0.5102, 1.6345, -0.7446, -0.3703, -0.9065, 0.4836, 0.1739],
  3046. [ 0.3287, -0.5379, 1.2386, -0.9834, -0.4621, -0.9095, 0.4226, 0.2145],
  3047. [-0.0653, -0.8610, 1.5051, -1.2168, -0.0975, -1.5968, 0.5487, 0.2045],
  3048. [ 0.6743, -0.3409, 1.6672, -0.1588, -0.3592, -0.2565, 0.5132, 0.1833],
  3049. [ 0.8930, -0.2093, 1.7921, 0.0180, -0.4376, -0.1262, 0.5343, 0.2037],
  3050. [ 0.2680, -0.5955, 1.4384, -0.9249, -0.3006, -1.0350, 0.5602, 0.1513],
  3051. [ 0.6325, -0.3450, 1.5545, -0.3512, -0.4066, -0.2343, 0.5168, 0.2220],
  3052. [ 0.6448, -0.3246, 1.5906, -0.2330, -0.3459, -0.0971, 0.5405, 0.2321]],
  3053. device='cuda:0', grad_fn=<AddmmBackward>)
  3054. landmarks are: tensor([[[ 0.5841, -0.3778, 1.6113, -0.6462, -0.6635, -0.2536, 0.3542,
  3055. 0.0802],
  3056. [ 0.5932, -0.3962, 1.4554, -0.9233, -0.6404, -0.4922, 0.4912,
  3057. 0.1159],
  3058. [ 0.6119, -0.3927, 1.6979, -0.7925, -0.2536, -1.3698, 0.4282,
  3059. -0.0368],
  3060. [ 0.6275, -0.4430, 1.7268, 0.1082, -0.4614, 0.0159, 0.6644,
  3061. 0.0851],
  3062. [ 0.5960, -0.3430, 1.7557, 0.2083, -0.5827, -0.0457, 0.3642,
  3063. 0.3469],
  3064. [ 0.6125, -0.4369, 1.9173, -0.5384, -0.2594, -0.8386, 0.9741,
  3065. 0.1821],
  3066. [ 0.5746, -0.4706, 1.8476, -0.2365, -0.5068, 0.2245, 0.6069,
  3067. 0.1449],
  3068. [ 0.6421, -0.3912, 1.9115, -0.0842, -0.4730, 0.1544, 1.1824,
  3069. 0.2035]]], device='cuda:0')
  3070. loss_train_step before backward: tensor(0.0645, device='cuda:0', grad_fn=<MseLossBackward>)
  3071. loss_train_step after backward: tensor(0.0645, device='cuda:0', grad_fn=<MseLossBackward>)
  3072. loss_train: 8.848596323281527
  3073. step: 62
  3074. running loss: 0.14271929553679882
  3075. Train Steps: 62/90 Loss: 0.1427 torch.Size([8, 600, 800])
  3076. torch.Size([8, 8])
  3077. tensor([[0.6263, 0.4233, 0.7924, 0.4626, 0.3788, 0.2883, 0.5573, 0.6047],
  3078. [0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407],
  3079. [0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240],
  3080. [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050],
  3081. [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167],
  3082. [0.6219, 0.3934, 0.8688, 0.5267, 0.4313, 0.4967, 0.5988, 0.4983],
  3083. [0.6159, 0.4085, 0.6900, 0.2283, 0.4088, 0.1950, 0.5123, 0.5397],
  3084. [0.6128, 0.4022, 0.8738, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064]],
  3085. device='cuda:0', dtype=torch.float64)
  3086. predictions are: tensor([[ 6.0193e-01, -3.5355e-01, 1.4660e+00, -3.8460e-01, -4.2701e-01,
  3087. -3.7003e-01, 4.9220e-01, 2.1553e-01],
  3088. [ 6.1210e-01, -3.9221e-01, 1.6649e+00, -4.7105e-01, -4.3883e-01,
  3089. -6.3568e-01, 4.8713e-01, 1.8266e-01],
  3090. [-1.6378e-01, -8.3230e-01, 1.1472e+00, -1.3570e+00, -2.2156e-01,
  3091. -1.4473e+00, 4.2822e-01, 2.1547e-01],
  3092. [ 7.4028e-01, -3.1413e-01, 1.8390e+00, -2.3177e-01, -4.3150e-01,
  3093. -3.5873e-01, 6.1900e-01, 1.3988e-01],
  3094. [ 3.8143e-01, -5.2839e-01, 1.5582e+00, -8.0401e-01, -3.4977e-01,
  3095. -9.7758e-01, 5.2392e-01, 1.7169e-01],
  3096. [ 7.3339e-01, -3.0600e-01, 1.7870e+00, -5.2512e-02, -3.5248e-01,
  3097. -1.5096e-01, 5.9570e-01, 1.4974e-01],
  3098. [ 1.3114e-03, -7.5873e-01, 1.0977e+00, -1.2836e+00, -3.4830e-01,
  3099. -1.3368e+00, 3.7091e-01, 2.0337e-01],
  3100. [ 4.9602e-01, -4.2798e-01, 1.7384e+00, -2.2984e-01, -1.6082e-01,
  3101. -3.1048e-01, 6.1138e-01, 1.8955e-01]], device='cuda:0',
  3102. grad_fn=<AddmmBackward>)
  3103. landmarks are: tensor([[[ 0.6069, -0.3309, 1.3742, -0.1493, -0.5365, -0.9541, 0.2884,
  3104. 0.5071],
  3105. [ 0.5470, -0.3952, 1.6377, -0.4253, -0.6289, -0.8079, 0.0249,
  3106. 0.2116],
  3107. [ 0.5676, -0.4112, 1.1898, -1.2467, -0.2940, -1.4622, 0.2103,
  3108. 0.1343],
  3109. [ 0.5844, -0.4466, 1.8423, -0.1997, -0.5942, -0.3998, 0.4219,
  3110. 0.0467],
  3111. [ 0.5483, -0.4249, 1.5305, -0.7386, -0.6115, -0.8694, 0.3353,
  3112. 0.1005],
  3113. [ 0.5864, -0.4690, 1.7268, 0.1467, -0.2940, 0.0082, 0.4797,
  3114. 0.0159],
  3115. [ 0.5591, -0.3990, 0.9012, -1.2313, -0.3979, -1.3852, 0.0804,
  3116. 0.2071],
  3117. [ 0.5446, -0.4280, 1.7499, 0.0543, 0.0156, 0.1301, 0.1918,
  3118. 0.0532]]], device='cuda:0')
  3119. loss_train_step before backward: tensor(0.0488, device='cuda:0', grad_fn=<MseLossBackward>)
  3120. loss_train_step after backward: tensor(0.0488, device='cuda:0', grad_fn=<MseLossBackward>)
  3121. loss_train: 8.897438570857048
  3122. step: 63
  3123. running loss: 0.1412291836643976
  3124. Train Steps: 63/90 Loss: 0.1412 torch.Size([8, 600, 800])
  3125. torch.Size([8, 8])
  3126. tensor([[0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5837, 0.5500],
  3127. [ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600],
  3128. [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938],
  3129. [0.6276, 0.4120, 0.8738, 0.3133, 0.4225, 0.2217, 0.6203, 0.4892],
  3130. [0.6135, 0.3994, 0.7913, 0.3050, 0.3625, 0.3050, 0.5837, 0.5050],
  3131. [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6088, 0.5583],
  3132. [0.6204, 0.4110, 0.7913, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367],
  3133. [0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400]],
  3134. device='cuda:0', dtype=torch.float64)
  3135. predictions are: tensor([[ 0.6887, -0.2958, 1.7633, -0.0927, -0.3535, -0.0904, 0.5729, 0.1684],
  3136. [-0.0433, -0.7717, 1.3271, -1.0488, -0.1998, -1.1762, 0.4160, 0.2186],
  3137. [ 0.9128, -0.2367, 1.9098, 0.0604, -0.4814, -0.3571, 0.5593, 0.1122],
  3138. [ 0.2893, -0.6126, 1.7348, -0.7418, -0.1768, -1.0713, 0.6419, 0.1588],
  3139. [ 0.4247, -0.5085, 1.3427, -0.8894, -0.4398, -0.8603, 0.4474, 0.1686],
  3140. [ 0.3004, -0.5806, 1.2293, -0.9643, -0.3975, -1.0585, 0.3702, 0.1898],
  3141. [ 0.3562, -0.5280, 1.3323, -0.9654, -0.3721, -1.0282, 0.4855, 0.1664],
  3142. [ 0.7345, -0.2874, 1.6705, -0.2287, -0.4608, -0.1024, 0.5201, 0.1640]],
  3143. device='cuda:0', grad_fn=<AddmmBackward>)
  3144. landmarks are: tensor([[[ 0.5604, -0.4620, 1.7961, -0.1997, -0.3229, 0.1082, 0.4104,
  3145. 0.2545],
  3146. [-2.2859, -2.2859, 1.4266, -1.1389, -0.4499, -1.1235, 0.2891,
  3147. 0.3007],
  3148. [ 0.6273, -0.4105, 1.8480, 0.1082, -0.5538, -0.5076, 0.6414,
  3149. -0.0049],
  3150. [ 0.6128, -0.3828, 1.7499, -0.8386, -0.3344, -1.2620, 0.5792,
  3151. -0.0263],
  3152. [ 0.5477, -0.4413, 1.3688, -0.8771, -0.6115, -0.8771, 0.4104,
  3153. 0.0467],
  3154. [ 0.5992, -0.3968, 0.9307, -1.3497, -0.4730, -1.0465, 0.5259,
  3155. 0.2930],
  3156. [ 0.5796, -0.3878, 1.3688, -1.0542, -0.4095, -1.1312, 0.5894,
  3157. 0.1929],
  3158. [ 0.5778, -0.4393, 1.8018, -0.4614, -0.6693, -0.1381, 0.5490,
  3159. 0.2083]]], device='cuda:0')
  3160. loss_train_step before backward: tensor(0.1364, device='cuda:0', grad_fn=<MseLossBackward>)
  3161. loss_train_step after backward: tensor(0.1364, device='cuda:0', grad_fn=<MseLossBackward>)
  3162. loss_train: 9.033822163939476
  3163. step: 64
  3164. running loss: 0.1411534713115543
  3165.  
  3166. Train Steps: 64/90 Loss: 0.1412 torch.Size([8, 600, 800])
  3167. torch.Size([8, 8])
  3168. tensor([[0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882],
  3169. [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869],
  3170. [0.6168, 0.4029, 0.8523, 0.3417, 0.3588, 0.5000, 0.6125, 0.5400],
  3171. [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986],
  3172. [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200],
  3173. [ nan, nan, 0.7648, 0.2722, 0.3962, 0.2183, 0.5060, 0.5422],
  3174. [0.6250, 0.4106, 0.8700, 0.3717, 0.3588, 0.4967, 0.6038, 0.5167],
  3175. [0.6250, 0.4054, 0.8770, 0.4723, 0.4662, 0.5367, 0.6162, 0.5433]],
  3176. device='cuda:0', dtype=torch.float64)
  3177. predictions are: tensor([[ 0.6446, -0.3956, 1.6840, -0.4547, -0.5014, -0.7255, 0.4978, 0.1312],
  3178. [ 0.0775, -0.7255, 1.5045, -1.0049, -0.1928, -1.3421, 0.4969, 0.1804],
  3179. [ 0.5250, -0.3771, 1.4217, -0.5524, -0.4841, -0.3258, 0.4748, 0.1695],
  3180. [ 0.2667, -0.6293, 1.7410, -0.6791, -0.1763, -1.1061, 0.6038, 0.1466],
  3181. [ 0.8183, -0.2901, 1.7640, -0.0631, -0.5073, -0.2813, 0.4108, 0.1293],
  3182. [-0.0212, -0.7648, 1.1398, -1.1932, -0.3444, -1.3294, 0.3225, 0.1775],
  3183. [ 0.6146, -0.3061, 1.4930, -0.4160, -0.4981, -0.2606, 0.4851, 0.1693],
  3184. [ 0.5536, -0.3818, 1.6883, -0.1470, -0.2797, -0.1758, 0.5672, 0.1822]],
  3185. device='cuda:0', grad_fn=<AddmmBackward>)
  3186. landmarks are: tensor([[[ 0.5595, -0.3988, 1.7672, -0.4460, -0.5538, -0.5384, 0.0828,
  3187. -0.0310],
  3188. [ 0.6119, -0.3927, 1.6979, -0.7925, -0.2536, -1.3698, 0.4282,
  3189. -0.0368],
  3190. [ 0.5631, -0.4249, 1.6509, -0.7078, -0.6289, 0.0236, 0.5432,
  3191. 0.2083],
  3192. [ 0.6054, -0.3641, 1.8711, -0.5153, -0.2305, -1.3005, 0.5948,
  3193. 0.0171],
  3194. [ 0.5865, -0.3973, 1.8423, -0.0688, -0.5192, -0.2305, 0.4162,
  3195. 0.1159],
  3196. [-2.2859, -2.2859, 1.2469, -1.0288, -0.4557, -1.2774, 0.0511,
  3197. 0.2183],
  3198. [ 0.6010, -0.3896, 1.7326, -0.5692, -0.6289, 0.0082, 0.5028,
  3199. 0.1005],
  3200. [ 0.6009, -0.4135, 1.7651, -0.1043, -0.1323, 0.1929, 0.5605,
  3201. 0.2237]]], device='cuda:0')
  3202. loss_train_step before backward: tensor(0.1453, device='cuda:0', grad_fn=<MseLossBackward>)
  3203. loss_train_step after backward: tensor(0.1453, device='cuda:0', grad_fn=<MseLossBackward>)
  3204. loss_train: 9.179143756628036
  3205. step: 65
  3206. running loss: 0.14121759625581595
  3207. Train Steps: 65/90 Loss: 0.1412 torch.Size([8, 600, 800])
  3208. torch.Size([8, 8])
  3209. tensor([[0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204],
  3210. [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852],
  3211. [0.6364, 0.4154, 0.8938, 0.3717, 0.4500, 0.2583, 0.6448, 0.5285],
  3212. [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332],
  3213. [0.6274, 0.4003, 0.8638, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661],
  3214. [0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367],
  3215. [ nan, nan, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533],
  3216. [0.6208, 0.4082, 0.8538, 0.3067, 0.3588, 0.3717, 0.6112, 0.5517]],
  3217. device='cuda:0', dtype=torch.float64)
  3218. predictions are: tensor([[ 0.6267, -0.3419, 1.6333, -0.1439, -0.4652, -0.1813, 0.4128, 0.1475],
  3219. [ 0.4693, -0.4492, 1.6629, -0.2391, -0.3128, -0.3149, 0.4396, 0.1641],
  3220. [ 0.0783, -0.7178, 1.6822, -0.9112, -0.1434, -1.2153, 0.5995, 0.1530],
  3221. [ 0.6831, -0.3662, 1.6256, -0.3334, -0.5366, -0.4167, 0.4806, 0.1187],
  3222. [ 0.7415, -0.2970, 1.6891, -0.0164, -0.5241, -0.1477, 0.4616, 0.1052],
  3223. [ 0.2370, -0.6096, 1.6264, -0.7659, -0.3008, -1.0667, 0.4651, 0.1323],
  3224. [ 0.0431, -0.7028, 1.2359, -1.1322, -0.4204, -1.2392, 0.3013, 0.1724],
  3225. [ 0.4519, -0.4859, 1.5367, -0.7710, -0.4870, -0.8818, 0.4508, 0.1412]],
  3226. device='cuda:0', grad_fn=<AddmmBackward>)
  3227. landmarks are: tensor([[[ 0.6042, -0.4273, 1.7198, 0.2184, -0.3478, 0.1149, 0.8062,
  3228. 0.1176],
  3229. [ 0.5500, -0.4060, 1.7326, 0.0236, -0.1554, -0.0226, 0.1438,
  3230. 0.4171],
  3231. [ 0.6537, -0.3671, 1.8423, -0.5692, -0.2074, -1.0927, 0.6926,
  3232. 0.1554],
  3233. [ 0.6520, -0.4056, 1.9173, -0.0765, -0.5596, -0.4537, 0.7949,
  3234. 0.1768],
  3235. [ 0.6120, -0.4371, 1.7037, 0.4701, -0.5827, -0.0226, 0.5354,
  3236. -0.1331],
  3237. [ 0.5900, -0.3932, 1.8307, -0.3921, -0.4268, -1.1851, 0.3758,
  3238. 0.1929],
  3239. [-2.2859, -2.2859, 1.5074, -1.0388, -0.5423, -0.9849, 0.2199,
  3240. 0.2699],
  3241. [ 0.5814, -0.4003, 1.6575, -0.8694, -0.6289, -0.5692, 0.5374,
  3242. 0.2622]]], device='cuda:0')
  3243. loss_train_step before backward: tensor(0.1668, device='cuda:0', grad_fn=<MseLossBackward>)
  3244. loss_train_step after backward: tensor(0.1668, device='cuda:0', grad_fn=<MseLossBackward>)
  3245. loss_train: 9.345949053764343
  3246. step: 66
  3247. running loss: 0.14160528869339914
  3248. Train Steps: 66/90 Loss: 0.1416 torch.Size([8, 600, 800])
  3249. torch.Size([8, 8])
  3250. tensor([[0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6063, 0.5617],
  3251. [0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6188, 0.5433],
  3252. [0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507],
  3253. [0.6198, 0.3997, 0.8582, 0.5361, 0.4117, 0.5016, 0.5942, 0.5134],
  3254. [ nan, nan, 0.8850, 0.3000, 0.5363, 0.2250, 0.7343, 0.5771],
  3255. [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896],
  3256. [0.6223, 0.4028, 0.8988, 0.4200, 0.3763, 0.5733, 0.6375, 0.5167],
  3257. [0.6282, 0.4034, 0.7830, 0.2080, 0.4532, 0.2080, 0.6404, 0.5323]],
  3258. device='cuda:0', dtype=torch.float64)
  3259. predictions are: tensor([[ 0.7712, -0.2923, 1.8211, -0.0302, -0.6109, -0.2295, 0.3809, 0.1126],
  3260. [ 0.7003, -0.3199, 1.8258, -0.0810, -0.5108, -0.0843, 0.5129, 0.1062],
  3261. [ 0.1465, -0.6928, 1.1311, -1.1692, -0.4661, -1.2716, 0.2790, 0.1480],
  3262. [ 0.6411, -0.3872, 1.7449, -0.0299, -0.4848, -0.1530, 0.3797, 0.1211],
  3263. [-0.1579, -0.8550, 1.6510, -1.0446, -0.0177, -1.3773, 0.6750, 0.1100],
  3264. [ 0.3622, -0.5523, 1.4921, -0.6718, -0.5310, -0.9092, 0.2859, 0.1646],
  3265. [ 0.6469, -0.3301, 1.7399, -0.1981, -0.4915, -0.0087, 0.5131, 0.1263],
  3266. [ 0.0991, -0.7330, 1.4066, -1.1085, -0.3121, -1.4014, 0.4390, 0.1324]],
  3267. device='cuda:0', grad_fn=<AddmmBackward>)
  3268. landmarks are: tensor([[[ 0.5797, -0.3965, 1.8480, -0.0765, -0.6173, -0.1535, 0.5143,
  3269. 0.3084],
  3270. [ 0.5773, -0.4105, 1.8192, -0.1304, -0.5307, 0.0467, 0.5721,
  3271. 0.2237],
  3272. [ 0.5185, -0.4252, 0.9647, -1.2928, -0.4788, -1.2390, 0.2617,
  3273. 0.2576],
  3274. [ 0.5767, -0.4396, 1.6782, 0.1905, -0.3844, 0.0308, 0.4588,
  3275. 0.0855],
  3276. [-2.2859, -2.2859, 1.8018, -0.9002, 0.1910, -1.2467, 1.1057,
  3277. 0.3799],
  3278. [ 0.5485, -0.3997, 1.4445, -0.4895, -0.6000, -0.8309, 0.1878,
  3279. 0.4374],
  3280. [ 0.5886, -0.4253, 1.8654, -0.3460, -0.5480, 0.3623, 0.6587,
  3281. 0.1005],
  3282. [ 0.6158, -0.4225, 1.3307, -1.3253, -0.1924, -1.3252, 0.6721,
  3283. 0.1727]]], device='cuda:0')
  3284. loss_train_step before backward: tensor(0.1310, device='cuda:0', grad_fn=<MseLossBackward>)
  3285. loss_train_step after backward: tensor(0.1310, device='cuda:0', grad_fn=<MseLossBackward>)
  3286. loss_train: 9.476928368210793
  3287. step: 67
  3288. running loss: 0.14144669206284766
  3289. Train Steps: 67/90 Loss: 0.1414 torch.Size([8, 600, 800])
  3290. torch.Size([8, 8])
  3291. tensor([[0.6216, 0.4167, 0.8588, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667],
  3292. [0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133],
  3293. [0.6250, 0.4146, 0.8838, 0.3933, 0.3588, 0.4283, 0.6162, 0.5367],
  3294. [0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350],
  3295. [0.6156, 0.4125, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084],
  3296. [0.6278, 0.4253, 0.8875, 0.5017, 0.4113, 0.2750, 0.5413, 0.6196],
  3297. [0.6126, 0.4039, 0.8237, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138],
  3298. [0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124]],
  3299. device='cuda:0', dtype=torch.float64)
  3300. predictions are: tensor([[ 0.7303, -0.2955, 1.7954, 0.1157, -0.5662, 0.0109, 0.4479, 0.1301],
  3301. [ 0.6625, -0.3700, 1.7461, -0.0592, -0.5969, -0.2779, 0.3915, 0.0902],
  3302. [ 0.3183, -0.5735, 1.6481, -0.6196, -0.5000, -0.7327, 0.3463, 0.1323],
  3303. [-0.0706, -0.8395, 1.4050, -1.1966, -0.2504, -1.4328, 0.4267, 0.1387],
  3304. [ 0.3940, -0.4962, 1.7431, -0.1674, -0.3207, -0.1481, 0.4893, 0.1618],
  3305. [ 0.3869, -0.5238, 1.6733, -0.3139, -0.4044, -0.6556, 0.4237, 0.1367],
  3306. [ 0.5155, -0.4132, 1.5255, -0.5827, -0.6067, -0.5232, 0.4523, 0.1157],
  3307. [-0.1190, -0.8626, 1.4404, -1.1726, -0.2574, -1.4463, 0.4442, 0.1302]],
  3308. device='cuda:0', grad_fn=<AddmmBackward>)
  3309. landmarks are: tensor([[[ 0.5853, -0.3614, 1.6806, 0.2930, -0.4499, 0.1005, 0.3815,
  3310. 0.3315],
  3311. [ 0.6148, -0.4130, 1.6864, 0.1698, -0.5307, -0.1150, 0.6125,
  3312. 0.0851],
  3313. [ 0.6009, -0.3710, 1.7961, -0.4691, -0.6289, -0.3075, 0.5605,
  3314. 0.1929],
  3315. [ 0.5768, -0.4031, 1.3838, -1.1527, -0.2188, -1.4216, 0.4379,
  3316. 0.1850],
  3317. [ 0.5573, -0.3808, 1.8018, -0.0534, -0.1843, 0.4008, 0.3769,
  3318. 0.0622],
  3319. [ 0.6139, -0.3216, 1.8134, 0.0313, -0.3864, -1.0157, 0.2144,
  3320. 0.5762],
  3321. [ 0.5438, -0.4206, 1.5189, -0.4537, -0.6115, -0.6231, 0.4365,
  3322. 0.5491],
  3323. [ 0.5790, -0.4079, 1.5929, -1.0630, -0.4729, -1.0725, 0.4137,
  3324. 0.0807]]], device='cuda:0')
  3325. loss_train_step before backward: tensor(0.0537, device='cuda:0', grad_fn=<MseLossBackward>)
  3326.  
  3327. loss_train_step after backward: tensor(0.0537, device='cuda:0', grad_fn=<MseLossBackward>)
  3328. loss_train: 9.530608255416155
  3329. step: 68
  3330. running loss: 0.14015600375611992
  3331. Train Steps: 68/90 Loss: 0.1402 torch.Size([8, 600, 800])
  3332. torch.Size([8, 8])
  3333. tensor([[0.6198, 0.4130, 0.8762, 0.4117, 0.3650, 0.4900, 0.5707, 0.5103],
  3334. [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973],
  3335. [0.6261, 0.4029, 0.8720, 0.3364, 0.3665, 0.3753, 0.6531, 0.5183],
  3336. [0.6266, 0.4101, 0.8350, 0.2333, 0.3950, 0.2950, 0.6264, 0.4921],
  3337. [0.6307, 0.4029, 0.8988, 0.4817, 0.3937, 0.3500, 0.7311, 0.5378],
  3338. [0.6200, 0.4071, 0.7338, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517],
  3339. [0.6268, 0.4029, 0.8500, 0.2683, 0.3937, 0.3500, 0.6860, 0.5297],
  3340. [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483]],
  3341. device='cuda:0', dtype=torch.float64)
  3342. predictions are: tensor([[ 0.4384, -0.5000, 1.7996, -0.2838, -0.5043, -0.3255, 0.3941, 0.1576],
  3343. [ 0.7036, -0.3673, 1.8411, 0.1240, -0.5864, -0.0798, 0.4007, 0.1136],
  3344. [ 0.4413, -0.5175, 1.5479, -0.5979, -0.5562, -0.5699, 0.4458, 0.1313],
  3345. [-0.0993, -0.8470, 1.5847, -1.0225, -0.2267, -1.3204, 0.4917, 0.1482],
  3346. [ 0.3940, -0.5685, 1.9010, -0.2202, -0.3972, -0.7088, 0.4700, 0.1093],
  3347. [ 0.0720, -0.7794, 1.3627, -0.9925, -0.3594, -1.2348, 0.3514, 0.1815],
  3348. [ 0.4026, -0.5244, 1.4602, -0.7272, -0.5536, -0.7108, 0.4587, 0.1436],
  3349. [ 0.4534, -0.4552, 1.6432, -0.2746, -0.5154, -0.1603, 0.4155, 0.1531]],
  3350. device='cuda:0', grad_fn=<AddmmBackward>)
  3351. landmarks are: tensor([[[ 0.5771, -0.3784, 1.7614, -0.3844, -0.6000, -0.0226, 0.3501,
  3352. 0.0712],
  3353. [ 0.6240, -0.4321, 1.8423, 0.1852, -0.5885, -0.1612, 0.6962,
  3354. 0.0111],
  3355. [ 0.6059, -0.4249, 1.7420, -0.7321, -0.5931, -0.5523, 0.7307,
  3356. 0.1080],
  3357. [ 0.6081, -0.3918, 1.5709, -1.2082, -0.4614, -0.9233, 0.6072,
  3358. -0.0129],
  3359. [ 0.6273, -0.4249, 1.8654, -0.0611, -0.4672, -0.6693, 1.0910,
  3360. 0.1982],
  3361. [ 0.5779, -0.4054, 1.1032, -1.4006, -0.3460, -1.1543, 0.5547,
  3362. 0.2622],
  3363. [ 0.6092, -0.4249, 1.6402, -1.0465, -0.4672, -0.6693, 0.8827,
  3364. 0.1608],
  3365. [ 0.5546, -0.4620, 1.7788, -0.4229, -0.5192, -0.0226, 0.4277,
  3366. 0.2468]]], device='cuda:0')
  3367. loss_train_step before backward: tensor(0.0482, device='cuda:0', grad_fn=<MseLossBackward>)
  3368. loss_train_step after backward: tensor(0.0482, device='cuda:0', grad_fn=<MseLossBackward>)
  3369. loss_train: 9.57880049943924
  3370. step: 69
  3371. running loss: 0.13882319564404694
  3372. Train Steps: 69/90 Loss: 0.1388 torch.Size([8, 600, 800])
  3373. torch.Size([8, 8])
  3374. tensor([[0.6199, 0.4102, 0.8950, 0.4417, 0.4012, 0.5367, 0.6112, 0.5967],
  3375. [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170],
  3376. [ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600],
  3377. [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5637, 0.5633],
  3378. [0.6053, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168],
  3379. [0.6251, 0.4163, 0.8662, 0.4467, 0.3625, 0.3567, 0.6038, 0.5533],
  3380. [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938],
  3381. [0.6143, 0.4034, 0.8800, 0.4833, 0.4512, 0.5367, 0.5289, 0.5097]],
  3382. device='cuda:0', dtype=torch.float64)
  3383. predictions are: tensor([[ 0.5717, -0.4289, 1.9086, 0.0682, -0.4359, 0.2100, 0.5311, 0.1928],
  3384. [ 0.1212, -0.7306, 1.5996, -0.9908, -0.2890, -1.1521, 0.6839, 0.1284],
  3385. [-0.2105, -0.8776, 1.3527, -1.0402, -0.3279, -1.1361, 0.3717, 0.2074],
  3386. [ 0.6163, -0.3671, 1.7900, -0.0668, -0.6327, -0.2917, 0.4187, 0.1631],
  3387. [-0.0168, -0.8197, 1.1640, -1.2255, -0.4815, -1.3067, 0.3112, 0.1773],
  3388. [ 0.3989, -0.5297, 1.7450, -0.3858, -0.5510, -0.6769, 0.3761, 0.1441],
  3389. [ 0.6817, -0.3759, 1.8699, -0.0261, -0.6450, -0.3888, 0.4625, 0.0988],
  3390. [ 0.3942, -0.5177, 1.7776, -0.0976, -0.3729, -0.1446, 0.4075, 0.1746]],
  3391. device='cuda:0', grad_fn=<AddmmBackward>)
  3392. landmarks are: tensor([[[ 0.5772, -0.3913, 1.8480, -0.2459, -0.4326, 0.1929, 0.5374,
  3393. 0.4701],
  3394. [ 0.6471, -0.3840, 1.5767, -1.0311, -0.0457, -1.5007, 0.6889,
  3395. 0.1020],
  3396. [-2.2859, -2.2859, 1.4266, -1.1389, -0.4499, -1.1235, 0.2891,
  3397. 0.3007],
  3398. [ 0.5902, -0.3493, 1.7961, -0.0072, -0.5942, -0.5615, 0.3180,
  3399. 0.3161],
  3400. [ 0.5102, -0.4223, 0.8999, -1.4011, -0.4383, -1.3082, 0.2267,
  3401. 0.1013],
  3402. [ 0.6016, -0.3633, 1.7152, -0.2228, -0.6115, -0.6385, 0.5028,
  3403. 0.2699],
  3404. [ 0.6273, -0.4105, 1.8480, 0.1082, -0.5538, -0.5076, 0.6414,
  3405. -0.0049],
  3406. [ 0.5515, -0.4225, 1.7788, -0.0534, -0.2016, 0.1929, 0.1568,
  3407. 0.0682]]], device='cuda:0')
  3408. loss_train_step before backward: tensor(0.1287, device='cuda:0', grad_fn=<MseLossBackward>)
  3409. loss_train_step after backward: tensor(0.1287, device='cuda:0', grad_fn=<MseLossBackward>)
  3410. loss_train: 9.707480400800705
  3411. step: 70
  3412. running loss: 0.13867829144001007
  3413. Train Steps: 70/90 Loss: 0.1387 torch.Size([8, 600, 800])
  3414. torch.Size([8, 8])
  3415. tensor([[0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717],
  3416. [0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567],
  3417. [0.6231, 0.3973, 0.8650, 0.3950, 0.3625, 0.3183, 0.5837, 0.5167],
  3418. [ nan, nan, 0.7625, 0.2433, 0.3713, 0.2867, 0.5235, 0.5220],
  3419. [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351],
  3420. [0.6245, 0.4115, 0.8700, 0.4883, 0.4625, 0.5517, 0.6100, 0.5217],
  3421. [0.6204, 0.4110, 0.7913, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367],
  3422. [0.6243, 0.4128, 0.7762, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417]],
  3423. device='cuda:0', dtype=torch.float64)
  3424. predictions are: tensor([[ 0.1662, -0.6768, 1.4210, -0.7968, -0.5248, -0.9291, 0.3372, 0.1923],
  3425. [ 0.4368, -0.4664, 1.9125, 0.0626, -0.3291, 0.1259, 0.5564, 0.1895],
  3426. [ 0.7224, -0.3700, 1.8125, -0.1950, -0.6755, -0.3210, 0.5398, 0.1174],
  3427. [-0.0464, -0.7945, 1.4098, -0.9435, -0.4752, -1.0650, 0.3346, 0.1966],
  3428. [ 0.1214, -0.7137, 1.4418, -0.9443, -0.4571, -1.1249, 0.4019, 0.1869],
  3429. [ 0.6116, -0.4209, 1.9821, 0.2965, -0.4020, 0.2547, 0.5225, 0.2192],
  3430. [ 0.4644, -0.5158, 1.5609, -0.7510, -0.5658, -0.8439, 0.5432, 0.1503],
  3431. [ 0.0048, -0.7716, 1.5697, -0.8414, -0.3493, -1.0282, 0.4463, 0.1949]],
  3432. device='cuda:0', grad_fn=<AddmmBackward>)
  3433. landmarks are: tensor([[[ 0.5746, -0.3882, 1.1436, -1.2005, -0.4903, -1.0157, 0.4393,
  3434. 0.3546],
  3435. [ 0.5538, -0.4139, 1.7557, -0.1843, -0.0459, 0.1242, 0.4219,
  3436. 0.2853],
  3437. [ 0.5924, -0.4507, 1.7095, -0.4614, -0.6115, -0.8156, 0.4104,
  3438. 0.1005],
  3439. [-2.2859, -2.2859, 1.2360, -1.1620, -0.5711, -0.9618, 0.1322,
  3440. 0.1253],
  3441. [ 0.5664, -0.4321, 1.2862, -1.0003, -0.2189, -1.4608, 0.3883,
  3442. 0.1855],
  3443. [ 0.5987, -0.3852, 1.7326, -0.0303, -0.1497, 0.2622, 0.5316,
  3444. 0.1236],
  3445. [ 0.5796, -0.3878, 1.3688, -1.0542, -0.4095, -1.1312, 0.5894,
  3446. 0.1929],
  3447. [ 0.5977, -0.3792, 1.2995, -1.0311, -0.5192, -0.8386, 0.5836,
  3448. 0.2160]]], device='cuda:0')
  3449. loss_train_step before backward: tensor(0.1591, device='cuda:0', grad_fn=<MseLossBackward>)
  3450. loss_train_step after backward: tensor(0.1591, device='cuda:0', grad_fn=<MseLossBackward>)
  3451. loss_train: 9.866620302200317
  3452. step: 71
  3453. running loss: 0.13896648312958193
  3454. Train Steps: 71/90 Loss: 0.1390 torch.Size([8, 600, 800])
  3455. torch.Size([8, 8])
  3456. tensor([[0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285],
  3457. [0.6272, 0.4045, 0.8538, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649],
  3458. [0.6254, 0.3993, 0.8988, 0.4767, 0.3987, 0.5517, 0.6955, 0.5285],
  3459. [ nan, nan, 0.7268, 0.2333, 0.4125, 0.1933, 0.5112, 0.5383],
  3460. [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773],
  3461. [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852],
  3462. [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123],
  3463. [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083]],
  3464. device='cuda:0', dtype=torch.float64)
  3465. predictions are: tensor([[ 0.4457, -0.4978, 1.7293, -0.5260, -0.5322, -0.3158, 0.6324, 0.1981],
  3466. [ 0.6052, -0.4613, 1.9613, 0.0222, -0.5985, -0.3985, 0.4678, 0.1336],
  3467. [ 0.5346, -0.4371, 1.9981, 0.1041, -0.4223, 0.2140, 0.6126, 0.2160],
  3468. [-0.1207, -0.8934, 1.2949, -1.2738, -0.4610, -1.4685, 0.3517, 0.1985],
  3469. [ 0.1044, -0.7318, 1.3309, -1.0767, -0.4852, -1.1226, 0.3573, 0.2500],
  3470. [ 0.3721, -0.5522, 1.9450, 0.0344, -0.3227, -0.1119, 0.4806, 0.2193],
  3471. [ 0.1638, -0.7299, 1.3604, -1.0877, -0.5575, -1.2349, 0.3445, 0.1811],
  3472. [ 0.5890, -0.4262, 1.8733, 0.0180, -0.5640, -0.0120, 0.4816, 0.2140]],
  3473. device='cuda:0', grad_fn=<AddmmBackward>)
  3474. landmarks are: tensor([[[ 0.6108, -0.4273, 1.8711, -0.6616, -0.5769, -0.1997, 0.9156,
  3475. 0.1554],
  3476. [ 0.6109, -0.4177, 1.6575, 0.4393, -0.5538, -0.2459, 0.4805,
  3477. -0.1385],
  3478. [ 0.6026, -0.4417, 1.8654, -0.0842, -0.4441, 0.2622, 0.9265,
  3479. 0.1554],
  3480. [-2.2859, -2.2859, 1.0712, -1.2085, -0.3806, -1.3929, 0.0755,
  3481. 0.2006],
  3482. [ 0.5841, -0.3574, 1.0859, -0.9541, -0.2824, -1.2851, 0.3460,
  3483. 0.3808],
  3484. [ 0.5500, -0.4060, 1.7326, 0.0236, -0.1554, -0.0226, 0.1438,
  3485. 0.4171],
  3486. [ 0.5342, -0.4280, 1.0051, -1.0619, -0.5711, -1.0388, 0.2226,
  3487. 0.0802],
  3488. [ 0.5713, -0.4105, 1.7557, 0.0467, -0.6520, -0.2690, 0.3988,
  3489. 0.5239]]], device='cuda:0')
  3490. loss_train_step before backward: tensor(0.1390, device='cuda:0', grad_fn=<MseLossBackward>)
  3491.  
  3492. loss_train_step after backward: tensor(0.1390, device='cuda:0', grad_fn=<MseLossBackward>)
  3493. loss_train: 10.005621805787086
  3494. step: 72
  3495. running loss: 0.13896696952482065
  3496. Train Steps: 72/90 Loss: 0.1390 torch.Size([8, 600, 800])
  3497. torch.Size([8, 8])
  3498. tensor([[0.6202, 0.4064, 0.7879, 0.2179, 0.4567, 0.1725, 0.5955, 0.5478],
  3499. [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426],
  3500. [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356],
  3501. [0.6229, 0.4066, 0.7612, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350],
  3502. [0.6134, 0.4090, 0.6926, 0.2819, 0.3538, 0.3233, 0.5563, 0.5667],
  3503. [0.6256, 0.4199, 0.8638, 0.5800, 0.3987, 0.4383, 0.5600, 0.5950],
  3504. [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220],
  3505. [0.6081, 0.3950, 0.8538, 0.4667, 0.3850, 0.4917, 0.5342, 0.4954]],
  3506. device='cuda:0', dtype=torch.float64)
  3507. predictions are: tensor([[ 0.0624, -0.7368, 1.4473, -0.9337, -0.3668, -1.0164, 0.4630, 0.2204],
  3508. [ 0.5218, -0.4831, 1.9114, -0.0674, -0.5392, -0.3152, 0.4470, 0.1608],
  3509. [-0.0386, -0.7876, 1.6309, -0.7558, -0.2925, -0.8943, 0.5597, 0.2360],
  3510. [ 0.2419, -0.6400, 1.4512, -0.9229, -0.4733, -0.8848, 0.5008, 0.2013],
  3511. [ 0.2882, -0.6012, 1.3015, -0.7879, -0.6360, -0.6736, 0.3322, 0.2447],
  3512. [ 0.7135, -0.3395, 1.9640, 0.4515, -0.5205, 0.2684, 0.5442, 0.2294],
  3513. [-0.0281, -0.8188, 1.2931, -1.1416, -0.4590, -1.2413, 0.3957, 0.2089],
  3514. [ 0.6038, -0.4154, 1.8860, 0.0039, -0.5550, 0.0999, 0.5708, 0.1776]],
  3515. device='cuda:0', grad_fn=<AddmmBackward>)
  3516. landmarks are: tensor([[[ 0.5787, -0.4086, 1.3535, -1.2794, -0.1764, -1.4891, 0.4645,
  3517. 0.2442],
  3518. [ 0.5788, -0.4135, 1.8214, -0.2408, -0.6039, -0.7854, 0.4115,
  3519. 0.2203],
  3520. [ 0.6174, -0.4225, 1.4975, -1.1709, -0.3174, -1.1806, 0.6539,
  3521. 0.1879],
  3522. [ 0.5911, -0.4080, 1.2303, -0.9156, -0.3229, -1.2851, 0.4508,
  3523. 0.1852],
  3524. [ 0.5473, -0.3966, 0.9131, -0.9838, -0.6520, -0.7925, 0.2834,
  3525. 0.3315],
  3526. [ 0.6038, -0.3464, 1.7037, 0.3931, -0.4441, -0.2613, 0.3007,
  3527. 0.4624],
  3528. [ 0.5566, -0.4160, 0.9935, -1.3313, -0.2824, -1.5161, 0.2144,
  3529. 0.1253],
  3530. [ 0.5227, -0.4615, 1.6575, -0.1304, -0.5076, -0.0149, 0.1815,
  3531. 0.0021]]], device='cuda:0')
  3532. loss_train_step before backward: tensor(0.0649, device='cuda:0', grad_fn=<MseLossBackward>)
  3533. loss_train_step after backward: tensor(0.0649, device='cuda:0', grad_fn=<MseLossBackward>)
  3534. loss_train: 10.070557050406933
  3535. step: 73
  3536. running loss: 0.1379528363069443
  3537. Train Steps: 73/90 Loss: 0.1380 torch.Size([8, 600, 800])
  3538. torch.Size([8, 8])
  3539. tensor([[0.6108, 0.4011, 0.8037, 0.3400, 0.3700, 0.2933, 0.5658, 0.5617],
  3540. [0.6182, 0.3972, 0.8552, 0.5914, 0.3683, 0.4181, 0.5688, 0.5378],
  3541. [0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567],
  3542. [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200],
  3543. [0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320],
  3544. [ nan, nan, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621],
  3545. [0.6193, 0.4108, 0.7438, 0.2700, 0.3650, 0.3683, 0.6238, 0.5717],
  3546. [0.6090, 0.4010, 0.7838, 0.3483, 0.3538, 0.3783, 0.5462, 0.5077]],
  3547. device='cuda:0', dtype=torch.float64)
  3548. predictions are: tensor([[ 0.3564, -0.5746, 1.4583, -0.8204, -0.5787, -0.8137, 0.4769, 0.2382],
  3549. [ 0.7135, -0.3818, 1.8126, 0.1473, -0.6671, -0.2749, 0.4425, 0.1993],
  3550. [ 0.2154, -0.6045, 1.7846, -0.2033, -0.2362, -0.1025, 0.5369, 0.2716],
  3551. [ 0.5455, -0.4859, 1.8217, -0.0442, -0.5644, -0.1549, 0.3641, 0.2224],
  3552. [ 0.5563, -0.4529, 1.7759, -0.2288, -0.5570, -0.2708, 0.5477, 0.1839],
  3553. [-0.3794, -1.0313, 1.6012, -1.1200, -0.0692, -1.3016, 0.7081, 0.2265],
  3554. [ 0.2850, -0.6013, 1.3278, -0.8779, -0.5838, -0.7285, 0.4025, 0.2657],
  3555. [ 0.3451, -0.5976, 1.4313, -0.8132, -0.6137, -0.7096, 0.3903, 0.2457]],
  3556. device='cuda:0', grad_fn=<AddmmBackward>)
  3557. landmarks are: tensor([[[ 0.5355, -0.4335, 1.4266, -0.7155, -0.5769, -0.9310, 0.3275,
  3558. 0.3087],
  3559. [ 0.5697, -0.4514, 1.6642, 0.4459, -0.5850, -0.3546, 0.3414,
  3560. 0.1982],
  3561. [ 0.5538, -0.4139, 1.7557, -0.1843, -0.0459, 0.1242, 0.4219,
  3562. 0.2853],
  3563. [ 0.5865, -0.3973, 1.8423, -0.0688, -0.5192, -0.2305, 0.4162,
  3564. 0.1159],
  3565. [ 0.6174, -0.4490, 1.8885, -0.0996, -0.4845, -0.3691, 0.9814,
  3566. 0.1715],
  3567. [-2.2859, -2.2859, 1.7210, -0.9772, 0.1852, -1.3698, 0.7986,
  3568. 0.3104],
  3569. [ 0.5747, -0.3886, 1.1494, -1.0388, -0.6000, -0.5846, 0.5952,
  3570. 0.3546],
  3571. [ 0.5270, -0.4336, 1.3342, -0.6770, -0.6520, -0.5384, 0.2370,
  3572. 0.0592]]], device='cuda:0')
  3573. loss_train_step before backward: tensor(0.1023, device='cuda:0', grad_fn=<MseLossBackward>)
  3574. loss_train_step after backward: tensor(0.1023, device='cuda:0', grad_fn=<MseLossBackward>)
  3575. loss_train: 10.172822825610638
  3576. step: 74
  3577. running loss: 0.13747057872446808
  3578. Train Steps: 74/90 Loss: 0.1375 torch.Size([8, 600, 800])
  3579. torch.Size([8, 8])
  3580. tensor([[0.6170, 0.4102, 0.7468, 0.3695, 0.3463, 0.3767, 0.5238, 0.5823],
  3581. [0.6164, 0.4102, 0.8850, 0.4867, 0.4213, 0.5983, 0.5609, 0.5038],
  3582. [0.6193, 0.4034, 0.7757, 0.2347, 0.3733, 0.2919, 0.5930, 0.4926],
  3583. [0.6252, 0.4158, 0.8988, 0.4083, 0.3788, 0.4783, 0.6225, 0.5633],
  3584. [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927],
  3585. [ nan, nan, 0.7225, 0.2167, 0.3987, 0.2283, 0.5427, 0.5181],
  3586. [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817],
  3587. [0.6293, 0.4024, 0.8750, 0.5000, 0.4012, 0.5733, 0.7121, 0.5633]],
  3588. device='cuda:0', dtype=torch.float64)
  3589. predictions are: tensor([[ 0.4214, -0.5060, 1.3804, -0.6720, -0.6546, -0.6298, 0.3875, 0.2875],
  3590. [ 0.4671, -0.4624, 1.8166, 0.0909, -0.3127, 0.1770, 0.5518, 0.2869],
  3591. [ 0.2761, -0.6391, 1.3203, -1.0693, -0.5863, -1.1693, 0.4733, 0.1975],
  3592. [ 0.5017, -0.4808, 1.8436, -0.1449, -0.5294, -0.1855, 0.4829, 0.2564],
  3593. [ 0.7856, -0.3420, 1.7747, -0.2400, -0.6585, -0.3377, 0.6026, 0.1817],
  3594. [-0.2893, -0.9720, 1.1546, -1.2933, -0.3962, -1.4782, 0.3583, 0.2215],
  3595. [-0.2827, -0.9134, 1.4449, -0.9290, -0.2064, -1.0815, 0.4354, 0.2955],
  3596. [ 0.6116, -0.4134, 1.9124, 0.2665, -0.3739, 0.3431, 0.5887, 0.2829]],
  3597. device='cuda:0', grad_fn=<AddmmBackward>)
  3598. landmarks are: tensor([[[ 0.5639, -0.3911, 1.1634, -0.5794, -0.6866, -0.5461, 0.1334,
  3599. 0.4036],
  3600. [ 0.5614, -0.3912, 1.8018, -0.0380, -0.3402, 0.4778, 0.3049,
  3601. 0.0412],
  3602. [ 0.5746, -0.4225, 1.2968, -1.2019, -0.5615, -0.9374, 0.4533,
  3603. -0.0108],
  3604. [ 0.6017, -0.3654, 1.8654, -0.3998, -0.5365, -0.0765, 0.5894,
  3605. 0.3161],
  3606. [ 0.6010, -0.4345, 1.8480, -0.3537, -0.6462, -0.2613, 0.6524,
  3607. -0.0099],
  3608. [-2.2859, -2.2859, 1.0513, -1.2851, -0.4441, -1.2313, 0.2206,
  3609. 0.1073],
  3610. [ 0.5964, -0.3380, 1.4785, -0.8386, -0.2420, -1.0619, 0.3238,
  3611. 0.4008],
  3612. [ 0.6207, -0.4273, 1.7557, 0.0236, -0.4326, 0.3623, 1.0033,
  3613. 0.3157]]], device='cuda:0')
  3614. loss_train_step before backward: tensor(0.1267, device='cuda:0', grad_fn=<MseLossBackward>)
  3615. loss_train_step after backward: tensor(0.1267, device='cuda:0', grad_fn=<MseLossBackward>)
  3616. loss_train: 10.29955194145441
  3617. step: 75
  3618. running loss: 0.13732735921939215
  3619. Train Steps: 75/90 Loss: 0.1373 torch.Size([8, 600, 800])
  3620. torch.Size([8, 8])
  3621. tensor([[0.6132, 0.4118, 0.8200, 0.3633, 0.3563, 0.5400, 0.5787, 0.5136],
  3622. [0.6251, 0.4163, 0.8662, 0.4467, 0.3625, 0.3567, 0.6038, 0.5533],
  3623. [0.6229, 0.4198, 0.7662, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783],
  3624. [0.6200, 0.4070, 0.8938, 0.4183, 0.3538, 0.4567, 0.6175, 0.5400],
  3625. [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760],
  3626. [ nan, nan, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563],
  3627. [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200],
  3628. [0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320]],
  3629. device='cuda:0', dtype=torch.float64)
  3630. predictions are: tensor([[ 0.3176, -0.5450, 1.4536, -0.6316, -0.4627, -0.1740, 0.5203, 0.3189],
  3631. [ 0.2611, -0.6328, 1.6570, -0.4461, -0.4967, -0.8020, 0.4033, 0.2421],
  3632. [-0.3042, -0.9527, 1.2764, -1.1099, -0.2248, -1.2061, 0.4493, 0.3028],
  3633. [ 0.6198, -0.4280, 1.7268, -0.3493, -0.5287, -0.2621, 0.6554, 0.2258],
  3634. [ 0.6195, -0.3894, 1.6452, 0.0312, -0.5016, -0.0553, 0.5120, 0.2784],
  3635. [-0.1235, -0.8596, 1.2788, -0.9454, -0.4956, -1.0478, 0.2828, 0.2828],
  3636. [ 0.5220, -0.4900, 1.7215, -0.1318, -0.5259, -0.2601, 0.3993, 0.2380],
  3637. [ 0.5877, -0.4383, 1.7150, -0.3114, -0.5164, -0.3794, 0.6195, 0.1953]],
  3638. device='cuda:0', grad_fn=<AddmmBackward>)
  3639. landmarks are: tensor([[[ 0.5466, -0.3840, 1.5016, -0.6077, -0.6404, 0.2083, 0.3871,
  3640. 0.0862],
  3641. [ 0.6016, -0.3633, 1.7152, -0.2228, -0.6115, -0.6385, 0.5028,
  3642. 0.2699],
  3643. [ 0.5912, -0.3467, 1.2533, -1.0388, -0.1150, -1.3005, 0.3353,
  3644. 0.3854],
  3645. [ 0.5778, -0.4061, 1.8423, -0.3537, -0.6520, -0.1766, 0.5663,
  3646. 0.2083],
  3647. [ 0.6421, -0.4008, 1.6055, 0.2160, -0.5076, -0.0534, 1.1020,
  3648. 0.3745],
  3649. [-2.2859, -2.2859, 1.4006, -0.8105, -0.6115, -0.8232, 0.0419,
  3650. 0.2837],
  3651. [ 0.5865, -0.3973, 1.8423, -0.0688, -0.5192, -0.2305, 0.4162,
  3652. 0.1159],
  3653. [ 0.6174, -0.4490, 1.8885, -0.0996, -0.4845, -0.3691, 0.9814,
  3654. 0.1715]]], device='cuda:0')
  3655. loss_train_step before backward: tensor(0.1469, device='cuda:0', grad_fn=<MseLossBackward>)
  3656. loss_train_step after backward: tensor(0.1469, device='cuda:0', grad_fn=<MseLossBackward>)
  3657. loss_train: 10.446426160633564
  3658. step: 76
  3659. running loss: 0.13745297579781005
  3660.  
  3661. Train Steps: 76/90 Loss: 0.1375 torch.Size([8, 600, 800])
  3662. torch.Size([8, 8])
  3663. tensor([[0.6099, 0.4030, 0.8638, 0.5117, 0.4983, 0.4965, 0.5086, 0.5388],
  3664. [0.6205, 0.4016, 0.8350, 0.2717, 0.3987, 0.2550, 0.5787, 0.5133],
  3665. [0.6185, 0.4129, 0.8900, 0.4567, 0.3937, 0.5417, 0.5734, 0.5110],
  3666. [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760],
  3667. [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5463, 0.5800],
  3668. [0.6122, 0.4006, 0.8850, 0.4217, 0.4088, 0.5517, 0.6063, 0.5517],
  3669. [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183],
  3670. [0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204]],
  3671. device='cuda:0', dtype=torch.float64)
  3672. predictions are: tensor([[ 0.1982, -0.6836, 1.6525, -0.2713, -0.2641, -0.4663, 0.4898, 0.2545],
  3673. [-0.2407, -0.9340, 1.4422, -1.1110, -0.3239, -1.3324, 0.5128, 0.2232],
  3674. [ 0.3009, -0.5716, 1.6889, -0.2674, -0.4470, -0.1971, 0.4458, 0.2760],
  3675. [ 0.6223, -0.3788, 1.5415, -0.1216, -0.5765, -0.2211, 0.5047, 0.2707],
  3676. [-0.1364, -0.8355, 1.0600, -1.1444, -0.5423, -1.2415, 0.2674, 0.2796],
  3677. [ 0.4898, -0.4481, 1.5487, -0.4247, -0.4844, -0.1815, 0.5657, 0.2896],
  3678. [ 0.5429, -0.4586, 1.7326, -0.2233, -0.4238, 0.0051, 0.6718, 0.2817],
  3679. [ 0.5129, -0.4504, 1.5689, -0.1454, -0.5120, -0.1783, 0.4519, 0.2615]],
  3680. device='cuda:0', grad_fn=<AddmmBackward>)
  3681. landmarks are: tensor([[[ 0.5309, -0.4246, 1.7037, 0.0774, 0.0158, 0.0075, 0.0635,
  3682. 0.2026],
  3683. [ 0.5800, -0.4312, 1.5709, -1.0311, -0.4441, -1.1081, 0.3873,
  3684. 0.0851],
  3685. [ 0.5711, -0.3788, 1.8249, -0.1766, -0.4672, 0.2160, 0.3625,
  3686. 0.0742],
  3687. [ 0.6421, -0.4008, 1.6055, 0.2160, -0.5076, -0.0534, 1.1020,
  3688. 0.3745],
  3689. [ 0.5548, -0.3936, 1.1634, -0.8105, -0.5192, -1.0696, 0.2372,
  3690. 0.3931],
  3691. [ 0.5417, -0.4355, 1.8018, -0.3383, -0.3979, 0.2622, 0.5143,
  3692. 0.2622],
  3693. [ 0.5934, -0.4276, 1.8192, -0.1458, -0.4499, 0.3777, 0.6702,
  3694. 0.1082],
  3695. [ 0.6042, -0.4273, 1.7198, 0.2184, -0.3478, 0.1149, 0.8062,
  3696. 0.1176]]], device='cuda:0')
  3697. loss_train_step before backward: tensor(0.0694, device='cuda:0', grad_fn=<MseLossBackward>)
  3698. loss_train_step after backward: tensor(0.0694, device='cuda:0', grad_fn=<MseLossBackward>)
  3699. loss_train: 10.51586477458477
  3700. step: 77
  3701. running loss: 0.1365696723972048
  3702. Train Steps: 77/90 Loss: 0.1366 torch.Size([8, 600, 800])
  3703. torch.Size([8, 8])
  3704. tensor([[0.6261, 0.4029, 0.8720, 0.3364, 0.3665, 0.3753, 0.6531, 0.5183],
  3705. [0.6198, 0.3997, 0.8582, 0.5361, 0.4117, 0.5016, 0.5942, 0.5134],
  3706. [0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6262, 0.5167],
  3707. [0.6200, 0.3978, 0.8900, 0.4550, 0.3775, 0.5200, 0.6150, 0.5367],
  3708. [0.6219, 0.4097, 0.8738, 0.3400, 0.3563, 0.4117, 0.5975, 0.5683],
  3709. [0.6201, 0.4064, 0.8688, 0.5050, 0.4225, 0.5100, 0.6138, 0.5500],
  3710. [0.6143, 0.4040, 0.8237, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973],
  3711. [0.6364, 0.4165, 0.9088, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297]],
  3712. device='cuda:0', dtype=torch.float64)
  3713. predictions are: tensor([[ 0.4612, -0.5073, 1.4651, -0.8233, -0.5214, -0.6658, 0.6923, 0.2148],
  3714. [ 0.5246, -0.4641, 1.5652, -0.0813, -0.4644, -0.0231, 0.4570, 0.2592],
  3715. [ 0.3270, -0.5956, 1.6400, -0.3893, -0.4486, -0.6250, 0.4818, 0.2420],
  3716. [ 0.6205, -0.3809, 1.6541, -0.1866, -0.4975, -0.0112, 0.6581, 0.2468],
  3717. [ 0.2243, -0.6179, 1.4108, -0.6065, -0.5477, -0.5506, 0.3383, 0.3082],
  3718. [ 0.5293, -0.4360, 1.5617, -0.0832, -0.4276, 0.0412, 0.4615, 0.2706],
  3719. [-0.2347, -0.8771, 1.2084, -1.0134, -0.3904, -1.1015, 0.3591, 0.2437],
  3720. [-0.0426, -0.7577, 1.6079, -0.4859, -0.2439, -0.7029, 0.5177, 0.2462]],
  3721. device='cuda:0', grad_fn=<AddmmBackward>)
  3722. landmarks are: tensor([[[ 0.6059, -0.4249, 1.7420, -0.7321, -0.5931, -0.5523, 0.7307,
  3723. 0.1080],
  3724. [ 0.5767, -0.4396, 1.6782, 0.1905, -0.3844, 0.0308, 0.4588,
  3725. 0.0855],
  3726. [ 0.6094, -0.3947, 1.8885, -0.2998, -0.5769, -0.6770, 0.6067,
  3727. 0.1005],
  3728. [ 0.5776, -0.4484, 1.8249, -0.1843, -0.5423, 0.1159, 0.5547,
  3729. 0.1929],
  3730. [ 0.5867, -0.3937, 1.7499, -0.7155, -0.6404, -0.3844, 0.4739,
  3731. 0.3392],
  3732. [ 0.5784, -0.4088, 1.7268, 0.0467, -0.3344, 0.0697, 0.5490,
  3733. 0.2545],
  3734. [ 0.5515, -0.4201, 1.5189, -0.7463, -0.4037, -1.3082, 0.0869,
  3735. 0.0111],
  3736. [ 0.6537, -0.3623, 1.9115, -0.2690, -0.4037, -0.8309, 0.6926,
  3737. 0.1608]]], device='cuda:0')
  3738. loss_train_step before backward: tensor(0.0475, device='cuda:0', grad_fn=<MseLossBackward>)
  3739. loss_train_step after backward: tensor(0.0475, device='cuda:0', grad_fn=<MseLossBackward>)
  3740. loss_train: 10.563358839601278
  3741. step: 78
  3742. running loss: 0.13542767743078563
  3743. Train Steps: 78/90 Loss: 0.1354 torch.Size([8, 600, 800])
  3744. torch.Size([8, 8])
  3745. tensor([[0.6131, 0.4064, 0.8638, 0.5200, 0.4788, 0.4783, 0.5258, 0.5867],
  3746. [ nan, nan, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533],
  3747. [0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742],
  3748. [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209],
  3749. [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461],
  3750. [0.6346, 0.4165, 0.9138, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471],
  3751. [ nan, nan, 0.7412, 0.2200, 0.4450, 0.1517, 0.5312, 0.4983],
  3752. [0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350]],
  3753. device='cuda:0', dtype=torch.float64)
  3754. predictions are: tensor([[ 0.6535, -0.3603, 1.6467, 0.0109, -0.3417, 0.1054, 0.5565, 0.2526],
  3755. [-0.1297, -0.7888, 1.1310, -0.9446, -0.5002, -0.8548, 0.2892, 0.2735],
  3756. [ 0.7948, -0.2646, 1.6848, 0.0678, -0.6741, -0.2577, 0.5078, 0.1569],
  3757. [ 0.2547, -0.5520, 1.5519, -0.3196, -0.3354, -0.4262, 0.5298, 0.2917],
  3758. [ 0.6743, -0.3393, 1.6766, 0.0688, -0.3600, 0.1798, 0.5958, 0.2401],
  3759. [ 0.3992, -0.5258, 1.8062, -0.2436, -0.3569, -0.1706, 0.6703, 0.2025],
  3760. [-0.3108, -0.9624, 1.0015, -1.3031, -0.3757, -1.3097, 0.3483, 0.2303],
  3761. [ 0.0091, -0.7416, 1.2802, -0.9353, -0.3341, -0.8951, 0.4326, 0.2514]],
  3762. device='cuda:0', grad_fn=<AddmmBackward>)
  3763. landmarks are: tensor([[[ 0.5462, -0.4090, 1.7037, 0.1159, -0.0746, -0.0765, 0.1427,
  3764. 0.4239],
  3765. [-2.2859, -2.2859, 1.5074, -1.0388, -0.5423, -0.9849, 0.2199,
  3766. 0.2699],
  3767. [ 0.6113, -0.4057, 1.7499, 0.3007, -0.5885, -0.5384, 0.4513,
  3768. -0.0957],
  3769. [ 0.6174, -0.3118, 1.6402, -0.2074, -0.1958, -1.0927, 0.2267,
  3770. 0.5822],
  3771. [ 0.5443, -0.3831, 1.7095, 0.1621, -0.2016, 0.1390, 0.1437,
  3772. 0.2364],
  3773. [ 0.6454, -0.3623, 1.9346, -0.4460, -0.4961, -0.2921, 1.1642,
  3774. 0.2409],
  3775. [-2.2859, -2.2859, 1.1379, -1.2697, -0.2305, -1.5854, 0.1679,
  3776. 0.0159],
  3777. [ 0.5768, -0.4031, 1.3838, -1.1527, -0.2188, -1.4216, 0.4379,
  3778. 0.1850]]], device='cuda:0')
  3779. loss_train_step before backward: tensor(0.2474, device='cuda:0', grad_fn=<MseLossBackward>)
  3780. loss_train_step after backward: tensor(0.2474, device='cuda:0', grad_fn=<MseLossBackward>)
  3781. loss_train: 10.810774769634008
  3782. step: 79
  3783. running loss: 0.13684525024853175
  3784. Train Steps: 79/90 Loss: 0.1368 torch.Size([8, 600, 800])
  3785. torch.Size([8, 8])
  3786. tensor([[0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617],
  3787. [0.6204, 0.4049, 0.7975, 0.2700, 0.3937, 0.2567, 0.5700, 0.5183],
  3788. [0.6346, 0.4144, 0.9088, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899],
  3789. [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366],
  3790. [ nan, nan, 0.7725, 0.2611, 0.3675, 0.2733, 0.5413, 0.5167],
  3791. [0.6083, 0.3957, 0.8638, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980],
  3792. [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133],
  3793. [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391]],
  3794. device='cuda:0', dtype=torch.float64)
  3795. predictions are: tensor([[ 0.5802, -0.3822, 1.6703, -0.0592, -0.3859, 0.1253, 0.4767, 0.2593],
  3796. [-0.0557, -0.7866, 1.2931, -0.9026, -0.4576, -1.0742, 0.3170, 0.2135],
  3797. [ 0.4856, -0.4461, 1.7900, -0.0398, -0.3855, -0.2339, 0.5998, 0.1871],
  3798. [ 0.5420, -0.3848, 1.5297, -0.1514, -0.4714, -0.0996, 0.5090, 0.2390],
  3799. [-0.4367, -1.0202, 1.1679, -1.1038, -0.3655, -1.1330, 0.2830, 0.2454],
  3800. [ 0.5202, -0.4176, 1.7353, -0.0741, -0.3475, 0.0032, 0.5747, 0.1865],
  3801. [ 0.5335, -0.4008, 1.6311, 0.0209, -0.3461, 0.0446, 0.4941, 0.2434],
  3802. [ 0.3662, -0.5787, 1.3590, -1.0290, -0.4626, -1.0988, 0.6388, 0.1773]],
  3803. device='cuda:0', grad_fn=<AddmmBackward>)
  3804. landmarks are: tensor([[[ 5.6374e-01, -4.1432e-01, 1.7519e+00, -7.8656e-02, -3.0554e-01,
  3805. -1.4935e-02, 3.7575e-01, 3.0839e-01],
  3806. [ 5.7985e-01, -4.1555e-01, 1.3977e+00, -1.0388e+00, -4.6721e-01,
  3807. -1.1004e+00, 3.4688e-01, 1.0824e-01],
  3808. [ 6.4542e-01, -3.7194e-01, 1.9115e+00, -1.3041e-01, -5.0762e-01,
  3809. -2.8437e-01, 1.0033e+00, 4.3864e-01],
  3810. [ 6.2895e-01, -4.3453e-01, 1.3794e+00, 3.6792e-01, -4.8453e-01,
  3811. 3.8953e-02, 9.2654e-01, 1.9283e-01],
  3812. [-2.2859e+00, -2.2859e+00, 1.2820e+00, -1.0801e+00, -5.8845e-01,
  3813. -1.0234e+00, 2.1409e-01, 1.0054e-01],
  3814. [ 5.2379e-01, -4.5797e-01, 1.7037e+00, 4.6189e-04, -2.7090e-01,
  3815. 6.2048e-02, 1.8356e-01, 1.4106e-02],
  3816. [ 5.7864e-01, -4.1409e-01, 1.7037e+00, 1.5443e-01, -1.8624e-01,
  3817. 7.3556e-02, 4.3926e-01, 8.5142e-02],
  3818. [ 5.7846e-01, -4.2587e-01, 1.4228e+00, -1.0261e+00, -4.1903e-01,
  3819. -1.2189e+00, 4.7633e-01, 2.0428e-01]]], device='cuda:0')
  3820. loss_train_step before backward: tensor(0.1085, device='cuda:0', grad_fn=<MseLossBackward>)
  3821. loss_train_step after backward: tensor(0.1085, device='cuda:0', grad_fn=<MseLossBackward>)
  3822. loss_train: 10.91929730400443
  3823. step: 80
  3824. running loss: 0.1364912163000554
  3825.  
  3826. Train Steps: 80/90 Loss: 0.1365 torch.Size([8, 600, 800])
  3827. torch.Size([8, 8])
  3828. tensor([[0.6200, 0.3978, 0.8900, 0.4550, 0.3775, 0.5200, 0.6150, 0.5367],
  3829. [0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411],
  3830. [0.6085, 0.4008, 0.8588, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283],
  3831. [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320],
  3832. [0.6133, 0.4066, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358],
  3833. [0.6127, 0.4066, 0.8550, 0.5567, 0.4662, 0.5141, 0.5070, 0.5412],
  3834. [0.6263, 0.4065, 0.9038, 0.4317, 0.3588, 0.4550, 0.6325, 0.5250],
  3835. [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329]],
  3836. device='cuda:0', dtype=torch.float64)
  3837. predictions are: tensor([[ 0.6595, -0.3411, 1.7133, -0.1565, -0.4309, 0.0339, 0.6960, 0.1918],
  3838. [-0.1699, -0.8419, 1.3545, -0.8984, -0.3419, -1.0833, 0.3931, 0.2011],
  3839. [ 0.3309, -0.5462, 1.7022, -0.0703, -0.1565, -0.1186, 0.5422, 0.2256],
  3840. [ 0.6394, -0.3299, 1.6054, -0.1054, -0.4234, 0.0326, 0.5868, 0.2038],
  3841. [-0.1806, -0.9008, 1.0618, -1.1263, -0.4600, -1.2427, 0.2815, 0.1938],
  3842. [ 0.5022, -0.4241, 1.6264, 0.0332, -0.2760, -0.0482, 0.5249, 0.2056],
  3843. [ 0.4720, -0.4786, 1.7559, -0.1731, -0.4580, -0.1977, 0.5077, 0.1997],
  3844. [ 0.3408, -0.5362, 1.4742, -0.6118, -0.5730, -0.7129, 0.3912, 0.1708]],
  3845. device='cuda:0', grad_fn=<AddmmBackward>)
  3846. landmarks are: tensor([[[ 0.5776, -0.4484, 1.8249, -0.1843, -0.5423, 0.1159, 0.5547,
  3847. 0.1929],
  3848. [ 0.5713, -0.4321, 1.4975, -0.8134, -0.3094, -1.3345, 0.3779,
  3849. 0.2134],
  3850. [ 0.5245, -0.4347, 1.6806, 0.1159, 0.0046, 0.0129, 0.1044,
  3851. 0.1544],
  3852. [ 0.6273, -0.4393, 1.6402, 0.1313, -0.5076, 0.0467, 1.1532,
  3853. 0.1715],
  3854. [ 0.5470, -0.4081, 0.8492, -1.0773, -0.5307, -1.1620, 0.0912,
  3855. 0.1890],
  3856. [ 0.5443, -0.4079, 1.6633, 0.2853, -0.1325, 0.0888, 0.0558,
  3857. 0.2138],
  3858. [ 0.6070, -0.4085, 1.8885, -0.2921, -0.6289, -0.1843, 0.6356,
  3859. 0.1390],
  3860. [ 0.5680, -0.4393, 1.5920, -0.6672, -0.6453, -0.5457, 0.5149,
  3861. 0.1753]]], device='cuda:0')
  3862. loss_train_step before backward: tensor(0.0473, device='cuda:0', grad_fn=<MseLossBackward>)
  3863. loss_train_step after backward: tensor(0.0473, device='cuda:0', grad_fn=<MseLossBackward>)
  3864. loss_train: 10.966615419834852
  3865. step: 81
  3866. running loss: 0.13539031382512162
  3867. Train Steps: 81/90 Loss: 0.1354 torch.Size([8, 600, 800])
  3868. torch.Size([8, 8])
  3869. tensor([[0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153],
  3870. [0.6346, 0.4144, 0.9088, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899],
  3871. [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114],
  3872. [0.6099, 0.4030, 0.8638, 0.5117, 0.4983, 0.4965, 0.5086, 0.5388],
  3873. [0.6200, 0.4098, 0.8237, 0.2917, 0.4012, 0.2967, 0.6000, 0.5683],
  3874. [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478],
  3875. [ nan, nan, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633],
  3876. [ nan, nan, 0.7515, 0.2708, 0.3987, 0.2267, 0.5162, 0.5567]],
  3877. device='cuda:0', dtype=torch.float64)
  3878. predictions are: tensor([[ 0.6577, -0.3297, 1.6868, 0.0782, -0.3042, 0.1605, 0.5671, 0.2163],
  3879. [ 0.6820, -0.3201, 1.8332, 0.0495, -0.4130, -0.1604, 0.6454, 0.1406],
  3880. [ 0.6271, -0.3454, 1.7065, 0.1052, -0.2852, 0.2045, 0.5946, 0.2010],
  3881. [ 0.4953, -0.4632, 1.7364, 0.0277, -0.2168, -0.0399, 0.5300, 0.1936],
  3882. [ 0.1193, -0.6399, 1.4553, -0.7747, -0.3613, -0.8269, 0.5000, 0.1925],
  3883. [ 0.4106, -0.5026, 1.6519, -0.4772, -0.4520, -0.6562, 0.5282, 0.1420],
  3884. [-0.2983, -0.9217, 1.1456, -0.9940, -0.4474, -1.0575, 0.2386, 0.2039],
  3885. [-0.3803, -0.9827, 1.1318, -1.0392, -0.4033, -1.1177, 0.2494, 0.1996]],
  3886. device='cuda:0', grad_fn=<AddmmBackward>)
  3887. landmarks are: tensor([[[ 0.5548, -0.4682, 1.7309, 0.0966, -0.1394, 0.1757, 0.5045,
  3888. 0.0942],
  3889. [ 0.6454, -0.3719, 1.9115, -0.1304, -0.5076, -0.2844, 1.0033,
  3890. 0.4386],
  3891. [ 0.5450, -0.4730, 1.7420, 0.1372, -0.1919, 0.2614, 0.4976,
  3892. 0.0764],
  3893. [ 0.5309, -0.4246, 1.7037, 0.0774, 0.0158, 0.0075, 0.0635,
  3894. 0.2026],
  3895. [ 0.5778, -0.3930, 1.5189, -0.9387, -0.4326, -0.9156, 0.4855,
  3896. 0.3392],
  3897. [ 0.5792, -0.4052, 1.8214, -0.6587, -0.5384, -0.8924, 0.4381,
  3898. 0.2442],
  3899. [-2.2859, -2.2859, 1.2030, -1.0288, -0.4961, -1.1081, 0.0813,
  3900. 0.3161],
  3901. [-2.2859, -2.2859, 1.1854, -1.0352, -0.4441, -1.2390, 0.0986,
  3902. 0.2853]]], device='cuda:0')
  3903. loss_train_step before backward: tensor(0.1947, device='cuda:0', grad_fn=<MseLossBackward>)
  3904. loss_train_step after backward: tensor(0.1947, device='cuda:0', grad_fn=<MseLossBackward>)
  3905. loss_train: 11.161314766854048
  3906. step: 82
  3907. running loss: 0.1361135947177323
  3908. Train Steps: 82/90 Loss: 0.1361 torch.Size([8, 600, 800])
  3909. torch.Size([8, 8])
  3910. tensor([[0.6371, 0.4092, 0.8337, 0.5850, 0.3950, 0.5117, 0.6559, 0.5262],
  3911. [ nan, nan, 0.6900, 0.1917, 0.3937, 0.2367, 0.5240, 0.5246],
  3912. [0.6300, 0.4133, 0.8538, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413],
  3913. [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050],
  3914. [0.6193, 0.4034, 0.7757, 0.2347, 0.3733, 0.2919, 0.5930, 0.4926],
  3915. [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750],
  3916. [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726],
  3917. [ nan, nan, 0.7725, 0.2611, 0.3675, 0.2733, 0.5413, 0.5167]],
  3918. device='cuda:0', dtype=torch.float64)
  3919. predictions are: tensor([[ 1.0737, -0.0740, 1.8738, 0.5913, -0.4413, 0.5408, 0.5651, 0.1580],
  3920. [-0.5130, -1.0604, 1.2191, -0.9690, -0.2829, -1.0806, 0.2641, 0.2092],
  3921. [-0.1248, -0.8598, 1.8897, -0.6127, 0.0917, -0.7495, 0.8417, 0.1448],
  3922. [ 0.8726, -0.2283, 1.9341, 0.3120, -0.4677, 0.1949, 0.5191, 0.1281],
  3923. [ 0.4474, -0.4906, 1.3738, -0.7276, -0.4931, -0.7259, 0.4807, 0.1319],
  3924. [ 0.5036, -0.4894, 1.3357, -0.5319, -0.4911, -0.5718, 0.4147, 0.2147],
  3925. [ 0.4899, -0.4758, 1.3907, -0.6307, -0.4970, -0.6554, 0.4247, 0.1343],
  3926. [-0.4548, -1.0278, 1.3212, -0.8165, -0.3190, -0.8877, 0.2281, 0.1962]],
  3927. device='cuda:0', grad_fn=<AddmmBackward>)
  3928. landmarks are: tensor([[[ 0.6569, -0.3960, 1.5651, 0.4162, -0.4614, 0.0774, 0.7438,
  3929. 0.1447],
  3930. [-2.2859, -2.2859, 0.9012, -1.4006, -0.4672, -1.1928, 0.1342,
  3931. 0.1373],
  3932. [ 0.6240, -0.3768, 1.6575, -1.2851, 0.2949, -1.2467, 1.1276,
  3933. 0.2142],
  3934. [ 0.5844, -0.4466, 1.8423, -0.1997, -0.5942, -0.3998, 0.4219,
  3935. 0.0467],
  3936. [ 0.5746, -0.4225, 1.2968, -1.2019, -0.5615, -0.9374, 0.4533,
  3937. -0.0108],
  3938. [ 0.5805, -0.3818, 1.0282, -1.1774, -0.4903, -0.9310, 0.5894,
  3939. 0.3700],
  3940. [ 0.4997, -0.4446, 1.1610, -0.9772, -0.6058, -1.0311, 0.1404,
  3941. -0.1031],
  3942. [-2.2859, -2.2859, 1.2820, -1.0801, -0.5885, -1.0234, 0.2141,
  3943. 0.1005]]], device='cuda:0')
  3944. loss_train_step before backward: tensor(0.2300, device='cuda:0', grad_fn=<MseLossBackward>)
  3945. loss_train_step after backward: tensor(0.2300, device='cuda:0', grad_fn=<MseLossBackward>)
  3946. loss_train: 11.391309630125761
  3947. step: 83
  3948. running loss: 0.1372446943388646
  3949. Train Steps: 83/90 Loss: 0.1372 torch.Size([8, 600, 800])
  3950. torch.Size([8, 8])
  3951. tensor([[0.6193, 0.4050, 0.7313, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656],
  3952. [0.6346, 0.4092, 0.7712, 0.5917, 0.4037, 0.4767, 0.7343, 0.5725],
  3953. [0.6206, 0.4001, 0.8900, 0.3933, 0.3588, 0.3567, 0.5837, 0.5083],
  3954. [0.6147, 0.4026, 0.6600, 0.2467, 0.4088, 0.2150, 0.5489, 0.5773],
  3955. [0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461],
  3956. [0.6203, 0.4096, 0.8862, 0.4267, 0.3538, 0.4117, 0.6025, 0.5650],
  3957. [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133],
  3958. [0.6136, 0.4117, 0.8700, 0.5167, 0.4188, 0.5083, 0.5147, 0.5495]],
  3959. device='cuda:0', dtype=torch.float64)
  3960. predictions are: tensor([[-0.0929, -0.8711, 1.2389, -1.0993, -0.3887, -1.2384, 0.4037, 0.1447],
  3961. [ 0.3426, -0.5333, 1.5412, -0.1081, -0.3541, -0.0805, 0.4267, 0.2252],
  3962. [ 0.4016, -0.5626, 1.8441, -0.3404, -0.4835, -0.6410, 0.5482, 0.0721],
  3963. [-0.2117, -0.9704, 1.2032, -1.0936, -0.4017, -1.2933, 0.2922, 0.1594],
  3964. [ 0.5968, -0.3992, 1.6112, -0.5510, -0.5232, -0.4294, 0.6442, 0.1186],
  3965. [ 0.2000, -0.6920, 1.8410, -0.2265, -0.3198, -0.3455, 0.4740, 0.1843],
  3966. [ 0.3778, -0.5193, 1.7693, 0.1073, -0.1769, 0.0373, 0.4870, 0.1666],
  3967. [ 0.5991, -0.3642, 1.7881, 0.2127, -0.2926, 0.1899, 0.5110, 0.1492]],
  3968. device='cuda:0', grad_fn=<AddmmBackward>)
  3969. landmarks are: tensor([[[ 0.5746, -0.4153, 1.0917, -1.1620, -0.4037, -1.3082, 0.3234,
  3970. 0.3267],
  3971. [ 0.6454, -0.3960, 1.2764, 0.4470, -0.4210, -0.0842, 1.1057,
  3972. 0.3585],
  3973. [ 0.5807, -0.4378, 1.8249, -0.4691, -0.6289, -0.6385, 0.4104,
  3974. 0.0620],
  3975. [ 0.5532, -0.4264, 0.7626, -1.1466, -0.3979, -1.2928, 0.2494,
  3976. 0.3808],
  3977. [ 0.5124, -0.4483, 1.5420, -0.6231, -0.7155, -0.4152, 0.2679,
  3978. 0.2365],
  3979. [ 0.5790, -0.3940, 1.8076, -0.3152, -0.6520, -0.3844, 0.4970,
  3980. 0.3238],
  3981. [ 0.5786, -0.4141, 1.7037, 0.1544, -0.1862, 0.0736, 0.4393,
  3982. 0.0851],
  3983. [ 0.5482, -0.3841, 1.7326, 0.1005, -0.3517, 0.0620, 0.0912,
  3984. 0.2522]]], device='cuda:0')
  3985. loss_train_step before backward: tensor(0.0595, device='cuda:0', grad_fn=<MseLossBackward>)
  3986. loss_train_step after backward: tensor(0.0595, device='cuda:0', grad_fn=<MseLossBackward>)
  3987. loss_train: 11.450813557952642
  3988. step: 84
  3989. running loss: 0.13631920902324574
  3990.  
  3991. Train Steps: 84/90 Loss: 0.1363 torch.Size([8, 600, 800])
  3992. torch.Size([8, 8])
  3993. tensor([[0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398],
  3994. [0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767],
  3995. [0.6127, 0.4084, 0.8700, 0.4467, 0.3987, 0.4317, 0.5013, 0.5471],
  3996. [0.6193, 0.4108, 0.7438, 0.2700, 0.3650, 0.3683, 0.6238, 0.5717],
  3997. [0.6257, 0.4034, 0.8287, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901],
  3998. [0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400],
  3999. [ nan, nan, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737],
  4000. [0.6040, 0.4002, 0.7338, 0.2267, 0.3975, 0.2100, 0.5231, 0.4778]],
  4001. device='cuda:0', dtype=torch.float64)
  4002. predictions are: tensor([[ 1.0472, -0.1337, 1.8173, 0.4242, -0.5189, 0.1716, 0.5610, 0.1083],
  4003. [ 0.8067, -0.3118, 1.8637, 0.3343, -0.3867, 0.3309, 0.5161, 0.1508],
  4004. [ 0.3921, -0.5289, 1.6973, -0.0309, -0.4470, -0.0295, 0.2790, 0.2103],
  4005. [ 0.4866, -0.4627, 1.3325, -0.6769, -0.5362, -0.4760, 0.4466, 0.1793],
  4006. [ 0.3842, -0.5916, 1.5074, -0.8420, -0.4493, -0.9451, 0.6339, 0.0836],
  4007. [-0.4744, -1.0623, 1.4592, -0.7937, -0.2200, -0.9876, 0.3152, 0.1966],
  4008. [-0.6824, -1.2032, 1.6402, -0.9928, 0.0528, -1.1821, 0.6461, 0.1668],
  4009. [-0.1324, -0.9212, 1.2158, -0.9944, -0.4907, -1.1888, 0.2543, 0.1415]],
  4010. device='cuda:0', grad_fn=<AddmmBackward>)
  4011. landmarks are: tensor([[[ 0.5786, -0.4463, 1.6655, 0.2216, -0.5115, -0.2675, 0.4236,
  4012. 0.2075],
  4013. [ 0.5610, -0.4381, 1.7730, 0.0390, -0.4326, -0.1458, 0.1794,
  4014. 0.3777],
  4015. [ 0.5441, -0.3997, 1.7326, -0.2228, -0.4441, -0.2921, 0.0296,
  4016. 0.2409],
  4017. [ 0.5747, -0.3886, 1.1494, -1.0388, -0.6000, -0.5846, 0.5952,
  4018. 0.3546],
  4019. [ 0.6042, -0.4225, 1.5420, -1.2082, -0.4730, -1.0311, 0.6380,
  4020. -0.0220],
  4021. [ 0.5911, -0.3888, 1.4727, -0.9541, -0.0919, -1.4930, 0.3988,
  4022. 0.2083],
  4023. [-2.2859, -2.2859, 1.6979, -1.1081, 0.2487, -1.2697, 1.1020,
  4024. 0.3638],
  4025. [ 0.5037, -0.4375, 1.1032, -1.2390, -0.4499, -1.3159, 0.1301,
  4026. -0.0791]]], device='cuda:0')
  4027. loss_train_step before backward: tensor(0.1343, device='cuda:0', grad_fn=<MseLossBackward>)
  4028. loss_train_step after backward: tensor(0.1343, device='cuda:0', grad_fn=<MseLossBackward>)
  4029. loss_train: 11.58513193950057
  4030. step: 85
  4031. running loss: 0.1362956698764773
  4032. Train Steps: 85/90 Loss: 0.1363 torch.Size([8, 600, 800])
  4033. torch.Size([8, 8])
  4034. tensor([[0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284],
  4035. [0.6112, 0.4029, 0.8638, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567],
  4036. [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5513, 0.5750],
  4037. [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290],
  4038. [0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540],
  4039. [ nan, nan, 0.8938, 0.2850, 0.4662, 0.3117, 0.7406, 0.5528],
  4040. [0.6230, 0.4113, 0.7213, 0.1983, 0.4325, 0.2367, 0.6262, 0.5400],
  4041. [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656]],
  4042. device='cuda:0', dtype=torch.float64)
  4043. predictions are: tensor([[-6.5229e-02, -8.8510e-01, 1.2439e+00, -9.6002e-01, -4.1075e-01,
  4044. -1.1328e+00, 2.9228e-01, 1.5930e-01],
  4045. [ 5.0083e-01, -4.7344e-01, 1.8331e+00, 1.2776e-01, -1.5367e-01,
  4046. 1.6144e-01, 5.3884e-01, 1.9881e-01],
  4047. [-5.0049e-01, -1.1383e+00, 1.0559e+00, -1.1379e+00, -4.0530e-01,
  4048. -1.2439e+00, 2.0949e-01, 1.9963e-01],
  4049. [ 8.3195e-01, -2.8182e-01, 1.5013e+00, -5.1990e-01, -6.1314e-01,
  4050. -2.9477e-01, 6.0160e-01, 1.3491e-01],
  4051. [ 3.4898e-01, -5.4086e-01, 1.8459e+00, -2.4536e-02, -3.6836e-01,
  4052. -4.6690e-01, 5.0389e-01, 1.1189e-01],
  4053. [-2.3398e-01, -9.3486e-01, 1.8301e+00, -6.1287e-01, -1.3023e-01,
  4054. -8.4830e-01, 6.8406e-01, 1.4688e-01],
  4055. [-2.9074e-01, -9.9740e-01, 1.3294e+00, -8.7116e-01, -3.7710e-01,
  4056. -1.1082e+00, 3.2404e-01, 1.8692e-01],
  4057. [ 9.3682e-01, -1.8088e-01, 1.6402e+00, 1.7329e-03, -5.3351e-01,
  4058. 1.7422e-01, 5.1208e-01, 1.9848e-01]], device='cuda:0',
  4059. grad_fn=<AddmmBackward>)
  4060. landmarks are: tensor([[[ 0.5761, -0.4070, 1.2452, -1.2541, -0.1725, -1.4835, 0.4511,
  4061. 0.1545],
  4062. [ 0.5371, -0.4252, 1.7037, -0.0688, -0.0342, 0.0620, 0.3758,
  4063. 0.2853],
  4064. [ 0.5663, -0.4396, 0.8261, -1.1312, -0.2940, -1.3929, 0.2603,
  4065. 0.3700],
  4066. [ 0.5697, -0.4538, 1.5308, -0.8703, -0.6572, -0.3639, 0.5739,
  4067. 0.1576],
  4068. [ 0.6240, -0.3912, 1.9115, -0.2382, -0.3979, -0.8694, 0.8644,
  4069. 0.2730],
  4070. [-2.2859, -2.2859, 1.8423, -0.9695, -0.1323, -0.8463, 1.1349,
  4071. 0.2676],
  4072. [ 0.5917, -0.3861, 1.0455, -1.3698, -0.2882, -1.1928, 0.6067,
  4073. 0.2083],
  4074. [ 0.5283, -0.4429, 1.5940, -0.2844, -0.5827, -0.1458, 0.2823,
  4075. 0.3267]]], device='cuda:0')
  4076. loss_train_step before backward: tensor(0.1866, device='cuda:0', grad_fn=<MseLossBackward>)
  4077. loss_train_step after backward: tensor(0.1866, device='cuda:0', grad_fn=<MseLossBackward>)
  4078. loss_train: 11.77172925695777
  4079. step: 86
  4080. running loss: 0.1368805727553229
  4081. Train Steps: 86/90 Loss: 0.1369 torch.Size([8, 600, 800])
  4082. torch.Size([8, 8])
  4083. tensor([[0.6189, 0.4049, 0.8888, 0.4417, 0.4213, 0.5200, 0.5988, 0.5633],
  4084. [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118],
  4085. [ nan, nan, 0.7097, 0.2346, 0.4250, 0.1850, 0.5175, 0.5583],
  4086. [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082],
  4087. [0.6182, 0.4058, 0.8738, 0.4350, 0.3563, 0.3400, 0.5290, 0.5822],
  4088. [0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117],
  4089. [0.6122, 0.3993, 0.8738, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927],
  4090. [0.6219, 0.4114, 0.8175, 0.2817, 0.3925, 0.2783, 0.5900, 0.5350]],
  4091. device='cuda:0', dtype=torch.float64)
  4092. predictions are: tensor([[ 4.4188e-01, -5.1116e-01, 1.7300e+00, -2.0806e-01, -3.4683e-01,
  4093. -4.7736e-02, 4.7247e-01, 2.4311e-01],
  4094. [ 7.5987e-01, -3.2086e-01, 1.5680e+00, -9.0686e-04, -4.7657e-01,
  4095. -1.4873e-01, 4.9563e-01, 2.3884e-01],
  4096. [-9.0501e-01, -1.4015e+00, 1.0909e+00, -1.4329e+00, -3.3467e-01,
  4097. -1.6594e+00, 2.6124e-01, 1.6862e-01],
  4098. [ 6.2424e-01, -4.0011e-01, 1.6721e+00, -4.6716e-02, -2.9070e-01,
  4099. 3.5843e-02, 5.2984e-01, 2.0643e-01],
  4100. [ 2.5168e-01, -6.7213e-01, 1.5804e+00, -4.8583e-01, -5.6401e-01,
  4101. -7.9968e-01, 3.3252e-01, 1.5927e-01],
  4102. [ 7.6404e-01, -3.3559e-01, 1.7661e+00, -1.6401e-01, -4.4549e-01,
  4103. -1.6386e-01, 6.5402e-01, 1.4741e-01],
  4104. [ 3.2009e-01, -6.1624e-01, 1.7472e+00, -2.7111e-01, -2.9466e-01,
  4105. -3.7289e-01, 5.0791e-01, 1.5344e-01],
  4106. [-7.9375e-01, -1.3188e+00, 1.3903e+00, -1.2725e+00, -2.4440e-01,
  4107. -1.5361e+00, 3.7710e-01, 1.7771e-01]], device='cuda:0',
  4108. grad_fn=<AddmmBackward>)
  4109. landmarks are: tensor([[[ 0.5726, -0.4159, 1.8192, -0.2459, -0.3402, 0.1159, 0.4797,
  4110. 0.3161],
  4111. [ 0.6121, -0.3844, 1.4556, 0.3936, -0.4383, -0.1689, 0.2925,
  4112. 0.5401],
  4113. [-2.2859, -2.2859, 0.9922, -1.2021, -0.3229, -1.4314, 0.1044,
  4114. 0.2930],
  4115. [ 0.5450, -0.4706, 1.7643, 0.0722, -0.3708, 0.3200, 0.4854,
  4116. 0.0612],
  4117. [ 0.5697, -0.4113, 1.7499, -0.2767, -0.6404, -0.7155, 0.1576,
  4118. 0.4032],
  4119. [ 0.6262, -0.4461, 1.8480, -0.0534, -0.5827, -0.1227, 0.6587,
  4120. 0.0774],
  4121. [ 0.5417, -0.4417, 1.7499, -0.1304, -0.1994, -0.0324, 0.0951,
  4122. -0.0099],
  4123. [ 0.5868, -0.3858, 1.4901, -0.9849, -0.4730, -1.0003, 0.4393,
  4124. 0.1852]]], device='cuda:0')
  4125. loss_train_step before backward: tensor(0.1162, device='cuda:0', grad_fn=<MseLossBackward>)
  4126. loss_train_step after backward: tensor(0.1162, device='cuda:0', grad_fn=<MseLossBackward>)
  4127. loss_train: 11.88794718310237
  4128. step: 87
  4129. running loss: 0.13664307107014217
  4130. Train Steps: 87/90 Loss: 0.1366 torch.Size([8, 600, 800])
  4131. torch.Size([8, 8])
  4132. tensor([[0.6185, 0.4042, 0.7700, 0.2250, 0.4062, 0.2117, 0.5763, 0.5150],
  4133. [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220],
  4134. [0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631],
  4135. [0.6267, 0.4080, 0.8438, 0.2633, 0.4763, 0.1800, 0.6259, 0.5240],
  4136. [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118],
  4137. [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5963, 0.6217],
  4138. [0.6219, 0.4097, 0.8738, 0.3400, 0.3563, 0.4117, 0.5975, 0.5683],
  4139. [0.6296, 0.4045, 0.9138, 0.4100, 0.4232, 0.4242, 0.7422, 0.5297]],
  4140. device='cuda:0', dtype=torch.float64)
  4141. predictions are: tensor([[-0.4403, -1.1100, 1.3071, -1.0556, -0.4015, -1.3357, 0.2999, 0.1572],
  4142. [-0.4792, -1.1614, 1.2105, -1.1959, -0.3678, -1.4565, 0.3273, 0.1607],
  4143. [ 0.5990, -0.3910, 1.5966, -0.3210, -0.4170, 0.0495, 0.5252, 0.2310],
  4144. [-0.2928, -1.0263, 1.6465, -0.8846, -0.2008, -1.1616, 0.5395, 0.1867],
  4145. [ 0.7680, -0.3232, 1.6906, 0.2699, -0.4037, 0.1270, 0.4906, 0.2611],
  4146. [ 0.1373, -0.7692, 1.2821, -0.8456, -0.4277, -0.9613, 0.3662, 0.2293],
  4147. [ 0.2150, -0.6974, 1.6454, -0.4520, -0.5265, -0.5073, 0.3394, 0.2385],
  4148. [ 0.9976, -0.1960, 1.7188, -0.2152, -0.4970, -0.1576, 0.6804, 0.1570]],
  4149. device='cuda:0', grad_fn=<AddmmBackward>)
  4150. landmarks are: tensor([[[ 0.5707, -0.4189, 1.2707, -1.2467, -0.4095, -1.3082, 0.3758,
  4151. 0.0928],
  4152. [ 0.5566, -0.4160, 0.9935, -1.3313, -0.2824, -1.5161, 0.2144,
  4153. 0.1253],
  4154. [ 0.5027, -0.4289, 1.5478, -0.4229, -0.6346, -0.0303, 0.3223,
  4155. 0.3148],
  4156. [ 0.6088, -0.4015, 1.6113, -1.0696, -0.0861, -1.4545, 0.6051,
  4157. 0.1343],
  4158. [ 0.6121, -0.3844, 1.4556, 0.3936, -0.4383, -0.1689, 0.2925,
  4159. 0.5401],
  4160. [ 0.5687, -0.4088, 1.0397, -1.1466, -0.3113, -1.1928, 0.4681,
  4161. 0.5855],
  4162. [ 0.5867, -0.3937, 1.7499, -0.7155, -0.6404, -0.3844, 0.4739,
  4163. 0.3392],
  4164. [ 0.6224, -0.4177, 1.9346, -0.3921, -0.3314, -0.3264, 1.1422,
  4165. 0.1608]]], device='cuda:0')
  4166. loss_train_step before backward: tensor(0.1034, device='cuda:0', grad_fn=<MseLossBackward>)
  4167. loss_train_step after backward: tensor(0.1034, device='cuda:0', grad_fn=<MseLossBackward>)
  4168. loss_train: 11.991319220513105
  4169. step: 88
  4170. running loss: 0.13626499114219437
  4171.  
  4172. Train Steps: 88/90 Loss: 0.1363 torch.Size([8, 600, 800])
  4173. torch.Size([8, 8])
  4174. tensor([[0.6246, 0.4008, 0.8757, 0.5088, 0.4101, 0.5392, 0.6644, 0.5133],
  4175. [0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6063, 0.6083],
  4176. [0.6081, 0.3950, 0.8538, 0.4667, 0.3850, 0.4917, 0.5342, 0.4954],
  4177. [0.6117, 0.4019, 0.8538, 0.4067, 0.3513, 0.3583, 0.5663, 0.5133],
  4178. [0.6182, 0.3930, 0.8841, 0.3892, 0.3556, 0.4967, 0.6222, 0.5279],
  4179. [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869],
  4180. [0.6276, 0.4120, 0.8738, 0.3133, 0.4225, 0.2217, 0.6203, 0.4892],
  4181. [0.6293, 0.4097, 0.8800, 0.2517, 0.5262, 0.2600, 0.7430, 0.5378]],
  4182. device='cuda:0', dtype=torch.float64)
  4183. predictions are: tensor([[ 0.4388, -0.5066, 1.5570, -0.0962, -0.3215, -0.0344, 0.4251, 0.2929],
  4184. [ 0.4409, -0.5333, 1.5423, -0.5185, -0.5618, -0.4258, 0.3752, 0.2538],
  4185. [ 0.4461, -0.5290, 1.4972, -0.4042, -0.4689, -0.3502, 0.4340, 0.2296],
  4186. [ 0.1839, -0.7194, 1.4752, -0.6965, -0.6311, -0.9276, 0.2914, 0.2059],
  4187. [ 0.3816, -0.5760, 1.4603, -0.5821, -0.4492, -0.2598, 0.5169, 0.2633],
  4188. [ 0.4607, -0.5290, 1.5899, -0.0559, -0.5009, -0.4150, 0.3185, 0.2172],
  4189. [-0.2723, -1.0297, 1.5811, -0.9711, -0.3396, -1.3682, 0.4855, 0.1946],
  4190. [-0.1325, -0.9935, 1.5606, -1.2217, -0.1039, -1.4336, 0.7863, 0.1774]],
  4191. device='cuda:0', grad_fn=<AddmmBackward>)
  4192. landmarks are: tensor([[[ 0.5993, -0.4345, 1.7587, 0.0641, -0.3917, 0.2048, 0.7827,
  4193. 0.0852],
  4194. [ 0.5774, -0.3868, 1.6286, -0.5692, -0.6462, -0.2767, 0.5143,
  4195. 0.5239],
  4196. [ 0.5227, -0.4615, 1.6575, -0.1304, -0.5076, -0.0149, 0.1815,
  4197. 0.0021],
  4198. [ 0.5393, -0.4294, 1.6575, -0.4075, -0.6635, -0.6308, 0.3296,
  4199. 0.0851],
  4200. [ 0.5697, -0.4706, 1.7976, -0.4884, -0.6433, 0.0081, 0.5878,
  4201. 0.1525],
  4202. [ 0.5320, -0.4488, 1.6633, 0.3315, -0.5018, -0.2459, 0.0828,
  4203. -0.0370],
  4204. [ 0.6128, -0.3828, 1.7499, -0.8386, -0.3344, -1.2620, 0.5792,
  4205. -0.0263],
  4206. [ 0.6207, -0.3936, 1.7788, -1.1235, 0.1448, -1.0850, 1.1459,
  4207. 0.1982]]], device='cuda:0')
  4208. loss_train_step before backward: tensor(0.0703, device='cuda:0', grad_fn=<MseLossBackward>)
  4209. loss_train_step after backward: tensor(0.0703, device='cuda:0', grad_fn=<MseLossBackward>)
  4210. loss_train: 12.061648171395063
  4211. step: 89
  4212. running loss: 0.1355241367572479
  4213. Train Steps: 89/90 Loss: 0.1355 torch.Size([8, 600, 800])
  4214. torch.Size([8, 8])
  4215. tensor([[0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719],
  4216. [0.6095, 0.3970, 0.8688, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940],
  4217. [0.6199, 0.4071, 0.7600, 0.2117, 0.4037, 0.2767, 0.6138, 0.5550],
  4218. [0.6102, 0.4020, 0.8638, 0.3717, 0.3625, 0.5017, 0.6038, 0.5500],
  4219. [0.6161, 0.4024, 0.8662, 0.4683, 0.4935, 0.5364, 0.6063, 0.5567],
  4220. [ nan, nan, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563],
  4221. [0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742],
  4222. [0.6122, 0.3993, 0.8738, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927]],
  4223. device='cuda:0', dtype=torch.float64)
  4224. predictions are: tensor([[ 0.6735, -0.3933, 1.4704, -0.5424, -0.5828, -0.4557, 0.4849, 0.2398],
  4225. [ 0.2983, -0.6639, 1.7263, -0.3416, -0.2465, -0.5271, 0.5216, 0.2376],
  4226. [-0.2158, -0.9857, 1.3755, -1.1546, -0.3586, -1.3447, 0.3959, 0.2534],
  4227. [ 0.5483, -0.4424, 1.4602, -0.5501, -0.4750, -0.1833, 0.5094, 0.3046],
  4228. [ 0.3909, -0.5643, 1.7396, -0.2520, -0.2310, -0.3395, 0.5488, 0.2893],
  4229. [-0.4585, -1.1134, 1.1996, -1.0510, -0.5404, -1.2248, 0.1798, 0.2508],
  4230. [ 0.7076, -0.4017, 1.5375, -0.7291, -0.5402, -0.9808, 0.5706, 0.1742],
  4231. [ 0.3275, -0.6220, 1.6696, -0.2631, -0.3304, -0.3368, 0.3925, 0.2430]],
  4232. device='cuda:0', grad_fn=<AddmmBackward>)
  4233. landmarks are: tensor([[[ 0.5367, -0.4294, 1.5709, -0.4999, -0.6693, -0.3075, 0.2455,
  4234. 0.3559],
  4235. [ 0.5292, -0.4520, 1.7268, -0.0842, -0.0413, -0.0324, 0.1116,
  4236. -0.0039],
  4237. [ 0.5775, -0.4054, 1.2245, -1.3082, -0.4210, -1.0080, 0.5490,
  4238. 0.2776],
  4239. [ 0.5324, -0.4293, 1.7037, -0.5692, -0.6115, 0.0313, 0.5028,
  4240. 0.2545],
  4241. [ 0.5598, -0.4273, 1.7152, -0.1227, -0.0065, 0.1917, 0.5143,
  4242. 0.2853],
  4243. [-2.2859, -2.2859, 1.4006, -0.8105, -0.6115, -0.8232, 0.0419,
  4244. 0.2837],
  4245. [ 0.6134, -0.4218, 1.7268, -0.6154, -0.4730, -1.0850, 0.5463,
  4246. -0.0957],
  4247. [ 0.5417, -0.4417, 1.7499, -0.1304, -0.1994, -0.0324, 0.0951,
  4248. -0.0099]]], device='cuda:0')
  4249. loss_train_step before backward: tensor(0.1250, device='cuda:0', grad_fn=<MseLossBackward>)
  4250. loss_train_step after backward: tensor(0.1250, device='cuda:0', grad_fn=<MseLossBackward>)
  4251. loss_train: 12.186646927148104
  4252. step: 90
  4253. running loss: 0.13540718807942337
  4254. Valid Steps: 10/10 Loss: nan 7.3225
  4255. --------------------------------------------------
  4256. Epoch: 1 Train Loss: 0.1354 Valid Loss: nan
  4257. --------------------------------------------------
  4258. size of train loader is: 90
  4259. torch.Size([8, 600, 800])
  4260. torch.Size([8, 8])
  4261. tensor([[0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373],
  4262. [0.6162, 0.4134, 0.6700, 0.2467, 0.3962, 0.2533, 0.5737, 0.5467],
  4263. [0.6138, 0.4054, 0.8750, 0.4750, 0.4363, 0.5017, 0.5086, 0.5822],
  4264. [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170],
  4265. [0.6197, 0.4050, 0.7527, 0.2000, 0.4042, 0.2249, 0.5895, 0.4995],
  4266. [0.6219, 0.4097, 0.8738, 0.3400, 0.3563, 0.4117, 0.5975, 0.5683],
  4267. [0.6097, 0.3988, 0.8650, 0.5250, 0.4213, 0.5200, 0.5675, 0.5050],
  4268. [0.6336, 0.4154, 0.8900, 0.2767, 0.4988, 0.2867, 0.7422, 0.5540]],
  4269. device='cuda:0', dtype=torch.float64)
  4270. predictions are: tensor([[ 7.1007e-01, -3.5610e-01, 1.6399e+00, 6.3848e-02, -5.3525e-01,
  4271. -1.6645e-01, 3.9725e-01, 2.2814e-01],
  4272. [ 5.5852e-02, -7.9860e-01, 1.1793e+00, -1.0058e+00, -5.5704e-01,
  4273. -1.0473e+00, 2.7695e-01, 2.8426e-01],
  4274. [ 3.4867e-01, -5.3798e-01, 1.6127e+00, -2.1721e-01, -3.6062e-01,
  4275. -9.5872e-02, 3.0168e-01, 2.9245e-01],
  4276. [ 1.2011e-01, -7.8341e-01, 1.5370e+00, -1.1360e+00, -2.5911e-01,
  4277. -1.2756e+00, 6.6877e-01, 1.9165e-01],
  4278. [ 2.9767e-01, -6.7336e-01, 1.2427e+00, -1.1123e+00, -5.4911e-01,
  4279. -1.2077e+00, 3.5533e-01, 1.8474e-01],
  4280. [ 2.9968e-01, -6.2822e-01, 1.5641e+00, -6.2807e-01, -5.8927e-01,
  4281. -5.5374e-01, 3.0163e-01, 2.9457e-01],
  4282. [ 5.1429e-01, -4.5790e-01, 1.7073e+00, 6.4440e-03, -3.9337e-01,
  4283. 4.3100e-04, 3.4778e-01, 2.6891e-01],
  4284. [ 2.0770e-01, -7.3136e-01, 1.6545e+00, -9.7071e-01, -1.7980e-01,
  4285. -1.0599e+00, 7.4962e-01, 2.0022e-01]], device='cuda:0',
  4286. grad_fn=<AddmmBackward>)
  4287. landmarks are: tensor([[[ 0.5680, -0.4562, 1.5697, 0.4947, -0.4904, -0.1503, 0.3536,
  4288. 0.1956],
  4289. [ 0.5603, -0.3764, 0.8088, -1.1466, -0.4557, -1.1158, 0.3642,
  4290. 0.2391],
  4291. [ 0.5491, -0.4132, 1.7557, -0.0919, -0.2709, 0.0313, 0.0635,
  4292. 0.4032],
  4293. [ 0.6471, -0.3840, 1.5767, -1.0311, -0.0457, -1.5007, 0.6889,
  4294. 0.1020],
  4295. [ 0.5762, -0.4153, 1.1908, -1.3622, -0.4190, -1.2471, 0.4368,
  4296. 0.0213],
  4297. [ 0.5867, -0.3937, 1.7499, -0.7155, -0.6404, -0.3844, 0.4739,
  4298. 0.3392],
  4299. [ 0.5303, -0.4440, 1.7095, 0.1390, -0.3402, 0.1159, 0.3353,
  4300. 0.0467],
  4301. [ 0.6405, -0.3671, 1.8249, -1.0080, 0.0178, -0.9618, 1.1422,
  4302. 0.2730]]], device='cuda:0')
  4303. loss_train_step before backward: tensor(0.0426, device='cuda:0', grad_fn=<MseLossBackward>)
  4304. loss_train_step after backward: tensor(0.0426, device='cuda:0', grad_fn=<MseLossBackward>)
  4305. loss_train: 0.04259268939495087
  4306. step: 1
  4307. running loss: 0.04259268939495087
  4308. Train Steps: 1/90 Loss: 0.0426 torch.Size([8, 600, 800])
  4309. torch.Size([8, 8])
  4310. tensor([[0.6157, 0.4102, 0.8513, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890],
  4311. [0.6084, 0.3981, 0.8588, 0.5233, 0.4600, 0.5367, 0.5680, 0.5006],
  4312. [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524],
  4313. [0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750],
  4314. [0.6185, 0.4129, 0.8900, 0.4567, 0.3937, 0.5417, 0.5734, 0.5110],
  4315. [0.6276, 0.4235, 0.8888, 0.5333, 0.3800, 0.3117, 0.5427, 0.6164],
  4316. [0.6138, 0.4054, 0.8750, 0.4750, 0.4363, 0.5017, 0.5086, 0.5822],
  4317. [ nan, nan, 0.7192, 0.2346, 0.4037, 0.2050, 0.5138, 0.5650]],
  4318. device='cuda:0', dtype=torch.float64)
  4319. predictions are: tensor([[ 0.4878, -0.4826, 1.4859, -0.7712, -0.6467, -0.7535, 0.3544, 0.2405],
  4320. [ 0.6440, -0.4132, 1.7307, -0.1418, -0.2576, -0.2243, 0.4596, 0.2307],
  4321. [ 0.8317, -0.2595, 1.6605, -0.2190, -0.5446, -0.2266, 0.4281, 0.2645],
  4322. [-0.1442, -0.9106, 1.3206, -1.2534, -0.4152, -1.3048, 0.3776, 0.2055],
  4323. [ 0.4407, -0.5200, 1.7622, -0.3675, -0.3601, -0.1808, 0.4552, 0.2645],
  4324. [ 0.8585, -0.2477, 1.6806, -0.0893, -0.4960, -0.5336, 0.4715, 0.2489],
  4325. [ 0.4703, -0.4759, 1.6254, -0.3660, -0.3193, -0.2449, 0.3914, 0.2745],
  4326. [-0.6149, -1.2198, 1.1038, -1.5154, -0.3264, -1.5851, 0.3474, 0.1998]],
  4327. device='cuda:0', grad_fn=<AddmmBackward>)
  4328. landmarks are: tensor([[[ 0.5581, -0.3912, 1.6460, -0.5230, -0.6173, -0.5923, 0.0681,
  4329. 0.4348],
  4330. [ 0.5241, -0.4470, 1.6806, 0.1313, -0.1612, 0.1929, 0.3378,
  4331. 0.0261],
  4332. [ 0.5529, -0.3811, 1.7788, -0.0380, -0.5307, -0.2074, 0.0727,
  4333. 0.2657],
  4334. [ 0.5695, -0.3927, 1.3226, -0.9002, -0.4672, -1.1928, 0.1737,
  4335. 0.3700],
  4336. [ 0.5711, -0.3788, 1.8249, -0.1766, -0.4672, 0.2160, 0.3625,
  4337. 0.0742],
  4338. [ 0.6131, -0.3299, 1.8192, 0.1775, -0.5307, -0.8463, 0.2206,
  4339. 0.5612],
  4340. [ 0.5491, -0.4132, 1.7557, -0.0919, -0.2709, 0.0313, 0.0635,
  4341. 0.4032],
  4342. [-2.2859, -2.2859, 1.0361, -1.2021, -0.4210, -1.3390, 0.0871,
  4343. 0.3238]]], device='cuda:0')
  4344. loss_train_step before backward: tensor(0.1098, device='cuda:0', grad_fn=<MseLossBackward>)
  4345. loss_train_step after backward: tensor(0.1098, device='cuda:0', grad_fn=<MseLossBackward>)
  4346. loss_train: 0.15242556482553482
  4347. step: 2
  4348. running loss: 0.07621278241276741
  4349.  
  4350. Train Steps: 2/90 Loss: 0.0762 torch.Size([8, 600, 800])
  4351. torch.Size([8, 8])
  4352. tensor([[0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367],
  4353. [0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412],
  4354. [0.6311, 0.3998, 0.7975, 0.5767, 0.3838, 0.4850, 0.7327, 0.5343],
  4355. [0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411],
  4356. [0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6262, 0.5167],
  4357. [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309],
  4358. [0.6201, 0.3970, 0.8413, 0.4950, 0.4413, 0.5183, 0.6088, 0.5400],
  4359. [0.6115, 0.3998, 0.7063, 0.2383, 0.4037, 0.1950, 0.5320, 0.4993]],
  4360. device='cuda:0', dtype=torch.float64)
  4361. predictions are: tensor([[ 0.2055, -0.6563, 1.3950, -0.9640, -0.4633, -0.9175, 0.3805, 0.2383],
  4362. [ 0.4255, -0.4887, 1.5916, -0.6542, -0.4395, -0.7613, 0.4005, 0.2200],
  4363. [ 0.5741, -0.4211, 1.5171, -0.2034, -0.3918, -0.0850, 0.3576, 0.2748],
  4364. [ 0.2696, -0.5973, 1.3929, -0.9257, -0.4207, -0.9935, 0.3744, 0.2466],
  4365. [ 0.5596, -0.4324, 1.7364, -0.3859, -0.4203, -0.5341, 0.4266, 0.2516],
  4366. [ 0.6552, -0.3283, 1.7068, -0.2231, -0.4293, -0.4643, 0.4657, 0.2142],
  4367. [ 0.3758, -0.5187, 1.6616, -0.1487, -0.2192, 0.0357, 0.3985, 0.2725],
  4368. [ 0.1727, -0.7338, 1.0763, -1.1915, -0.5032, -1.1693, 0.2439, 0.2383]],
  4369. device='cuda:0', grad_fn=<AddmmBackward>)
  4370. landmarks are: tensor([[[ 5.6966e-01, -4.4416e-01, 1.3529e+00, -9.5152e-01, -5.7742e-01,
  4371. -7.8011e-01, 5.2533e-01, 1.9310e-01],
  4372. [ 5.7841e-01, -4.0062e-01, 1.7911e+00, -5.7008e-01, -5.1916e-01,
  4373. -1.0331e+00, 4.1374e-01, 2.1391e-01],
  4374. [ 6.2895e-01, -4.3934e-01, 1.3977e+00, 3.7768e-01, -5.1339e-01,
  4375. -4.5727e-02, 1.0984e+00, 1.8214e-01],
  4376. [ 5.7131e-01, -4.3212e-01, 1.4975e+00, -8.1340e-01, -3.0942e-01,
  4377. -1.3345e+00, 3.7786e-01, 2.1339e-01],
  4378. [ 6.0935e-01, -3.9469e-01, 1.8885e+00, -2.9977e-01, -5.7691e-01,
  4379. -6.7698e-01, 6.0670e-01, 1.0054e-01],
  4380. [ 6.5201e-01, -3.6231e-01, 1.8885e+00, 3.1255e-02, -5.5381e-01,
  4381. -5.3841e-01, 6.9257e-01, 1.6611e-01],
  4382. [ 5.7812e-01, -4.5219e-01, 1.5998e+00, 4.6189e-04, -2.4781e-01,
  4383. 1.0824e-01, 5.2587e-01, 2.0831e-01],
  4384. [ 5.3837e-01, -4.3934e-01, 9.7621e-01, -1.1851e+00, -4.2102e-01,
  4385. -1.3852e+00, 1.7122e-01, 2.0118e-02]]], device='cuda:0')
  4386. loss_train_step before backward: tensor(0.0365, device='cuda:0', grad_fn=<MseLossBackward>)
  4387. loss_train_step after backward: tensor(0.0365, device='cuda:0', grad_fn=<MseLossBackward>)
  4388. loss_train: 0.1889284811913967
  4389. step: 3
  4390. running loss: 0.06297616039713223
  4391. Train Steps: 3/90 Loss: 0.0630 torch.Size([8, 600, 800])
  4392. torch.Size([8, 8])
  4393. tensor([[0.6263, 0.4039, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804],
  4394. [0.6154, 0.4112, 0.7037, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600],
  4395. [0.6114, 0.4018, 0.7213, 0.1967, 0.3763, 0.2700, 0.5875, 0.5533],
  4396. [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717],
  4397. [0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631],
  4398. [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378],
  4399. [0.6147, 0.4107, 0.8137, 0.3333, 0.3750, 0.2683, 0.5006, 0.5412],
  4400. [0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385]],
  4401. device='cuda:0', dtype=torch.float64)
  4402. predictions are: tensor([[ 0.9354, -0.1689, 1.7982, -0.0957, -0.5106, -0.0445, 0.4831, 0.2091],
  4403. [ 0.2110, -0.6863, 1.2094, -1.0299, -0.4531, -1.1977, 0.2558, 0.2339],
  4404. [ 0.4777, -0.5063, 1.2934, -0.9708, -0.5087, -0.9821, 0.3245, 0.2426],
  4405. [ 0.2455, -0.6200, 1.3273, -0.9011, -0.4809, -0.9806, 0.2810, 0.2534],
  4406. [ 0.4969, -0.4011, 1.6330, -0.3698, -0.3892, -0.0085, 0.3747, 0.2756],
  4407. [ 0.8093, -0.2633, 1.9032, 0.2491, -0.2979, 0.3559, 0.4084, 0.2968],
  4408. [ 0.1955, -0.6490, 1.4015, -0.7826, -0.4914, -0.8720, 0.2076, 0.2256],
  4409. [ 0.4164, -0.5204, 1.6887, -0.7912, -0.2706, -1.0754, 0.5404, 0.1997]],
  4410. device='cuda:0', grad_fn=<AddmmBackward>)
  4411. landmarks are: tensor([[[ 0.6072, -0.4206, 1.8711, -0.2536, -0.6115, -0.1304, 0.6812,
  4412. -0.0670],
  4413. [ 0.5565, -0.3864, 0.9647, -1.2159, -0.3286, -1.4391, 0.1852,
  4414. 0.3007],
  4415. [ 0.5380, -0.4302, 1.0455, -1.3775, -0.5480, -1.0388, 0.4277,
  4416. 0.2699],
  4417. [ 0.5746, -0.3882, 1.1436, -1.2005, -0.4903, -1.0157, 0.4393,
  4418. 0.3546],
  4419. [ 0.5027, -0.4289, 1.5478, -0.4229, -0.6346, -0.0303, 0.3223,
  4420. 0.3148],
  4421. [ 0.6421, -0.3864, 1.7961, 0.0543, -0.4383, 0.2237, 1.2007,
  4422. 0.1982],
  4423. [ 0.5532, -0.3888, 1.4727, -0.7463, -0.5538, -1.0465, 0.0265,
  4424. 0.2138],
  4425. [ 0.6174, -0.3936, 1.4586, -1.1709, -0.2420, -1.1389, 0.8296,
  4426. 0.2012]]], device='cuda:0')
  4427. loss_train_step before backward: tensor(0.0430, device='cuda:0', grad_fn=<MseLossBackward>)
  4428. loss_train_step after backward: tensor(0.0430, device='cuda:0', grad_fn=<MseLossBackward>)
  4429. loss_train: 0.23190448060631752
  4430. step: 4
  4431. running loss: 0.05797612015157938
  4432. Train Steps: 4/90 Loss: 0.0580 torch.Size([8, 600, 800])
  4433. torch.Size([8, 8])
  4434. tensor([[0.6161, 0.4076, 0.8900, 0.4667, 0.4125, 0.5917, 0.6262, 0.5367],
  4435. [0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183],
  4436. [0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631],
  4437. [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533],
  4438. [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656],
  4439. [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586],
  4440. [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301],
  4441. [0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131]],
  4442. device='cuda:0', dtype=torch.float64)
  4443. predictions are: tensor([[ 0.7001, -0.2766, 1.6908, -0.1484, -0.3457, -0.0808, 0.4303, 0.2640],
  4444. [ 0.4341, -0.5333, 1.4638, -0.8986, -0.4397, -1.1506, 0.3810, 0.2546],
  4445. [ 0.4200, -0.4447, 1.4699, -0.5509, -0.4799, -0.3041, 0.3056, 0.2709],
  4446. [ 0.5279, -0.4039, 1.6207, -0.2760, -0.3859, -0.1375, 0.2687, 0.2574],
  4447. [ 0.6383, -0.3270, 1.5034, -0.3866, -0.5343, -0.3790, 0.2957, 0.2652],
  4448. [ 0.2794, -0.6039, 1.6532, -1.1591, -0.1907, -1.4350, 0.7113, 0.1612],
  4449. [ 0.4900, -0.4513, 1.6026, -0.0751, -0.3859, -0.1690, 0.2521, 0.2527],
  4450. [ 0.6725, -0.3149, 1.4706, -0.6186, -0.5776, -0.5253, 0.3481, 0.2127]],
  4451. device='cuda:0', grad_fn=<AddmmBackward>)
  4452. landmarks are: tensor([[[ 0.5598, -0.4032, 1.8249, -0.1304, -0.3806, 0.4470, 0.6067,
  4453. 0.1929],
  4454. [ 0.5759, -0.3948, 1.2072, -0.9849, -0.4672, -0.8156, 0.4912,
  4455. 0.5701],
  4456. [ 0.5027, -0.4289, 1.5478, -0.4229, -0.6346, -0.0303, 0.3223,
  4457. 0.3148],
  4458. [ 0.5499, -0.4225, 1.8018, -0.2921, -0.3055, 0.0543, 0.4046,
  4459. 0.2699],
  4460. [ 0.5283, -0.4429, 1.5940, -0.2844, -0.5827, -0.1458, 0.2823,
  4461. 0.3267],
  4462. [ 0.6075, -0.4129, 1.6055, -1.0080, -0.2420, -1.0080, 0.9704,
  4463. 0.2944],
  4464. [ 0.6010, -0.4562, 1.7198, -0.0090, -0.3464, 0.0108, 0.6294,
  4465. 0.1627],
  4466. [ 0.5581, -0.4586, 1.5586, -0.3747, -0.6792, -0.2391, 0.4455,
  4467. 0.0840]]], device='cuda:0')
  4468. loss_train_step before backward: tensor(0.0299, device='cuda:0', grad_fn=<MseLossBackward>)
  4469. loss_train_step after backward: tensor(0.0299, device='cuda:0', grad_fn=<MseLossBackward>)
  4470. loss_train: 0.26177416928112507
  4471. step: 5
  4472. running loss: 0.05235483385622501
  4473. Train Steps: 5/90 Loss: 0.0524 torch.Size([8, 600, 800])
  4474. torch.Size([8, 8])
  4475. tensor([[0.6263, 0.4030, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771],
  4476. [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100],
  4477. [0.6225, 0.4116, 0.8662, 0.3517, 0.3663, 0.3233, 0.5837, 0.5317],
  4478. [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461],
  4479. [0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804],
  4480. [ nan, nan, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819],
  4481. [0.6092, 0.4001, 0.8638, 0.4867, 0.4288, 0.5367, 0.5484, 0.5064],
  4482. [0.6152, 0.4131, 0.6863, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305]],
  4483. device='cuda:0', dtype=torch.float64)
  4484. predictions are: tensor([[ 0.5591, -0.3862, 1.8141, -0.1409, -0.3558, -0.0230, 0.4684, 0.1989],
  4485. [ 0.7180, -0.3285, 1.7612, 0.0262, -0.4315, -0.1326, 0.3433, 0.2333],
  4486. [ 0.6214, -0.3498, 1.5698, -0.8629, -0.4971, -1.0170, 0.4340, 0.2186],
  4487. [ 0.7594, -0.2548, 1.7353, 0.0515, -0.2798, -0.1146, 0.3592, 0.2564],
  4488. [ 0.7539, -0.2979, 1.6140, -0.6767, -0.5285, -0.6289, 0.5362, 0.1763],
  4489. [-0.0267, -0.8053, 1.0535, -1.2585, -0.4485, -1.2836, 0.2414, 0.2479],
  4490. [ 0.6017, -0.3277, 1.7198, -0.1452, -0.3064, -0.0624, 0.3880, 0.2317],
  4491. [ 0.4863, -0.4454, 1.1515, -1.0345, -0.5895, -0.9306, 0.2529, 0.2732]],
  4492. device='cuda:0', grad_fn=<AddmmBackward>)
  4493. landmarks are: tensor([[[ 0.6072, -0.4247, 1.8711, -0.0842, -0.5307, 0.1005, 0.6771,
  4494. -0.0821],
  4495. [ 0.6125, -0.4273, 1.6402, 0.2006, -0.4788, -0.0303, 0.6182,
  4496. 0.0697],
  4497. [ 0.5893, -0.3847, 1.7152, -0.6616, -0.5942, -0.7925, 0.4104,
  4498. 0.1698],
  4499. [ 0.5443, -0.3831, 1.7095, 0.1621, -0.2016, 0.1390, 0.1437,
  4500. 0.2364],
  4501. [ 0.6071, -0.4119, 1.7788, -0.5153, -0.6000, -0.5692, 0.6586,
  4502. -0.0670],
  4503. [-2.2859, -2.2859, 0.9438, -0.9967, -0.4614, -1.1851, 0.2468,
  4504. 0.4019],
  4505. [ 0.5278, -0.4377, 1.7037, -0.0380, -0.3055, 0.1929, 0.2473,
  4506. 0.0532],
  4507. [ 0.5557, -0.3779, 0.8838, -1.1004, -0.6115, -0.7617, 0.3769,
  4508. 0.1644]]], device='cuda:0')
  4509. loss_train_step before backward: tensor(0.1336, device='cuda:0', grad_fn=<MseLossBackward>)
  4510. loss_train_step after backward: tensor(0.1336, device='cuda:0', grad_fn=<MseLossBackward>)
  4511. loss_train: 0.39533588476479053
  4512. step: 6
  4513. running loss: 0.06588931412746508
  4514.  
  4515. Train Steps: 6/90 Loss: 0.0659 torch.Size([8, 600, 800])
  4516. torch.Size([8, 8])
  4517. tensor([[0.6277, 0.4057, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823],
  4518. [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679],
  4519. [0.6213, 0.4131, 0.8438, 0.3550, 0.3513, 0.4400, 0.5716, 0.5123],
  4520. [0.6125, 0.3974, 0.7725, 0.2517, 0.3538, 0.3317, 0.5887, 0.5500],
  4521. [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650],
  4522. [0.6289, 0.4081, 0.8720, 0.3487, 0.3900, 0.3183, 0.6703, 0.5376],
  4523. [0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117],
  4524. [0.6030, 0.3969, 0.7988, 0.3917, 0.3450, 0.3667, 0.5266, 0.4700]],
  4525. device='cuda:0', dtype=torch.float64)
  4526. predictions are: tensor([[ 0.5827, -0.4379, 1.4441, -1.1022, -0.3418, -1.2180, 0.5851, 0.1748],
  4527. [ 0.6353, -0.2988, 1.7405, 0.2265, -0.2744, 0.2779, 0.4264, 0.2445],
  4528. [ 0.6568, -0.3217, 1.5801, -0.5005, -0.5412, -0.4082, 0.3244, 0.2201],
  4529. [ 0.4810, -0.4567, 1.2555, -0.9256, -0.5145, -0.7963, 0.3975, 0.2130],
  4530. [ 0.6530, -0.3305, 1.5795, -0.3256, -0.5694, -0.2815, 0.2549, 0.2598],
  4531. [ 0.6139, -0.3530, 1.6029, -0.5732, -0.3805, -0.8116, 0.4547, 0.2125],
  4532. [ 0.3853, -0.4860, 1.6758, -0.1242, -0.3653, -0.0897, 0.4411, 0.2086],
  4533. [ 0.5693, -0.3776, 1.4477, -0.5859, -0.5408, -0.5223, 0.3565, 0.1967]],
  4534. device='cuda:0', grad_fn=<AddmmBackward>)
  4535. landmarks are: tensor([[[ 0.6132, -0.4122, 1.5478, -1.0619, -0.2709, -1.4314, 0.5500,
  4536. -0.0583],
  4537. [ 0.6257, -0.4249, 1.5998, 0.1236, -0.3806, 0.3084, 0.9887,
  4538. 0.3371],
  4539. [ 0.5841, -0.3778, 1.6113, -0.6462, -0.6635, -0.2536, 0.3542,
  4540. 0.0802],
  4541. [ 0.5430, -0.4503, 1.2822, -1.1235, -0.6520, -0.7540, 0.4335,
  4542. 0.2545],
  4543. [ 0.5757, -0.3917, 1.7095, -0.4768, -0.6346, -0.4229, 0.3931,
  4544. 0.3238],
  4545. [ 0.6191, -0.4008, 1.7420, -0.6753, -0.4845, -0.8156, 0.8101,
  4546. 0.1974],
  4547. [ 0.6262, -0.4461, 1.8480, -0.0534, -0.5827, -0.1227, 0.6587,
  4548. 0.0774],
  4549. [ 0.4992, -0.4525, 1.4035, -0.4768, -0.6924, -0.5923, 0.1465,
  4550. -0.1151]]], device='cuda:0')
  4551. loss_train_step before backward: tensor(0.0207, device='cuda:0', grad_fn=<MseLossBackward>)
  4552. loss_train_step after backward: tensor(0.0207, device='cuda:0', grad_fn=<MseLossBackward>)
  4553. loss_train: 0.4160845186561346
  4554. step: 7
  4555. running loss: 0.05944064552230494
  4556. Train Steps: 7/90 Loss: 0.0594 torch.Size([8, 600, 800])
  4557. torch.Size([8, 8])
  4558. tensor([[0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350],
  4559. [0.6239, 0.4206, 0.8750, 0.5400, 0.3688, 0.4850, 0.5737, 0.5700],
  4560. [0.6261, 0.3987, 0.8688, 0.4917, 0.4300, 0.5333, 0.7010, 0.5309],
  4561. [0.6289, 0.4019, 0.8113, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332],
  4562. [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167],
  4563. [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600],
  4564. [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114],
  4565. [0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272]],
  4566. device='cuda:0', dtype=torch.float64)
  4567. predictions are: tensor([[ 0.7021, -0.3496, 1.2926, -1.1388, -0.4333, -1.1794, 0.4480, 0.1941],
  4568. [ 0.9963, -0.1069, 1.7337, 0.1366, -0.6265, -0.0216, 0.4571, 0.1893],
  4569. [ 0.4696, -0.4188, 1.6972, -0.1467, -0.3254, 0.0139, 0.5288, 0.1844],
  4570. [ 0.4513, -0.4183, 1.4989, -0.2108, -0.4803, -0.1647, 0.3797, 0.1903],
  4571. [ 0.6194, -0.3489, 1.4720, -0.8228, -0.5840, -0.7738, 0.4798, 0.2157],
  4572. [ 0.3232, -0.5147, 1.6214, -0.3481, -0.2429, -0.1752, 0.4896, 0.2023],
  4573. [ 0.4407, -0.4461, 1.6146, -0.1038, -0.3282, -0.0575, 0.4005, 0.1977],
  4574. [ 0.4908, -0.4550, 1.4114, -0.8419, -0.5530, -0.9475, 0.3662, 0.1874]],
  4575. device='cuda:0', grad_fn=<AddmmBackward>)
  4576. landmarks are: tensor([[[ 0.5768, -0.4031, 1.3838, -1.1527, -0.2188, -1.4216, 0.4379,
  4577. 0.1850],
  4578. [ 0.5960, -0.3430, 1.7557, 0.2083, -0.5827, -0.0457, 0.3642,
  4579. 0.3469],
  4580. [ 0.6059, -0.4442, 1.7268, -0.0149, -0.2998, 0.1775, 0.9521,
  4581. 0.1661],
  4582. [ 0.6191, -0.4297, 1.4612, 0.2391, -0.4961, 0.0313, 1.1166,
  4583. 0.1768],
  4584. [ 0.5768, -0.3899, 1.3861, -0.7771, -0.5885, -0.5461, 0.5028,
  4585. 0.5624],
  4586. [ 0.5652, -0.4325, 1.7388, -0.1429, -0.0688, 0.3469, 0.4970,
  4587. 0.3007],
  4588. [ 0.5450, -0.4730, 1.7420, 0.1372, -0.1919, 0.2614, 0.4976,
  4589. 0.0764],
  4590. [ 0.5711, -0.4015, 1.6979, -0.6770, -0.5365, -1.0619, 0.1712,
  4591. 0.1494]]], device='cuda:0')
  4592. loss_train_step before backward: tensor(0.0386, device='cuda:0', grad_fn=<MseLossBackward>)
  4593. loss_train_step after backward: tensor(0.0386, device='cuda:0', grad_fn=<MseLossBackward>)
  4594. loss_train: 0.4546631034463644
  4595. step: 8
  4596. running loss: 0.05683288793079555
  4597. Train Steps: 8/90 Loss: 0.0568 torch.Size([8, 600, 800])
  4598. torch.Size([8, 8])
  4599. tensor([[0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749],
  4600. [0.6179, 0.4118, 0.7278, 0.4237, 0.3588, 0.3400, 0.5675, 0.5917],
  4601. [0.6286, 0.4055, 0.9000, 0.4717, 0.3763, 0.4683, 0.7018, 0.5494],
  4602. [0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633],
  4603. [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482],
  4604. [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758],
  4605. [0.6207, 0.4081, 0.7662, 0.2067, 0.3962, 0.3200, 0.6312, 0.5300],
  4606. [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5837, 0.5500]],
  4607. device='cuda:0', dtype=torch.float64)
  4608. predictions are: tensor([[ 0.2553, -0.5819, 1.1939, -0.8524, -0.5313, -0.6417, 0.3589, 0.1984],
  4609. [ 0.7735, -0.2880, 1.4316, -0.4455, -0.6361, -0.5807, 0.2920, 0.2505],
  4610. [ 0.8888, -0.1807, 1.8123, 0.0572, -0.5599, 0.0372, 0.5111, 0.1406],
  4611. [ 0.5415, -0.4451, 1.6313, -0.9964, -0.1940, -1.0505, 0.8056, 0.1340],
  4612. [ 0.7210, -0.2990, 1.6302, -0.2831, -0.5936, -0.3882, 0.4315, 0.1563],
  4613. [ 0.4772, -0.4362, 1.7029, 0.0759, -0.3797, 0.1898, 0.4535, 0.1590],
  4614. [ 0.4588, -0.4792, 1.2327, -0.9956, -0.4738, -0.8281, 0.4887, 0.1862],
  4615. [ 0.5360, -0.3870, 1.7055, -0.0526, -0.3721, 0.0659, 0.5101, 0.1510]],
  4616. device='cuda:0', grad_fn=<AddmmBackward>)
  4617. landmarks are: tensor([[[ 0.5419, -0.4160, 1.1810, -0.8939, -0.6808, -0.7463, 0.2485,
  4618. 0.3694],
  4619. [ 0.5680, -0.3840, 1.0756, -0.3290, -0.6289, -0.7155, 0.3353,
  4620. 0.4470],
  4621. [ 0.6174, -0.4129, 1.8711, -0.1073, -0.5480, -0.1227, 0.9558,
  4622. 0.2516],
  4623. [ 0.6042, -0.4105, 1.5478, -1.2082, -0.1208, -1.0927, 0.9704,
  4624. 0.3157],
  4625. [ 0.5791, -0.4127, 1.8442, -0.3985, -0.6031, -0.6154, 0.4473,
  4626. 0.2464],
  4627. [ 0.6072, -0.4250, 1.8249, -0.0072, -0.4037, 0.1082, 0.6730,
  4628. -0.0881],
  4629. [ 0.5809, -0.4011, 1.2533, -1.3313, -0.4557, -0.8079, 0.6298,
  4630. 0.1621],
  4631. [ 0.5604, -0.4620, 1.7961, -0.1997, -0.3229, 0.1082, 0.4104,
  4632. 0.2545]]], device='cuda:0')
  4633. loss_train_step before backward: tensor(0.0231, device='cuda:0', grad_fn=<MseLossBackward>)
  4634. loss_train_step after backward: tensor(0.0231, device='cuda:0', grad_fn=<MseLossBackward>)
  4635. loss_train: 0.4777647405862808
  4636. step: 9
  4637. running loss: 0.05308497117625342
  4638. Train Steps: 9/90 Loss: 0.0531 torch.Size([8, 600, 800])
  4639. torch.Size([8, 8])
  4640. tensor([[0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400],
  4641. [0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748],
  4642. [0.6277, 0.4103, 0.8087, 0.5717, 0.4188, 0.4750, 0.5663, 0.6083],
  4643. [0.6190, 0.4135, 0.8000, 0.4883, 0.3566, 0.3647, 0.5613, 0.5900],
  4644. [0.6264, 0.4049, 0.8988, 0.4633, 0.3813, 0.4983, 0.6326, 0.4843],
  4645. [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586],
  4646. [0.6125, 0.4076, 0.8488, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505],
  4647. [ nan, nan, 0.7512, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617]],
  4648. device='cuda:0', dtype=torch.float64)
  4649. predictions are: tensor([[ 0.3739, -0.5290, 1.7073, -0.4004, -0.4418, -0.0480, 0.6693, 0.1278],
  4650. [ 0.5947, -0.3743, 1.6106, -0.0960, -0.4672, -0.2712, 0.5061, 0.1534],
  4651. [ 0.6945, -0.3319, 1.6409, 0.2558, -0.4384, 0.1308, 0.4694, 0.1777],
  4652. [ 0.9772, -0.1057, 1.5562, -0.1561, -0.7040, -0.3083, 0.3698, 0.1888],
  4653. [ 0.6909, -0.3311, 1.8099, -0.0290, -0.5085, 0.1348, 0.5433, 0.1238],
  4654. [ 0.5375, -0.4298, 1.6000, -1.0680, -0.1836, -0.9936, 0.8623, 0.1370],
  4655. [ 0.7472, -0.2899, 1.4580, -0.5038, -0.6770, -0.3817, 0.3422, 0.2079],
  4656. [-0.0050, -0.7495, 1.0129, -1.3194, -0.4006, -1.1948, 0.3865, 0.2216]],
  4657. device='cuda:0', grad_fn=<AddmmBackward>)
  4658. landmarks are: tensor([[[ 0.5778, -0.4393, 1.8018, -0.4614, -0.6693, -0.1381, 0.5490,
  4659. 0.2083],
  4660. [ 0.6421, -0.3647, 1.5940, 0.3084, -0.5192, -0.3691, 1.1057,
  4661. 0.3692],
  4662. [ 0.6134, -0.3910, 1.4497, 0.3546, -0.3517, -0.0919, 0.3296,
  4663. 0.5239],
  4664. [ 0.5732, -0.3761, 1.4092, -0.0303, -0.6388, -0.6012, 0.3065,
  4665. 0.4393],
  4666. [ 0.6075, -0.4159, 1.8654, -0.1458, -0.5249, 0.0159, 0.6359,
  4667. -0.0490],
  4668. [ 0.6075, -0.4129, 1.6055, -1.0080, -0.2420, -1.0080, 0.9704,
  4669. 0.2944],
  4670. [ 0.5433, -0.4032, 1.6344, -0.4922, -0.5769, -0.5846, 0.0357,
  4671. 0.2567],
  4672. [-2.2859, -2.2859, 1.1841, -1.3082, -0.3055, -1.3621, 0.3007,
  4673. 0.3084]]], device='cuda:0')
  4674. loss_train_step before backward: tensor(0.1472, device='cuda:0', grad_fn=<MseLossBackward>)
  4675. loss_train_step after backward: tensor(0.1472, device='cuda:0', grad_fn=<MseLossBackward>)
  4676. loss_train: 0.6249740868806839
  4677. step: 10
  4678. running loss: 0.06249740868806839
  4679.  
  4680. Train Steps: 10/90 Loss: 0.0625 torch.Size([8, 600, 800])
  4681. torch.Size([8, 8])
  4682. tensor([[0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266],
  4683. [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679],
  4684. [0.6202, 0.4066, 0.8746, 0.3376, 0.3717, 0.3090, 0.5842, 0.5165],
  4685. [0.6293, 0.4024, 0.8750, 0.5000, 0.4012, 0.5733, 0.7121, 0.5633],
  4686. [ nan, nan, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733],
  4687. [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150],
  4688. [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391],
  4689. [0.6266, 0.4070, 0.8712, 0.5600, 0.3713, 0.4783, 0.5775, 0.6100]],
  4690. device='cuda:0', dtype=torch.float64)
  4691. predictions are: tensor([[ 0.5993, -0.4172, 1.2918, -0.8599, -0.4895, -0.8689, 0.4674, 0.2569],
  4692. [ 0.6262, -0.3522, 1.7937, 0.2198, -0.4215, 0.3747, 0.6560, 0.1763],
  4693. [ 0.8618, -0.2339, 1.5645, -0.7135, -0.6392, -0.8297, 0.5180, 0.1674],
  4694. [ 0.5942, -0.3806, 1.8856, 0.2418, -0.3690, 0.4265, 0.7306, 0.1557],
  4695. [-0.1523, -0.8777, 1.0009, -1.3971, -0.3676, -1.3578, 0.4247, 0.2480],
  4696. [ 0.6371, -0.3410, 1.8052, 0.0200, -0.3949, 0.2762, 0.6559, 0.1790],
  4697. [ 0.6024, -0.4038, 1.3672, -1.0748, -0.5515, -1.0605, 0.6216, 0.1504],
  4698. [ 0.6503, -0.3460, 1.6743, 0.2830, -0.5657, 0.1409, 0.4786, 0.1767]],
  4699. device='cuda:0', grad_fn=<AddmmBackward>)
  4700. landmarks are: tensor([[[ 0.5746, -0.4021, 1.0801, -1.1312, -0.3229, -1.1081, 0.4803,
  4701. 0.6084],
  4702. [ 0.6257, -0.4249, 1.5998, 0.1236, -0.3806, 0.3084, 0.9887,
  4703. 0.3371],
  4704. [ 0.5787, -0.4077, 1.7537, -0.7263, -0.5692, -0.8586, 0.4126,
  4705. 0.1000],
  4706. [ 0.6207, -0.4273, 1.7557, 0.0236, -0.4326, 0.3623, 1.0033,
  4707. 0.3157],
  4708. [-2.2859, -2.2859, 0.7106, -1.4468, -0.2882, -1.4237, 0.2430,
  4709. 0.3623],
  4710. [ 0.6014, -0.3883, 1.8192, -0.1150, -0.2882, 0.4008, 0.5201,
  4711. 0.0928],
  4712. [ 0.5785, -0.4259, 1.4228, -1.0261, -0.4190, -1.2189, 0.4763,
  4713. 0.2043],
  4714. [ 0.6081, -0.4059, 1.7383, 0.3007, -0.5711, -0.0765, 0.3815,
  4715. 0.5316]]], device='cuda:0')
  4716. loss_train_step before backward: tensor(0.1232, device='cuda:0', grad_fn=<MseLossBackward>)
  4717. loss_train_step after backward: tensor(0.1232, device='cuda:0', grad_fn=<MseLossBackward>)
  4718. loss_train: 0.7481739446520805
  4719. step: 11
  4720. running loss: 0.06801581315018913
  4721. Train Steps: 11/90 Loss: 0.0680 torch.Size([8, 600, 800])
  4722. torch.Size([8, 8])
  4723. tensor([[0.6202, 0.4079, 0.8025, 0.2500, 0.3763, 0.3217, 0.6125, 0.5533],
  4724. [0.6272, 0.4045, 0.8538, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649],
  4725. [0.6157, 0.4102, 0.8513, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890],
  4726. [0.6095, 0.3970, 0.8688, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940],
  4727. [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467],
  4728. [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611],
  4729. [0.6193, 0.4050, 0.7313, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656],
  4730. [0.6090, 0.4010, 0.7838, 0.3483, 0.3538, 0.3783, 0.5462, 0.5077]],
  4731. device='cuda:0', dtype=torch.float64)
  4732. predictions are: tensor([[ 0.6062, -0.3971, 1.4841, -0.8442, -0.4335, -0.7972, 0.6544, 0.2369],
  4733. [ 0.7315, -0.3394, 1.7088, 0.1553, -0.5659, -0.1784, 0.6413, 0.1370],
  4734. [ 0.6299, -0.3791, 1.5284, -0.3892, -0.6555, -0.3467, 0.4660, 0.2540],
  4735. [ 0.4641, -0.4911, 1.8043, -0.0249, -0.2395, -0.0491, 0.7281, 0.1559],
  4736. [ 0.4449, -0.4664, 1.5657, -0.5493, -0.5251, -0.2868, 0.7594, 0.1848],
  4737. [ 0.3801, -0.5408, 1.7443, 0.1158, -0.3322, 0.1869, 0.5623, 0.2204],
  4738. [ 0.4160, -0.5228, 1.1209, -1.0819, -0.5001, -1.0193, 0.5185, 0.2376],
  4739. [ 0.4419, -0.5029, 1.3971, -0.7442, -0.5796, -0.5155, 0.5558, 0.1858]],
  4740. device='cuda:0', grad_fn=<AddmmBackward>)
  4741. landmarks are: tensor([[[ 0.5788, -0.4020, 1.4208, -1.1312, -0.5480, -0.8002, 0.5432,
  4742. 0.2699],
  4743. [ 0.6109, -0.4177, 1.6575, 0.4393, -0.5538, -0.2459, 0.4805,
  4744. -0.1385],
  4745. [ 0.5581, -0.3912, 1.6460, -0.5230, -0.6173, -0.5923, 0.0681,
  4746. 0.4348],
  4747. [ 0.5292, -0.4520, 1.7268, -0.0842, -0.0413, -0.0324, 0.1116,
  4748. -0.0039],
  4749. [ 0.5484, -0.4591, 1.5940, -0.7771, -0.6693, -0.3460, 0.4219,
  4750. 0.2391],
  4751. [ 0.5491, -0.4290, 1.7788, -0.1073, -0.2651, -0.1073, 0.2555,
  4752. 0.3057],
  4753. [ 0.5746, -0.4153, 1.0917, -1.1620, -0.4037, -1.3082, 0.3234,
  4754. 0.3267],
  4755. [ 0.5270, -0.4336, 1.3342, -0.6770, -0.6520, -0.5384, 0.2370,
  4756. 0.0592]]], device='cuda:0')
  4757. loss_train_step before backward: tensor(0.0295, device='cuda:0', grad_fn=<MseLossBackward>)
  4758. loss_train_step after backward: tensor(0.0295, device='cuda:0', grad_fn=<MseLossBackward>)
  4759. loss_train: 0.7777050696313381
  4760. step: 12
  4761. running loss: 0.06480875580261151
  4762. Train Steps: 12/90 Loss: 0.0648 torch.Size([8, 600, 800])
  4763. torch.Size([8, 8])
  4764. tensor([[0.6184, 0.4079, 0.8350, 0.3700, 0.3675, 0.2883, 0.5312, 0.5783],
  4765. [0.6109, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117],
  4766. [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609],
  4767. [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550],
  4768. [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896],
  4769. [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398],
  4770. [0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136],
  4771. [0.6339, 0.4123, 0.8638, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436]],
  4772. device='cuda:0', dtype=torch.float64)
  4773. predictions are: tensor([[ 0.4202, -0.5173, 1.3836, -0.6753, -0.5700, -0.7469, 0.4485, 0.2560],
  4774. [ 0.4411, -0.5239, 1.6127, -0.2479, -0.5575, -0.2546, 0.5313, 0.2155],
  4775. [ 0.3881, -0.5298, 1.6339, -0.9055, -0.2468, -0.8263, 0.8806, 0.2196],
  4776. [ 0.4624, -0.4736, 1.6928, -0.2386, -0.2085, -0.0825, 0.7356, 0.2267],
  4777. [ 0.3695, -0.5581, 1.3107, -0.6324, -0.6252, -0.6646, 0.4025, 0.3258],
  4778. [ 0.6355, -0.4044, 1.6213, -0.0035, -0.5570, -0.2023, 0.6437, 0.1830],
  4779. [ 0.4304, -0.5043, 1.4906, -0.5608, -0.6256, -0.4607, 0.4872, 0.2247],
  4780. [ 0.6903, -0.3622, 1.7952, 0.1668, -0.4360, 0.3059, 0.8035, 0.2034]],
  4781. device='cuda:0', grad_fn=<AddmmBackward>)
  4782. landmarks are: tensor([[[ 0.5704, -0.4019, 1.5709, -0.5769, -0.5885, -0.9541, 0.1679,
  4783. 0.3854],
  4784. [ 0.5356, -0.4343, 1.7441, -0.1150, -0.6462, -0.3614, 0.3238,
  4785. 0.0774],
  4786. [ 0.6026, -0.4032, 1.7326, -0.7771, -0.3691, -0.8617, 0.9704,
  4787. 0.3050],
  4788. [ 0.5351, -0.4321, 1.7326, -0.1381, 0.0463, 0.1049, 0.4162,
  4789. 0.2776],
  4790. [ 0.5485, -0.3997, 1.4445, -0.4895, -0.6000, -0.8309, 0.1878,
  4791. 0.4374],
  4792. [ 0.5786, -0.4463, 1.6655, 0.2216, -0.5115, -0.2675, 0.4236,
  4793. 0.2075],
  4794. [ 0.5235, -0.4273, 1.7499, -0.4306, -0.5827, -0.4614, 0.1651,
  4795. 0.0862],
  4796. [ 0.6421, -0.3816, 1.7037, 0.1929, -0.4037, 0.2391, 1.1861,
  4797. 0.2249]]], device='cuda:0')
  4798. loss_train_step before backward: tensor(0.0215, device='cuda:0', grad_fn=<MseLossBackward>)
  4799. loss_train_step after backward: tensor(0.0215, device='cuda:0', grad_fn=<MseLossBackward>)
  4800. loss_train: 0.7992481663823128
  4801. step: 13
  4802. running loss: 0.06148062818325483
  4803. Train Steps: 13/90 Loss: 0.0615 torch.Size([8, 600, 800])
  4804. torch.Size([8, 8])
  4805. tensor([[0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517],
  4806. [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5513, 0.5750],
  4807. [0.6148, 0.4076, 0.8666, 0.4820, 0.4138, 0.5067, 0.5250, 0.5767],
  4808. [0.6346, 0.4144, 0.9088, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899],
  4809. [0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283],
  4810. [0.6143, 0.4040, 0.8237, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973],
  4811. [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209],
  4812. [ nan, nan, 0.7335, 0.2569, 0.3788, 0.2667, 0.5066, 0.5578]],
  4813. device='cuda:0', dtype=torch.float64)
  4814. predictions are: tensor([[-0.0145, -0.7906, 1.1196, -1.0009, -0.4156, -0.9101, 0.4786, 0.3279],
  4815. [-0.1917, -0.9380, 1.0036, -1.1857, -0.4182, -1.1089, 0.4359, 0.3034],
  4816. [ 0.9721, -0.1997, 1.9877, 0.5234, -0.4360, 0.5324, 0.7571, 0.1728],
  4817. [ 0.9694, -0.2090, 2.0023, 0.2408, -0.5046, 0.1142, 0.8536, 0.1759],
  4818. [ 0.9472, -0.1856, 1.9436, -0.1935, -0.5401, -0.2110, 0.8041, 0.1836],
  4819. [ 0.3610, -0.5303, 1.3561, -0.7856, -0.5148, -0.7644, 0.4557, 0.2544],
  4820. [ 0.7199, -0.3162, 1.7110, -0.1089, -0.3733, -0.3602, 0.6635, 0.2948],
  4821. [-0.4062, -1.0408, 1.1087, -1.0144, -0.4563, -0.8119, 0.3634, 0.2917]],
  4822. device='cuda:0', grad_fn=<AddmmBackward>)
  4823. landmarks are: tensor([[[ 0.5713, -0.3671, 0.8665, -1.0696, -0.3691, -1.2236, 0.3527,
  4824. 0.2622],
  4825. [ 0.5663, -0.4396, 0.8261, -1.1312, -0.2940, -1.3929, 0.2603,
  4826. 0.3700],
  4827. [ 0.5539, -0.4031, 1.7168, -0.0594, -0.3748, 0.0543, 0.1390,
  4828. 0.3777],
  4829. [ 0.6454, -0.3719, 1.9115, -0.1304, -0.5076, -0.2844, 1.0033,
  4830. 0.4386],
  4831. [ 0.6073, -0.4058, 1.8885, -0.4999, -0.5942, -0.4768, 0.6413,
  4832. 0.1544],
  4833. [ 0.5515, -0.4201, 1.5189, -0.7463, -0.4037, -1.3082, 0.0869,
  4834. 0.0111],
  4835. [ 0.6174, -0.3118, 1.6402, -0.2074, -0.1958, -1.0927, 0.2267,
  4836. 0.5822],
  4837. [-2.2859, -2.2859, 1.1020, -1.0994, -0.5365, -1.0542, 0.0542,
  4838. 0.2905]]], device='cuda:0')
  4839. loss_train_step before backward: tensor(0.1652, device='cuda:0', grad_fn=<MseLossBackward>)
  4840.  
  4841. loss_train_step after backward: tensor(0.1652, device='cuda:0', grad_fn=<MseLossBackward>)
  4842. loss_train: 0.964469201862812
  4843. step: 14
  4844. running loss: 0.06889065727591515
  4845. Train Steps: 14/90 Loss: 0.0689 torch.Size([8, 600, 800])
  4846. torch.Size([8, 8])
  4847. tensor([[0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6063, 0.6083],
  4848. [0.6350, 0.4043, 0.8738, 0.5650, 0.3850, 0.4750, 0.6401, 0.4950],
  4849. [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517],
  4850. [0.6300, 0.4133, 0.8538, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413],
  4851. [0.6055, 0.4015, 0.7425, 0.2033, 0.4113, 0.1883, 0.5217, 0.4823],
  4852. [0.6245, 0.4100, 0.7762, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417],
  4853. [0.6239, 0.4206, 0.8750, 0.5400, 0.3688, 0.4850, 0.5737, 0.5700],
  4854. [0.6110, 0.3984, 0.8750, 0.4933, 0.4625, 0.4950, 0.5578, 0.5676]],
  4855. device='cuda:0', dtype=torch.float64)
  4856. predictions are: tensor([[ 0.7869, -0.3033, 1.7974, -0.1969, -0.6868, -0.0957, 0.6042, 0.2605],
  4857. [ 0.8107, -0.2904, 1.8674, 0.4519, -0.6166, 0.1297, 0.6507, 0.2021],
  4858. [-0.2218, -0.9150, 1.0375, -1.0969, -0.4347, -1.0455, 0.3649, 0.3677],
  4859. [ 0.2698, -0.6191, 1.7095, -1.0067, -0.0719, -0.8716, 0.9475, 0.2549],
  4860. [-0.1655, -0.9113, 1.0007, -1.1529, -0.5338, -1.2172, 0.2648, 0.3071],
  4861. [ 0.0616, -0.7328, 1.1565, -1.1089, -0.3360, -1.1049, 0.4130, 0.3369],
  4862. [ 1.0715, -0.1211, 1.9333, 0.5056, -0.6645, 0.2440, 0.6736, 0.2401],
  4863. [ 0.6491, -0.3910, 1.8553, 0.1901, -0.3428, 0.2149, 0.6684, 0.2235]],
  4864. device='cuda:0', grad_fn=<AddmmBackward>)
  4865. landmarks are: tensor([[[ 5.7742e-01, -3.8684e-01, 1.6286e+00, -5.6921e-01, -6.4619e-01,
  4866. -2.7667e-01, 5.1432e-01, 5.2394e-01],
  4867. [ 6.4706e-01, -4.1832e-01, 1.7499e+00, 3.2379e-01, -5.0762e-01,
  4868. -9.1917e-02, 6.7064e-01, 4.6189e-04],
  4869. [ 5.7131e-01, -3.6712e-01, 8.6651e-01, -1.0696e+00, -3.6905e-01,
  4870. -1.2236e+00, 3.5266e-01, 2.6220e-01],
  4871. [ 6.2401e-01, -3.7675e-01, 1.6575e+00, -1.2851e+00, 2.9492e-01,
  4872. -1.2467e+00, 1.1276e+00, 2.1421e-01],
  4873. [ 5.1085e-01, -4.3164e-01, 1.1436e+00, -1.3467e+00, -3.8637e-01,
  4874. -1.4160e+00, 1.2393e-01, -5.8033e-02],
  4875. [ 5.9850e-01, -3.9207e-01, 1.2995e+00, -1.0927e+00, 6.2356e-03,
  4876. -1.5854e+00, 4.2771e-01, 2.1601e-01],
  4877. [ 5.9601e-01, -3.4305e-01, 1.7557e+00, 2.0831e-01, -5.8268e-01,
  4878. -4.5727e-02, 3.6420e-01, 3.4688e-01],
  4879. [ 5.3637e-01, -4.4573e-01, 1.7557e+00, -7.2363e-03, -1.4965e-01,
  4880. 4.6189e-04, 2.9049e-01, 3.3573e-01]]], device='cuda:0')
  4881. loss_train_step before backward: tensor(0.0739, device='cuda:0', grad_fn=<MseLossBackward>)
  4882. loss_train_step after backward: tensor(0.0739, device='cuda:0', grad_fn=<MseLossBackward>)
  4883. loss_train: 1.038370206952095
  4884. step: 15
  4885. running loss: 0.069224680463473
  4886. Train Steps: 15/90 Loss: 0.0692 torch.Size([8, 600, 800])
  4887. torch.Size([8, 8])
  4888. tensor([[0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947],
  4889. [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817],
  4890. [0.6204, 0.4013, 0.8075, 0.2400, 0.4313, 0.2050, 0.5800, 0.5150],
  4891. [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5963, 0.6217],
  4892. [0.6042, 0.3990, 0.6831, 0.2875, 0.3500, 0.3133, 0.5143, 0.5510],
  4893. [0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038],
  4894. [0.6272, 0.4045, 0.8538, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649],
  4895. [ nan, nan, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552]],
  4896. device='cuda:0', dtype=torch.float64)
  4897. predictions are: tensor([[ 0.7944, -0.3463, 1.9906, 0.2476, -0.5201, 0.2336, 0.7704, 0.1918],
  4898. [ 0.0689, -0.7012, 1.3575, -0.7715, -0.3078, -0.7171, 0.4185, 0.3937],
  4899. [ 0.3230, -0.5789, 1.3084, -0.9413, -0.4369, -1.0296, 0.4259, 0.3471],
  4900. [ 0.5474, -0.4719, 1.2206, -0.8174, -0.4809, -0.9417, 0.4338, 0.3615],
  4901. [ 0.1219, -0.7330, 1.1250, -0.9224, -0.6118, -0.8752, 0.3285, 0.3161],
  4902. [ 0.5981, -0.4339, 1.8751, 0.1986, -0.4184, 0.1858, 0.5192, 0.2195],
  4903. [ 0.8358, -0.2984, 1.8211, 0.3606, -0.6419, -0.1551, 0.5716, 0.2108],
  4904. [-0.1780, -0.8768, 1.5876, -1.0652, -0.0232, -0.9913, 0.8239, 0.2795]],
  4905. device='cuda:0', grad_fn=<AddmmBackward>)
  4906. landmarks are: tensor([[[ 5.8031e-01, -4.3672e-01, 1.8423e+00, -3.0331e-02, -5.9423e-01,
  4907. 2.3557e-02, 6.5034e-01, -9.2270e-04],
  4908. [ 5.9636e-01, -3.3795e-01, 1.4785e+00, -8.3865e-01, -2.4203e-01,
  4909. -1.0619e+00, 3.2379e-01, 4.0077e-01],
  4910. [ 5.7962e-01, -4.3256e-01, 1.4439e+00, -1.1774e+00, -2.9400e-01,
  4911. -1.3390e+00, 3.9307e-01, 9.2841e-02],
  4912. [ 5.6871e-01, -4.0878e-01, 1.0397e+00, -1.1466e+00, -3.1132e-01,
  4913. -1.1928e+00, 4.6813e-01, 5.8553e-01],
  4914. [ 5.0491e-01, -4.4280e-01, 8.6919e-01, -9.5814e-01, -6.6928e-01,
  4915. -8.3865e-01, 8.9698e-02, 2.5891e-01],
  4916. [ 5.2494e-01, -4.4734e-01, 1.7326e+00, -9.1917e-02, -2.0162e-01,
  4917. 1.5443e-01, 1.7328e-01, 4.1158e-02],
  4918. [ 6.1085e-01, -4.1771e-01, 1.6575e+00, 4.3926e-01, -5.5381e-01,
  4919. -2.4588e-01, 4.8055e-01, -1.3847e-01],
  4920. [-2.2859e+00, -2.2859e+00, 1.7557e+00, -1.1466e+00, 8.7067e-02,
  4921. -1.0773e+00, 1.1239e+00, 2.7833e-01]]], device='cuda:0')
  4922. loss_train_step before backward: tensor(0.1397, device='cuda:0', grad_fn=<MseLossBackward>)
  4923. loss_train_step after backward: tensor(0.1397, device='cuda:0', grad_fn=<MseLossBackward>)
  4924. loss_train: 1.1780213713645935
  4925. step: 16
  4926. running loss: 0.0736263357102871
  4927. Train Steps: 16/90 Loss: 0.0736 torch.Size([8, 600, 800])
  4928. torch.Size([8, 8])
  4929. tensor([[0.6147, 0.4107, 0.8137, 0.3333, 0.3750, 0.2683, 0.5006, 0.5412],
  4930. [0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507],
  4931. [ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600],
  4932. [0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869],
  4933. [ nan, nan, 0.6688, 0.2513, 0.4113, 0.2117, 0.5193, 0.5933],
  4934. [0.6205, 0.4016, 0.8350, 0.2717, 0.3987, 0.2550, 0.5787, 0.5133],
  4935. [0.6184, 0.4079, 0.8350, 0.3700, 0.3675, 0.2883, 0.5312, 0.5783],
  4936. [0.6078, 0.4033, 0.8019, 0.3055, 0.3450, 0.4200, 0.6025, 0.5550]],
  4937. device='cuda:0', dtype=torch.float64)
  4938. predictions are: tensor([[ 0.3631, -0.5394, 1.4618, -0.3937, -0.5298, -0.5855, 0.3389, 0.2794],
  4939. [ 0.4413, -0.5507, 1.2269, -0.7110, -0.4430, -0.9009, 0.4244, 0.2799],
  4940. [-0.3335, -0.9625, 1.3309, -0.7680, -0.2863, -0.7579, 0.4095, 0.3167],
  4941. [ 0.4255, -0.4960, 1.6453, -0.6700, -0.1697, -0.6607, 0.6740, 0.2752],
  4942. [-0.3251, -0.9842, 1.0067, -0.9940, -0.3619, -1.0681, 0.3570, 0.3390],
  4943. [ 0.5777, -0.4170, 1.6715, -0.5359, -0.3518, -0.6406, 0.6042, 0.2708],
  4944. [ 0.6887, -0.3486, 1.7259, -0.0297, -0.5322, -0.4222, 0.4260, 0.2669],
  4945. [ 0.8123, -0.2894, 1.7652, -0.2682, -0.5185, -0.1183, 0.7801, 0.2465]],
  4946. device='cuda:0', grad_fn=<AddmmBackward>)
  4947. landmarks are: tensor([[[ 0.5532, -0.3888, 1.4727, -0.7463, -0.5538, -1.0465, 0.0265,
  4948. 0.2138],
  4949. [ 0.5185, -0.4252, 0.9647, -1.2928, -0.4788, -1.2390, 0.2617,
  4950. 0.2576],
  4951. [-2.2859, -2.2859, 1.4266, -1.1389, -0.4499, -1.1235, 0.2891,
  4952. 0.3007],
  4953. [ 0.6127, -0.3944, 1.5189, -1.2467, -0.1323, -1.4622, 0.5646,
  4954. -0.0369],
  4955. [-2.2859, -2.2859, 0.8033, -1.1250, -0.3864, -1.3082, 0.1126,
  4956. 0.4543],
  4957. [ 0.5800, -0.4312, 1.5709, -1.0311, -0.4441, -1.1081, 0.3873,
  4958. 0.0851],
  4959. [ 0.5704, -0.4019, 1.5709, -0.5769, -0.5885, -0.9541, 0.1679,
  4960. 0.3854],
  4961. [ 0.5215, -0.4232, 1.4182, -0.8747, -0.6924, -0.3460, 0.4970,
  4962. 0.2776]]], device='cuda:0')
  4963. loss_train_step before backward: tensor(0.2478, device='cuda:0', grad_fn=<MseLossBackward>)
  4964. loss_train_step after backward: tensor(0.2478, device='cuda:0', grad_fn=<MseLossBackward>)
  4965. loss_train: 1.425776869058609
  4966. step: 17
  4967. running loss: 0.08386922759168289
  4968. Train Steps: 17/90 Loss: 0.0839 torch.Size([8, 600, 800])
  4969. torch.Size([8, 8])
  4970. tensor([[0.6164, 0.4102, 0.8850, 0.4867, 0.4213, 0.5983, 0.5609, 0.5038],
  4971. [0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400],
  4972. [0.6091, 0.3997, 0.8314, 0.4334, 0.3788, 0.4550, 0.5213, 0.5656],
  4973. [0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407],
  4974. [0.6128, 0.4115, 0.7934, 0.3778, 0.3450, 0.4033, 0.5337, 0.5456],
  4975. [ nan, nan, 0.7525, 0.2291, 0.3838, 0.3017, 0.6050, 0.5667],
  4976. [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334],
  4977. [ nan, nan, 0.7097, 0.2346, 0.4250, 0.1850, 0.5175, 0.5583]],
  4978. device='cuda:0', dtype=torch.float64)
  4979. predictions are: tensor([[ 0.7392, -0.3159, 1.9126, 0.2965, -0.2309, 0.2665, 0.6292, 0.2335],
  4980. [ 0.6183, -0.4341, 1.8354, -0.2402, -0.4850, -0.2214, 0.6888, 0.2186],
  4981. [ 0.7088, -0.3281, 1.7322, -0.1589, -0.4614, -0.2889, 0.5233, 0.2228],
  4982. [ 0.3687, -0.5831, 1.6318, -0.3442, -0.5248, -0.8639, 0.3086, 0.3081],
  4983. [ 0.6640, -0.3452, 1.5032, -0.5834, -0.5535, -0.6198, 0.4556, 0.2848],
  4984. [-0.3263, -0.9667, 1.2706, -1.0857, -0.2673, -1.1992, 0.3529, 0.3348],
  4985. [ 0.3817, -0.5648, 1.2980, -0.9832, -0.4549, -1.1130, 0.4572, 0.2740],
  4986. [-0.7312, -1.2464, 1.0082, -1.2988, -0.2090, -1.4575, 0.2481, 0.3142]],
  4987. device='cuda:0', grad_fn=<AddmmBackward>)
  4988. landmarks are: tensor([[[ 0.5614, -0.3912, 1.8018, -0.0380, -0.3402, 0.4778, 0.3049,
  4989. 0.0412],
  4990. [ 0.5778, -0.4393, 1.8018, -0.4614, -0.6693, -0.1381, 0.5490,
  4991. 0.2083],
  4992. [ 0.5275, -0.4396, 1.5543, -0.2841, -0.5365, -0.1843, 0.1221,
  4993. 0.3265],
  4994. [ 0.5470, -0.3952, 1.6377, -0.4253, -0.6289, -0.8079, 0.0249,
  4995. 0.2116],
  4996. [ 0.5444, -0.3852, 1.3786, -0.5409, -0.6924, -0.4229, 0.1791,
  4997. 0.2341],
  4998. [-2.2859, -2.2859, 1.1898, -1.2278, -0.5134, -0.8925, 0.5085,
  4999. 0.3315],
  5000. [ 0.5730, -0.4562, 1.2195, -1.2440, -0.5497, -0.7711, 0.5704,
  5001. 0.1779],
  5002. [-2.2859, -2.2859, 0.9922, -1.2021, -0.3229, -1.4314, 0.1044,
  5003. 0.2930]]], device='cuda:0')
  5004. loss_train_step before backward: tensor(0.1672, device='cuda:0', grad_fn=<MseLossBackward>)
  5005. loss_train_step after backward: tensor(0.1672, device='cuda:0', grad_fn=<MseLossBackward>)
  5006. loss_train: 1.5929645597934723
  5007. step: 18
  5008. running loss: 0.08849803109963734
  5009.  
  5010. Train Steps: 18/90 Loss: 0.0885 torch.Size([8, 600, 800])
  5011. torch.Size([8, 8])
  5012. tensor([[0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879],
  5013. [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450],
  5014. [0.6109, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567],
  5015. [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285],
  5016. [0.6097, 0.4000, 0.7325, 0.2667, 0.3450, 0.3517, 0.5284, 0.5045],
  5017. [0.6250, 0.4146, 0.8838, 0.3933, 0.3588, 0.4283, 0.6162, 0.5367],
  5018. [0.6275, 0.4050, 0.9038, 0.3767, 0.3838, 0.3533, 0.7074, 0.5575],
  5019. [0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350]],
  5020. device='cuda:0', dtype=torch.float64)
  5021. predictions are: tensor([[ 0.7018, -0.3557, 1.7540, -0.1224, -0.4276, -0.0423, 0.5841, 0.2146],
  5022. [ 0.2970, -0.6503, 1.6967, -0.1300, -0.4445, -0.3633, 0.4110, 0.2402],
  5023. [ 0.0251, -0.7716, 1.1079, -1.0232, -0.4226, -1.2824, 0.2213, 0.3022],
  5024. [ 0.3834, -0.5366, 1.4415, -0.5750, -0.4902, -0.5628, 0.3763, 0.2302],
  5025. [-0.1063, -0.8539, 1.1570, -1.0314, -0.4734, -1.0130, 0.2999, 0.2568],
  5026. [ 0.0538, -0.7577, 1.6569, -0.3731, -0.4748, -0.4974, 0.3697, 0.2801],
  5027. [ 0.3470, -0.5624, 1.7853, -0.5192, -0.2951, -0.8880, 0.6386, 0.2389],
  5028. [ 0.1852, -0.6856, 1.3280, -1.0545, -0.2309, -1.3210, 0.3448, 0.3081]],
  5029. device='cuda:0', grad_fn=<AddmmBackward>)
  5030. landmarks are: tensor([[[ 0.5676, -0.4327, 1.8252, -0.2095, -0.4886, 0.3084, 0.4727,
  5031. -0.0322],
  5032. [ 0.5785, -0.4252, 1.7676, -0.1602, -0.5845, -0.3446, 0.4566,
  5033. 0.2314],
  5034. [ 0.5356, -0.4217, 1.0339, -1.4776, -0.5076, -1.1081, 0.4219,
  5035. 0.2853],
  5036. [ 0.5253, -0.4296, 1.3084, -0.3098, -0.6115, -0.2767, 0.0928,
  5037. 0.1552],
  5038. [ 0.5303, -0.4384, 1.0975, -1.0542, -0.6924, -0.6616, 0.1548,
  5039. 0.0442],
  5040. [ 0.6009, -0.3710, 1.7961, -0.4691, -0.6289, -0.3075, 0.5605,
  5041. 0.1929],
  5042. [ 0.6125, -0.4153, 1.8885, -0.5461, -0.5134, -0.6539, 0.9814,
  5043. 0.2890],
  5044. [ 0.5768, -0.4031, 1.3838, -1.1527, -0.2188, -1.4216, 0.4379,
  5045. 0.1850]]], device='cuda:0')
  5046. loss_train_step before backward: tensor(0.0528, device='cuda:0', grad_fn=<MseLossBackward>)
  5047. loss_train_step after backward: tensor(0.0528, device='cuda:0', grad_fn=<MseLossBackward>)
  5048. loss_train: 1.6457261517643929
  5049. step: 19
  5050. running loss: 0.08661716588233646
  5051. Train Steps: 19/90 Loss: 0.0866 torch.Size([8, 600, 800])
  5052. torch.Size([8, 8])
  5053. tensor([[0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633],
  5054. [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611],
  5055. [ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600],
  5056. [0.6113, 0.4088, 0.6859, 0.2208, 0.4363, 0.1700, 0.5188, 0.5533],
  5057. [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466],
  5058. [0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317],
  5059. [0.6317, 0.4038, 0.8287, 0.5900, 0.3800, 0.4717, 0.6295, 0.4986],
  5060. [0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433]],
  5061. device='cuda:0', dtype=torch.float64)
  5062. predictions are: tensor([[ 0.7591, -0.3409, 1.7338, -0.0023, -0.5642, -0.2221, 0.4435, 0.2436],
  5063. [ 0.3795, -0.5712, 1.7574, -0.1573, -0.3790, -0.1880, 0.3507, 0.2420],
  5064. [-0.8353, -1.2972, 1.2454, -1.2587, -0.2312, -1.3055, 0.2796, 0.2779],
  5065. [-0.5438, -1.1473, 1.0012, -1.3929, -0.3499, -1.5850, 0.1994, 0.2789],
  5066. [ 0.5224, -0.4489, 1.7406, -0.0551, -0.3338, -0.1273, 0.4229, 0.1991],
  5067. [-0.5449, -1.1069, 1.2491, -1.3066, -0.2089, -1.4304, 0.3174, 0.2808],
  5068. [ 0.9422, -0.2005, 1.7710, 0.0972, -0.6246, -0.3440, 0.4153, 0.1871],
  5069. [ 0.5999, -0.4022, 1.3268, -1.1283, -0.6029, -0.9938, 0.4009, 0.2491]],
  5070. device='cuda:0', grad_fn=<AddmmBackward>)
  5071. landmarks are: tensor([[[ 0.5879, -0.3591, 1.8018, 0.1236, -0.5596, -0.1612, 0.3469,
  5072. 0.3161],
  5073. [ 0.5491, -0.4290, 1.7788, -0.1073, -0.2651, -0.1073, 0.2555,
  5074. 0.3057],
  5075. [-2.2859, -2.2859, 1.4266, -1.1389, -0.4499, -1.1235, 0.2891,
  5076. 0.3007],
  5077. [ 0.5377, -0.3978, 0.8824, -1.2663, -0.2709, -1.5007, 0.1102,
  5078. 0.2699],
  5079. [ 0.5178, -0.4480, 1.6748, 0.0620, -0.2767, 0.2083, 0.1067,
  5080. 0.2386],
  5081. [ 0.5849, -0.3836, 1.2649, -1.2236, -0.3460, -1.2313, 0.4508,
  5082. 0.1698],
  5083. [ 0.6320, -0.4206, 1.5420, 0.4393, -0.5307, -0.1073, 0.6216,
  5084. 0.0171],
  5085. [ 0.5665, -0.3963, 1.3457, -0.9926, -0.6058, -0.6308, 0.5721,
  5086. 0.2237]]], device='cuda:0')
  5087. loss_train_step before backward: tensor(0.1236, device='cuda:0', grad_fn=<MseLossBackward>)
  5088. loss_train_step after backward: tensor(0.1236, device='cuda:0', grad_fn=<MseLossBackward>)
  5089. loss_train: 1.7693058922886848
  5090. step: 20
  5091. running loss: 0.08846529461443424
  5092. Train Steps: 20/90 Loss: 0.0885 torch.Size([8, 600, 800])
  5093. torch.Size([8, 8])
  5094. tensor([[0.6131, 0.4064, 0.8638, 0.5200, 0.4788, 0.4783, 0.5258, 0.5867],
  5095. [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938],
  5096. [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6188, 0.5283],
  5097. [0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133],
  5098. [0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162],
  5099. [0.6212, 0.4171, 0.7875, 0.3633, 0.3813, 0.2933, 0.5675, 0.5700],
  5100. [0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320],
  5101. [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837]],
  5102. device='cuda:0', dtype=torch.float64)
  5103. predictions are: tensor([[ 0.1886, -0.6836, 1.5729, -0.2674, -0.2233, -0.4119, 0.3165, 0.2492],
  5104. [ 0.4372, -0.5134, 1.6379, -0.3847, -0.5800, -0.9076, 0.2784, 0.1690],
  5105. [-0.0067, -0.8098, 1.6752, -0.9982, -0.3071, -1.3077, 0.4903, 0.2232],
  5106. [ 0.3528, -0.5931, 1.6171, -0.1919, -0.5364, -0.4231, 0.2934, 0.1907],
  5107. [ 0.3657, -0.5674, 1.6048, -0.6359, -0.4923, -0.3539, 0.4614, 0.1826],
  5108. [ 0.1345, -0.6927, 1.1944, -1.0451, -0.4435, -0.9648, 0.2830, 0.3123],
  5109. [ 0.3865, -0.5515, 1.5863, -0.5621, -0.4879, -0.5619, 0.4382, 0.1526],
  5110. [-0.4958, -1.1116, 0.9192, -1.2360, -0.4738, -1.2510, 0.0091, 0.3017]],
  5111. device='cuda:0', grad_fn=<AddmmBackward>)
  5112. landmarks are: tensor([[[ 0.5462, -0.4090, 1.7037, 0.1159, -0.0746, -0.0765, 0.1427,
  5113. 0.4239],
  5114. [ 0.6273, -0.4105, 1.8480, 0.1082, -0.5538, -0.5076, 0.6414,
  5115. -0.0049],
  5116. [ 0.6062, -0.3778, 1.8423, -0.6462, -0.4383, -1.0465, 0.5721,
  5117. 0.1544],
  5118. [ 0.6148, -0.4130, 1.6864, 0.1698, -0.5307, -0.1150, 0.6125,
  5119. 0.0851],
  5120. [ 0.6059, -0.4442, 1.8921, -0.3422, -0.6230, -0.1461, 0.7489,
  5121. 0.0985],
  5122. [ 0.5832, -0.3593, 1.3515, -0.6077, -0.5249, -0.9310, 0.3353,
  5123. 0.3469],
  5124. [ 0.6174, -0.4490, 1.8885, -0.0996, -0.4845, -0.3691, 0.9814,
  5125. 0.1715],
  5126. [ 0.5629, -0.3871, 0.7243, -0.9581, -0.5827, -0.9849, 0.1288,
  5127. 0.4103]]], device='cuda:0')
  5128. loss_train_step before backward: tensor(0.0934, device='cuda:0', grad_fn=<MseLossBackward>)
  5129. loss_train_step after backward: tensor(0.0934, device='cuda:0', grad_fn=<MseLossBackward>)
  5130. loss_train: 1.8627125546336174
  5131. step: 21
  5132. running loss: 0.08870059783969607
  5133. Train Steps: 21/90 Loss: 0.0887 torch.Size([8, 600, 800])
  5134. torch.Size([8, 8])
  5135. tensor([[0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320],
  5136. [ nan, nan, 0.7625, 0.2433, 0.3713, 0.2867, 0.5235, 0.5220],
  5137. [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837],
  5138. [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389],
  5139. [0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750],
  5140. [0.6201, 0.3970, 0.8413, 0.4950, 0.4413, 0.5183, 0.6088, 0.5400],
  5141. [ nan, nan, 0.8850, 0.3000, 0.5363, 0.2250, 0.7343, 0.5771],
  5142. [0.6304, 0.4024, 0.8925, 0.4800, 0.3937, 0.4817, 0.7485, 0.5297]],
  5143. device='cuda:0', dtype=torch.float64)
  5144. predictions are: tensor([[ 0.8606, -0.2453, 1.3886, -0.9505, -0.4922, -1.0851, 0.4113, 0.2025],
  5145. [-0.9464, -1.3543, 1.1559, -1.1529, -0.4029, -1.2134, 0.1268, 0.2243],
  5146. [ 0.1395, -0.6876, 1.0224, -0.8792, -0.6390, -0.9361, -0.0147, 0.2747],
  5147. [ 0.2228, -0.6074, 1.3296, -0.9617, -0.5155, -1.1569, 0.2396, 0.2520],
  5148. [ 0.0331, -0.7152, 1.2906, -0.9207, -0.5002, -1.0947, 0.1238, 0.2317],
  5149. [ 0.5004, -0.4776, 1.7450, 0.0990, -0.4102, 0.1276, 0.3315, 0.1553],
  5150. [-0.5480, -1.0970, 1.6360, -1.0426, 0.0103, -1.0901, 0.6953, 0.2123],
  5151. [ 0.7488, -0.3436, 1.7506, 0.0133, -0.5513, 0.1343, 0.4193, 0.1387]],
  5152. device='cuda:0', grad_fn=<AddmmBackward>)
  5153. landmarks are: tensor([[[ 6.1413e-01, -4.1527e-01, 1.4208e+00, -1.2697e+00, -2.9400e-01,
  5154. -1.0234e+00, 8.6439e-01, 1.7146e-01],
  5155. [-2.2859e+00, -2.2859e+00, 1.2360e+00, -1.1620e+00, -5.7113e-01,
  5156. -9.6182e-01, 1.3215e-01, 1.2532e-01],
  5157. [ 5.6293e-01, -3.8707e-01, 7.2426e-01, -9.5814e-01, -5.8268e-01,
  5158. -9.8491e-01, 1.2881e-01, 4.1034e-01],
  5159. [ 5.6966e-01, -4.4656e-01, 1.1973e+00, -1.1871e+00, -4.5712e-01,
  5160. -9.9653e-01, 5.2186e-01, 2.0324e-01],
  5161. [ 5.6951e-01, -3.9269e-01, 1.3226e+00, -9.0023e-01, -4.6721e-01,
  5162. -1.1928e+00, 1.7367e-01, 3.6998e-01],
  5163. [ 5.7812e-01, -4.5219e-01, 1.5998e+00, 4.6189e-04, -2.4781e-01,
  5164. 1.0824e-01, 5.2587e-01, 2.0831e-01],
  5165. [-2.2859e+00, -2.2859e+00, 1.8018e+00, -9.0023e-01, 1.9099e-01,
  5166. -1.2467e+00, 1.1057e+00, 3.7986e-01],
  5167. [ 6.2566e-01, -4.2731e-01, 1.8365e+00, -6.8822e-02, -4.6721e-01,
  5168. -6.1124e-02, 1.1715e+00, 1.6077e-01]]], device='cuda:0')
  5169. loss_train_step before backward: tensor(0.1541, device='cuda:0', grad_fn=<MseLossBackward>)
  5170.  
  5171. loss_train_step after backward: tensor(0.1541, device='cuda:0', grad_fn=<MseLossBackward>)
  5172. loss_train: 2.016780845820904
  5173. step: 22
  5174. running loss: 0.0916718566282229
  5175. Train Steps: 22/90 Loss: 0.0917 torch.Size([8, 600, 800])
  5176. torch.Size([8, 8])
  5177. tensor([[0.6263, 0.4065, 0.9038, 0.4317, 0.3588, 0.4550, 0.6325, 0.5250],
  5178. [0.6223, 0.4028, 0.8988, 0.4200, 0.3763, 0.5733, 0.6375, 0.5167],
  5179. [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758],
  5180. [0.6251, 0.4163, 0.8662, 0.4467, 0.3625, 0.3567, 0.6038, 0.5533],
  5181. [0.6200, 0.4070, 0.8938, 0.4183, 0.3538, 0.4567, 0.6175, 0.5400],
  5182. [0.6213, 0.4131, 0.8438, 0.3550, 0.3513, 0.4400, 0.5716, 0.5123],
  5183. [0.6161, 0.4024, 0.8662, 0.4683, 0.4935, 0.5364, 0.6063, 0.5567],
  5184. [0.6156, 0.4125, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084]],
  5185. device='cuda:0', dtype=torch.float64)
  5186. predictions are: tensor([[-0.0417, -0.8324, 1.5706, -0.6496, -0.5061, -0.7902, 0.3200, 0.1959],
  5187. [ 0.6402, -0.3714, 1.5161, -0.6605, -0.4793, -0.3589, 0.4657, 0.1758],
  5188. [ 0.0140, -0.7716, 1.4618, -0.3929, -0.4085, -0.4324, 0.2697, 0.1847],
  5189. [-0.0728, -0.8222, 1.4052, -0.7861, -0.5726, -1.2827, 0.1695, 0.2282],
  5190. [ 0.6133, -0.3940, 1.4564, -0.7158, -0.5758, -0.6346, 0.4125, 0.1667],
  5191. [-0.1707, -0.8710, 1.3833, -0.9728, -0.5769, -1.0311, 0.2213, 0.2043],
  5192. [ 0.1520, -0.6904, 1.5596, -0.5260, -0.2024, -0.6276, 0.3844, 0.1957],
  5193. [ 0.0610, -0.7315, 1.5109, -0.3846, -0.2611, -0.3896, 0.2713, 0.2347]],
  5194. device='cuda:0', grad_fn=<AddmmBackward>)
  5195. landmarks are: tensor([[[ 0.6070, -0.4085, 1.8885, -0.2921, -0.6289, -0.1843, 0.6356,
  5196. 0.1390],
  5197. [ 0.5886, -0.4253, 1.8654, -0.3460, -0.5480, 0.3623, 0.6587,
  5198. 0.1005],
  5199. [ 0.6072, -0.4250, 1.8249, -0.0072, -0.4037, 0.1082, 0.6730,
  5200. -0.0881],
  5201. [ 0.6016, -0.3633, 1.7152, -0.2228, -0.6115, -0.6385, 0.5028,
  5202. 0.2699],
  5203. [ 0.5778, -0.4061, 1.8423, -0.3537, -0.6520, -0.1766, 0.5663,
  5204. 0.2083],
  5205. [ 0.5841, -0.3778, 1.6113, -0.6462, -0.6635, -0.2536, 0.3542,
  5206. 0.0802],
  5207. [ 0.5598, -0.4273, 1.7152, -0.1227, -0.0065, 0.1917, 0.5143,
  5208. 0.2853],
  5209. [ 0.5573, -0.3808, 1.8018, -0.0534, -0.1843, 0.4008, 0.3769,
  5210. 0.0622]]], device='cuda:0')
  5211. loss_train_step before backward: tensor(0.1489, device='cuda:0', grad_fn=<MseLossBackward>)
  5212. loss_train_step after backward: tensor(0.1489, device='cuda:0', grad_fn=<MseLossBackward>)
  5213. loss_train: 2.1656840667128563
  5214. step: 23
  5215. running loss: 0.09416017681360245
  5216. Train Steps: 23/90 Loss: 0.0942 torch.Size([8, 600, 800])
  5217. torch.Size([8, 8])
  5218. tensor([[0.6202, 0.4079, 0.8025, 0.2500, 0.3763, 0.3217, 0.6125, 0.5533],
  5219. [0.6206, 0.4001, 0.8900, 0.3933, 0.3588, 0.3567, 0.5837, 0.5083],
  5220. [0.6142, 0.3982, 0.8650, 0.4883, 0.3912, 0.4317, 0.5315, 0.5350],
  5221. [0.6277, 0.4118, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938],
  5222. [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650],
  5223. [0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733],
  5224. [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150],
  5225. [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220]],
  5226. device='cuda:0', dtype=torch.float64)
  5227. predictions are: tensor([[ 0.0037, -0.7422, 1.3922, -1.1547, -0.4432, -1.1044, 0.3581, 0.2139],
  5228. [ 0.3885, -0.5133, 1.5515, -0.7567, -0.6069, -0.8293, 0.3334, 0.1393],
  5229. [ 0.2714, -0.6226, 1.5515, -0.2874, -0.5171, -0.2793, 0.2606, 0.1226],
  5230. [-0.1505, -0.8803, 1.5943, -0.8312, -0.3734, -1.0537, 0.4488, 0.1856],
  5231. [ 0.5403, -0.4329, 1.4621, -0.1111, -0.4986, -0.3244, 0.2106, 0.2074],
  5232. [ 0.0936, -0.6823, 1.5448, -0.2891, -0.5000, -0.3389, 0.3101, 0.2221],
  5233. [ 0.5081, -0.4559, 1.6541, -0.3323, -0.2829, 0.0522, 0.4847, 0.1648],
  5234. [-0.2603, -0.9413, 0.9716, -1.3636, -0.4094, -1.4996, 0.1687, 0.2274]],
  5235. device='cuda:0', grad_fn=<AddmmBackward>)
  5236. landmarks are: tensor([[[ 0.5788, -0.4020, 1.4208, -1.1312, -0.5480, -0.8002, 0.5432,
  5237. 0.2699],
  5238. [ 0.5807, -0.4378, 1.8249, -0.4691, -0.6289, -0.6385, 0.4104,
  5239. 0.0620],
  5240. [ 0.5513, -0.4467, 1.7095, -0.0303, -0.4788, -0.2921, 0.1692,
  5241. 0.1854],
  5242. [ 0.6135, -0.3841, 1.8654, -0.5153, -0.4614, -1.0619, 0.6195,
  5243. -0.0049],
  5244. [ 0.5770, -0.4036, 1.7095, 0.3084, -0.3691, -0.2690, 0.2314,
  5245. 0.3238],
  5246. [ 0.6121, -0.3138, 1.8423, 0.0082, -0.6462, -0.3075, 0.3469,
  5247. 0.3623],
  5248. [ 0.6014, -0.3883, 1.8192, -0.1150, -0.2882, 0.4008, 0.5201,
  5249. 0.0928],
  5250. [ 0.5566, -0.4160, 0.9935, -1.3313, -0.2824, -1.5161, 0.2144,
  5251. 0.1253]]], device='cuda:0')
  5252. loss_train_step before backward: tensor(0.0659, device='cuda:0', grad_fn=<MseLossBackward>)
  5253. loss_train_step after backward: tensor(0.0659, device='cuda:0', grad_fn=<MseLossBackward>)
  5254. loss_train: 2.2315543591976166
  5255. step: 24
  5256. running loss: 0.09298143163323402
  5257. Train Steps: 24/90 Loss: 0.0930 torch.Size([8, 600, 800])
  5258. torch.Size([8, 8])
  5259. tensor([[0.6084, 0.3981, 0.8588, 0.5233, 0.4600, 0.5367, 0.5680, 0.5006],
  5260. [0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183],
  5261. [0.6212, 0.4171, 0.7875, 0.3633, 0.3813, 0.2933, 0.5675, 0.5700],
  5262. [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333],
  5263. [0.6257, 0.4167, 0.8775, 0.3433, 0.3563, 0.4133, 0.6200, 0.5667],
  5264. [0.6127, 0.4066, 0.8550, 0.5567, 0.4662, 0.5141, 0.5070, 0.5412],
  5265. [0.6131, 0.4064, 0.8638, 0.5200, 0.4788, 0.4783, 0.5258, 0.5867],
  5266. [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320]],
  5267. device='cuda:0', dtype=torch.float64)
  5268. predictions are: tensor([[ 0.2694, -0.6120, 1.6079, -0.1341, -0.3535, -0.1856, 0.2777, 0.1239],
  5269. [-0.1129, -0.8312, 1.4104, -1.0356, -0.4240, -1.0812, 0.3827, 0.2333],
  5270. [ 0.3946, -0.4807, 1.1978, -0.9625, -0.5292, -0.8474, 0.3154, 0.2285],
  5271. [ 0.2245, -0.6238, 1.6883, -0.4121, -0.5650, -0.2460, 0.3334, 0.1264],
  5272. [-0.0742, -0.8103, 1.6389, -0.7168, -0.5823, -0.6310, 0.3868, 0.1623],
  5273. [ 0.3275, -0.5708, 1.5419, -0.0796, -0.3344, -0.2834, 0.2763, 0.1428],
  5274. [ 0.3363, -0.5500, 1.5343, -0.2309, -0.3338, -0.3383, 0.3287, 0.1582],
  5275. [ 0.4521, -0.4659, 1.3048, -1.2154, -0.4436, -1.2450, 0.4739, 0.1569]],
  5276. device='cuda:0', grad_fn=<AddmmBackward>)
  5277. landmarks are: tensor([[[ 0.5241, -0.4470, 1.6806, 0.1313, -0.1612, 0.1929, 0.3378,
  5278. 0.0261],
  5279. [ 0.5759, -0.3948, 1.2072, -0.9849, -0.4672, -0.8156, 0.4912,
  5280. 0.5701],
  5281. [ 0.5832, -0.3593, 1.3515, -0.6077, -0.5249, -0.9310, 0.3353,
  5282. 0.3469],
  5283. [ 0.6003, -0.3747, 1.8885, -0.1997, -0.5018, -0.0149, 0.5490,
  5284. 0.1775],
  5285. [ 0.6040, -0.3614, 1.7672, -0.7001, -0.6404, -0.3768, 0.5778,
  5286. 0.3315],
  5287. [ 0.5443, -0.4079, 1.6633, 0.2853, -0.1325, 0.0888, 0.0558,
  5288. 0.2138],
  5289. [ 0.5462, -0.4090, 1.7037, 0.1159, -0.0746, -0.0765, 0.1427,
  5290. 0.4239],
  5291. [ 0.6141, -0.4153, 1.4208, -1.2697, -0.2940, -1.0234, 0.8644,
  5292. 0.1715]]], device='cuda:0')
  5293. loss_train_step before backward: tensor(0.0605, device='cuda:0', grad_fn=<MseLossBackward>)
  5294. loss_train_step after backward: tensor(0.0605, device='cuda:0', grad_fn=<MseLossBackward>)
  5295. loss_train: 2.2920665852725506
  5296. step: 25
  5297. running loss: 0.09168266341090202
  5298. Train Steps: 25/90 Loss: 0.0917 torch.Size([8, 600, 800])
  5299. torch.Size([8, 8])
  5300. tensor([[0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118],
  5301. [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927],
  5302. [0.6182, 0.3998, 0.8793, 0.4191, 0.3552, 0.4285, 0.6038, 0.5312],
  5303. [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517],
  5304. [0.6099, 0.4030, 0.8638, 0.5117, 0.4983, 0.4965, 0.5086, 0.5388],
  5305. [ nan, nan, 0.6412, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700],
  5306. [0.6055, 0.4015, 0.7425, 0.2033, 0.4113, 0.1883, 0.5217, 0.4823],
  5307. [0.6182, 0.3972, 0.8552, 0.5914, 0.3683, 0.4181, 0.5688, 0.5378]],
  5308. device='cuda:0', dtype=torch.float64)
  5309. predictions are: tensor([[ 0.5360, -0.4152, 1.2727, -0.8990, -0.4217, -0.8114, 0.3388, 0.2447],
  5310. [ 0.7213, -0.2711, 1.7282, -0.4758, -0.5359, -0.0893, 0.5664, 0.0697],
  5311. [ 0.5419, -0.4243, 1.8018, -0.3781, -0.5639, -0.1564, 0.3643, 0.1712],
  5312. [-0.1072, -0.8073, 1.2532, -1.0415, -0.3916, -1.0363, 0.3168, 0.2298],
  5313. [ 0.1826, -0.6807, 1.8562, -0.0210, -0.1675, -0.0548, 0.4331, 0.1280],
  5314. [-0.2581, -0.9261, 1.1611, -1.1396, -0.3878, -1.1883, 0.2695, 0.2118],
  5315. [-0.1382, -0.8743, 1.2336, -1.0734, -0.4734, -1.1985, 0.2244, 0.1587],
  5316. [ 0.9495, -0.1562, 1.7437, 0.2298, -0.6145, -0.1484, 0.3448, 0.1113]],
  5317. device='cuda:0', grad_fn=<AddmmBackward>)
  5318. landmarks are: tensor([[[ 0.6077, -0.3226, 0.9993, -0.6462, -0.2651, -1.3082, 0.2946,
  5319. 0.5401],
  5320. [ 0.6010, -0.4345, 1.8480, -0.3537, -0.6462, -0.2613, 0.6524,
  5321. -0.0099],
  5322. [ 0.5697, -0.4393, 1.7754, -0.3503, -0.6453, -0.3067, 0.5028,
  5323. 0.1677],
  5324. [ 0.5713, -0.3671, 0.8665, -1.0696, -0.3691, -1.2236, 0.3527,
  5325. 0.2622],
  5326. [ 0.5309, -0.4246, 1.7037, 0.0774, 0.0158, 0.0075, 0.0635,
  5327. 0.2026],
  5328. [-2.2859, -2.2859, 0.6760, -1.4083, -0.3286, -1.4160, 0.2487,
  5329. 0.3469],
  5330. [ 0.5109, -0.4316, 1.1436, -1.3467, -0.3864, -1.4160, 0.1239,
  5331. -0.0580],
  5332. [ 0.5697, -0.4514, 1.6642, 0.4459, -0.5850, -0.3546, 0.3414,
  5333. 0.1982]]], device='cuda:0')
  5334. loss_train_step before backward: tensor(0.1472, device='cuda:0', grad_fn=<MseLossBackward>)
  5335.  
  5336. loss_train_step after backward: tensor(0.1472, device='cuda:0', grad_fn=<MseLossBackward>)
  5337. loss_train: 2.439281862229109
  5338. step: 26
  5339. running loss: 0.09381853316265804
  5340. Train Steps: 26/90 Loss: 0.0938 torch.Size([8, 600, 800])
  5341. torch.Size([8, 8])
  5342. tensor([[0.6164, 0.4119, 0.7913, 0.2650, 0.3538, 0.3500, 0.5614, 0.5038],
  5343. [0.6207, 0.4110, 0.8738, 0.5000, 0.4800, 0.5633, 0.6300, 0.5433],
  5344. [0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750],
  5345. [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852],
  5346. [0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748],
  5347. [0.6208, 0.4082, 0.8538, 0.3067, 0.3588, 0.3717, 0.6112, 0.5517],
  5348. [0.6143, 0.4040, 0.8237, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973],
  5349. [0.6286, 0.4078, 0.8063, 0.2267, 0.4788, 0.1533, 0.5953, 0.4913]],
  5350. device='cuda:0', dtype=torch.float64)
  5351. predictions are: tensor([[ 0.7377, -0.2847, 1.2891, -0.9831, -0.6037, -0.6774, 0.3743, 0.1758],
  5352. [ 0.7532, -0.3084, 1.7839, -0.0081, -0.2489, 0.1553, 0.4885, 0.1983],
  5353. [ 0.0920, -0.6706, 1.3227, -0.8929, -0.4635, -0.9294, 0.2370, 0.2070],
  5354. [ 0.4035, -0.5192, 1.7010, 0.0193, -0.2705, 0.0588, 0.4027, 0.1585],
  5355. [ 0.3594, -0.5149, 1.6956, 0.1053, -0.4258, -0.2073, 0.3934, 0.1345],
  5356. [ 0.7228, -0.2703, 1.5759, -0.6418, -0.6070, -0.4731, 0.3995, 0.1768],
  5357. [ 0.1192, -0.6348, 1.3703, -0.8587, -0.4873, -0.9933, 0.2480, 0.1672],
  5358. [-0.3907, -0.9933, 1.3532, -1.0529, -0.2807, -1.1499, 0.4252, 0.2074]],
  5359. device='cuda:0', grad_fn=<AddmmBackward>)
  5360. landmarks are: tensor([[[ 0.5614, -0.3834, 1.3688, -1.0619, -0.6520, -0.6693, 0.3069,
  5361. 0.0412],
  5362. [ 0.5811, -0.3878, 1.7499, 0.0236, -0.0688, 0.3161, 0.6240,
  5363. 0.2237],
  5364. [ 0.5695, -0.3927, 1.3226, -0.9002, -0.4672, -1.1928, 0.1737,
  5365. 0.3700],
  5366. [ 0.5500, -0.4060, 1.7326, 0.0236, -0.1554, -0.0226, 0.1438,
  5367. 0.4171],
  5368. [ 0.6421, -0.3647, 1.5940, 0.3084, -0.5192, -0.3691, 1.1057,
  5369. 0.3692],
  5370. [ 0.5814, -0.4003, 1.6575, -0.8694, -0.6289, -0.5692, 0.5374,
  5371. 0.2622],
  5372. [ 0.5515, -0.4201, 1.5189, -0.7463, -0.4037, -1.3082, 0.0869,
  5373. 0.0111],
  5374. [ 0.6177, -0.4022, 1.4381, -1.2390, -0.0746, -1.5777, 0.4639,
  5375. -0.0168]]], device='cuda:0')
  5376. loss_train_step before backward: tensor(0.0591, device='cuda:0', grad_fn=<MseLossBackward>)
  5377. loss_train_step after backward: tensor(0.0591, device='cuda:0', grad_fn=<MseLossBackward>)
  5378. loss_train: 2.498351190239191
  5379. step: 27
  5380. running loss: 0.09253152556441448
  5381. Train Steps: 27/90 Loss: 0.0925 torch.Size([8, 600, 800])
  5382. torch.Size([8, 8])
  5383. tensor([[0.6205, 0.4062, 0.8337, 0.2683, 0.3675, 0.4283, 0.6338, 0.5250],
  5384. [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500],
  5385. [ nan, nan, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819],
  5386. [0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200],
  5387. [0.6030, 0.3969, 0.7988, 0.3917, 0.3450, 0.3667, 0.5266, 0.4700],
  5388. [0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507],
  5389. [0.6187, 0.4104, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683],
  5390. [ nan, nan, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563]],
  5391. device='cuda:0', dtype=torch.float64)
  5392. predictions are: tensor([[ 0.7374, -0.2478, 1.5541, -0.5682, -0.4088, -0.1455, 0.5582, 0.1738],
  5393. [ 0.8026, -0.1954, 1.4651, -0.5861, -0.5108, -0.2191, 0.4381, 0.1958],
  5394. [-0.3435, -0.9865, 1.1572, -0.7504, -0.3741, -0.8815, 0.1974, 0.2048],
  5395. [ 0.4001, -0.5106, 1.9784, -0.5306, -0.1866, -0.6455, 0.7068, 0.1653],
  5396. [ 0.8539, -0.2140, 1.7453, -0.1278, -0.4997, -0.1058, 0.3985, 0.1125],
  5397. [ 0.6154, -0.3897, 1.2614, -0.7802, -0.4414, -0.8929, 0.3122, 0.2156],
  5398. [ 0.7323, -0.2899, 1.3559, -0.6403, -0.4691, -0.8427, 0.2395, 0.2310],
  5399. [-0.4492, -1.0322, 1.4731, -0.4906, -0.4800, -0.5407, 0.1952, 0.1919]],
  5400. device='cuda:0', grad_fn=<AddmmBackward>)
  5401. landmarks are: tensor([[[ 0.5799, -0.4099, 1.5651, -1.0465, -0.5885, -0.3075, 0.6413,
  5402. 0.1390],
  5403. [ 0.5763, -0.4147, 1.3226, -1.0619, -0.6635, -0.4152, 0.5374,
  5404. 0.2545],
  5405. [-2.2859, -2.2859, 0.9438, -0.9967, -0.4614, -1.1851, 0.2468,
  5406. 0.4019],
  5407. [ 0.6089, -0.3948, 1.7383, -0.8617, -0.2536, -1.2390, 0.6009,
  5408. 0.1159],
  5409. [ 0.4992, -0.4525, 1.4035, -0.4768, -0.6924, -0.5923, 0.1465,
  5410. -0.1151],
  5411. [ 0.5185, -0.4252, 0.9647, -1.2928, -0.4788, -1.2390, 0.2617,
  5412. 0.2576],
  5413. [ 0.5718, -0.3905, 1.0053, -1.3305, -0.4614, -1.1235, 0.4450,
  5414. 0.3392],
  5415. [-2.2859, -2.2859, 1.4006, -0.8105, -0.6115, -0.8232, 0.0419,
  5416. 0.2837]]], device='cuda:0')
  5417. loss_train_step before backward: tensor(0.2237, device='cuda:0', grad_fn=<MseLossBackward>)
  5418. loss_train_step after backward: tensor(0.2237, device='cuda:0', grad_fn=<MseLossBackward>)
  5419. loss_train: 2.722070474177599
  5420. step: 28
  5421. running loss: 0.09721680264919996
  5422. Train Steps: 28/90 Loss: 0.0972 torch.Size([8, 600, 800])
  5423. torch.Size([8, 8])
  5424. tensor([[0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217],
  5425. [0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136],
  5426. [0.6236, 0.3967, 0.8675, 0.5400, 0.3862, 0.4517, 0.5825, 0.5200],
  5427. [0.6293, 0.4097, 0.8800, 0.2517, 0.5262, 0.2600, 0.7430, 0.5378],
  5428. [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383],
  5429. [0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412],
  5430. [0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320],
  5431. [0.6147, 0.4026, 0.6600, 0.2467, 0.4088, 0.2150, 0.5489, 0.5773]],
  5432. device='cuda:0', dtype=torch.float64)
  5433. predictions are: tensor([[ 0.7305, -0.3132, 1.8139, 0.0944, -0.3673, 0.3312, 0.4564, 0.1943],
  5434. [ 0.4552, -0.4721, 1.6356, -0.4081, -0.5364, -0.3788, 0.2233, 0.1668],
  5435. [ 0.5072, -0.4483, 1.6649, 0.2623, -0.4183, 0.0033, 0.3367, 0.1549],
  5436. [-0.0679, -0.8478, 1.5491, -1.2164, -0.0420, -1.1702, 0.8427, 0.1657],
  5437. [ 0.4734, -0.4277, 1.4086, -0.7942, -0.5250, -0.4327, 0.4632, 0.2055],
  5438. [ 0.7182, -0.3058, 1.7048, -0.6412, -0.4689, -0.8560, 0.4083, 0.2025],
  5439. [ 0.4958, -0.4098, 1.5977, -0.2361, -0.4395, -0.1448, 0.4607, 0.1308],
  5440. [ 0.3944, -0.5354, 1.0374, -1.0903, -0.5026, -1.1602, 0.1620, 0.2851]],
  5441. device='cuda:0', grad_fn=<AddmmBackward>)
  5442. landmarks are: tensor([[[ 0.6047, -0.4209, 1.7557, -0.0303, -0.4845, 0.2545, 0.6587,
  5443. 0.1236],
  5444. [ 0.5235, -0.4273, 1.7499, -0.4306, -0.5827, -0.4614, 0.1651,
  5445. 0.0862],
  5446. [ 0.5944, -0.4538, 1.7210, 0.2083, -0.5018, -0.1997, 0.4046,
  5447. 0.1159],
  5448. [ 0.6207, -0.3936, 1.7788, -1.1235, 0.1448, -1.0850, 1.1459,
  5449. 0.1982],
  5450. [ 0.5779, -0.4275, 1.5894, -0.8362, -0.6577, -0.5153, 0.5605,
  5451. 0.2006],
  5452. [ 0.5784, -0.4006, 1.7911, -0.5701, -0.5192, -1.0331, 0.4137,
  5453. 0.2139],
  5454. [ 0.6174, -0.4490, 1.8885, -0.0996, -0.4845, -0.3691, 0.9814,
  5455. 0.1715],
  5456. [ 0.5532, -0.4264, 0.7626, -1.1466, -0.3979, -1.2928, 0.2494,
  5457. 0.3808]]], device='cuda:0')
  5458. loss_train_step before backward: tensor(0.0290, device='cuda:0', grad_fn=<MseLossBackward>)
  5459. loss_train_step after backward: tensor(0.0290, device='cuda:0', grad_fn=<MseLossBackward>)
  5460. loss_train: 2.751073746010661
  5461. step: 29
  5462. running loss: 0.09486461193140211
  5463. Train Steps: 29/90 Loss: 0.0949 torch.Size([8, 600, 800])
  5464. torch.Size([8, 8])
  5465. tensor([[0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100],
  5466. [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683],
  5467. [0.6275, 0.4081, 0.8063, 0.2017, 0.4825, 0.1583, 0.6156, 0.4869],
  5468. [0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585],
  5469. [0.6221, 0.4107, 0.7788, 0.3033, 0.3950, 0.2817, 0.6075, 0.5517],
  5470. [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083],
  5471. [0.6357, 0.4097, 0.9038, 0.3883, 0.4213, 0.2950, 0.6686, 0.5390],
  5472. [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5687, 0.5567]],
  5473. device='cuda:0', dtype=torch.float64)
  5474. predictions are: tensor([[ 0.6250, -0.4144, 1.6957, 0.1613, -0.4815, 0.1147, 0.3846, 0.1633],
  5475. [ 1.0344, -0.0511, 1.5689, 0.0435, -0.5933, -0.2652, 0.2690, 0.2165],
  5476. [ 0.4900, -0.4944, 1.4873, -1.2496, -0.2852, -1.3362, 0.6333, 0.2112],
  5477. [ 0.2213, -0.5748, 1.6522, -0.3171, -0.2605, 0.1559, 0.4833, 0.1932],
  5478. [ 0.6316, -0.3629, 1.5085, -0.9237, -0.4612, -1.0227, 0.4347, 0.2365],
  5479. [ 0.4009, -0.5173, 1.6601, 0.0104, -0.4300, 0.1274, 0.4248, 0.1939],
  5480. [ 0.6442, -0.3294, 1.5468, -0.6170, -0.3791, -0.6776, 0.6063, 0.1621],
  5481. [-0.3326, -0.9680, 1.2587, -1.1487, -0.3353, -1.1027, 0.3654, 0.2544]],
  5482. device='cuda:0', grad_fn=<AddmmBackward>)
  5483. landmarks are: tensor([[[ 0.6125, -0.4273, 1.6402, 0.2006, -0.4788, -0.0303, 0.6182,
  5484. 0.0697],
  5485. [ 0.5777, -0.3915, 1.7961, 0.1698, -0.5192, -0.5307, 0.2141,
  5486. 0.3392],
  5487. [ 0.6124, -0.4010, 1.4381, -1.3544, -0.0573, -1.5546, 0.5573,
  5488. -0.0369],
  5489. [ 0.5115, -0.4332, 1.7557, -0.3152, -0.2536, 0.3931, 0.4139,
  5490. 0.2936],
  5491. [ 0.5875, -0.3888, 1.3111, -0.8848, -0.4614, -0.9849, 0.5201,
  5492. 0.2622],
  5493. [ 0.5713, -0.4105, 1.7557, 0.0467, -0.6520, -0.2690, 0.3988,
  5494. 0.5239],
  5495. [ 0.6504, -0.3936, 1.8885, -0.4922, -0.3402, -0.9233, 0.8022,
  5496. 0.2035],
  5497. [ 0.5824, -0.3625, 1.0405, -0.9325, -0.4268, -1.1389, 0.3411,
  5498. 0.2853]]], device='cuda:0')
  5499. loss_train_step before backward: tensor(0.0458, device='cuda:0', grad_fn=<MseLossBackward>)
  5500. loss_train_step after backward: tensor(0.0458, device='cuda:0', grad_fn=<MseLossBackward>)
  5501. loss_train: 2.796903731301427
  5502. step: 30
  5503. running loss: 0.09323012437671423
  5504.  
  5505. Train Steps: 30/90 Loss: 0.0932 torch.Size([8, 600, 800])
  5506. torch.Size([8, 8])
  5507. tensor([[0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446],
  5508. [0.6224, 0.4179, 0.8700, 0.5683, 0.4037, 0.4683, 0.5650, 0.5600],
  5509. [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609],
  5510. [0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412],
  5511. [0.6164, 0.4119, 0.7913, 0.2650, 0.3538, 0.3500, 0.5614, 0.5038],
  5512. [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261],
  5513. [0.6133, 0.4066, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358],
  5514. [0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567]],
  5515. device='cuda:0', dtype=torch.float64)
  5516. predictions are: tensor([[ 0.2950, -0.5529, 1.5650, -0.1983, -0.3732, -0.0056, 0.4572, 0.2051],
  5517. [ 0.6665, -0.3533, 1.6722, 0.3198, -0.4098, 0.0561, 0.4334, 0.2532],
  5518. [ 0.3297, -0.5676, 1.8212, -0.9010, -0.2561, -1.0195, 0.8730, 0.1724],
  5519. [ 0.9322, -0.1943, 1.7126, -0.6697, -0.4805, -0.9836, 0.5364, 0.2048],
  5520. [ 0.6783, -0.3507, 1.2963, -1.0498, -0.5918, -0.8010, 0.4516, 0.2145],
  5521. [ 0.4780, -0.4556, 1.7030, -0.1402, -0.4341, -0.1432, 0.4271, 0.2051],
  5522. [ 0.1772, -0.6843, 1.0516, -1.1705, -0.5276, -1.2873, 0.2467, 0.2395],
  5523. [ 0.4062, -0.5089, 1.7335, -0.1380, -0.1525, -0.0088, 0.4933, 0.1989]],
  5524. device='cuda:0', grad_fn=<AddmmBackward>)
  5525. landmarks are: tensor([[[ 0.5444, -0.3846, 1.6171, -0.1689, -0.5885, -0.0380, 0.1791,
  5526. 0.2296],
  5527. [ 0.5891, -0.3557, 1.7326, 0.3392, -0.4210, -0.1227, 0.3238,
  5528. 0.3007],
  5529. [ 0.6026, -0.4032, 1.7326, -0.7771, -0.3691, -0.8617, 0.9704,
  5530. 0.3050],
  5531. [ 0.5784, -0.4006, 1.7911, -0.5701, -0.5192, -1.0331, 0.4137,
  5532. 0.2139],
  5533. [ 0.5614, -0.3834, 1.3688, -1.0619, -0.6520, -0.6693, 0.3069,
  5534. 0.0412],
  5535. [ 0.5746, -0.3623, 1.7961, -0.1150, -0.3691, -0.0380, 0.2208,
  5536. 0.1439],
  5537. [ 0.5470, -0.4081, 0.8492, -1.0773, -0.5307, -1.1620, 0.0912,
  5538. 0.1890],
  5539. [ 0.5538, -0.4139, 1.7557, -0.1843, -0.0459, 0.1242, 0.4219,
  5540. 0.2853]]], device='cuda:0')
  5541. loss_train_step before backward: tensor(0.0192, device='cuda:0', grad_fn=<MseLossBackward>)
  5542. loss_train_step after backward: tensor(0.0192, device='cuda:0', grad_fn=<MseLossBackward>)
  5543. loss_train: 2.8161120787262917
  5544. step: 31
  5545. running loss: 0.09084232512020296
  5546. Train Steps: 31/90 Loss: 0.0908 torch.Size([8, 600, 800])
  5547. torch.Size([8, 8])
  5548. tensor([[0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301],
  5549. [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600],
  5550. [0.6284, 0.4093, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167],
  5551. [0.6200, 0.3999, 0.8653, 0.5207, 0.4100, 0.5125, 0.5975, 0.5103],
  5552. [0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6138, 0.5450],
  5553. [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383],
  5554. [0.6164, 0.4119, 0.7913, 0.2650, 0.3538, 0.3500, 0.5614, 0.5038],
  5555. [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250]],
  5556. device='cuda:0', dtype=torch.float64)
  5557. predictions are: tensor([[ 0.3344, -0.5869, 1.6304, -0.2963, -0.3374, -0.3541, 0.4984, 0.2313],
  5558. [ 0.6908, -0.3503, 1.6515, -0.0700, -0.4126, -0.3634, 0.4649, 0.2478],
  5559. [ 0.7039, -0.3515, 1.7289, -0.5859, -0.5596, -1.0050, 0.6109, 0.1697],
  5560. [ 0.4358, -0.5206, 1.6912, -0.1203, -0.3773, -0.2881, 0.4537, 0.1943],
  5561. [ 0.6527, -0.3829, 1.6514, -0.3966, -0.3452, -0.1804, 0.4861, 0.2402],
  5562. [ 0.2703, -0.5918, 1.6975, -0.4348, -0.2091, -0.3389, 0.6179, 0.2223],
  5563. [ 0.6149, -0.3941, 1.2463, -1.3619, -0.5945, -1.1821, 0.4503, 0.2203],
  5564. [ 0.6631, -0.3669, 1.7404, -0.3028, -0.3693, -0.3110, 0.5155, 0.2463]],
  5565. device='cuda:0', grad_fn=<AddmmBackward>)
  5566. landmarks are: tensor([[[ 0.6010, -0.4562, 1.7198, -0.0090, -0.3464, 0.0108, 0.6294,
  5567. 0.1627],
  5568. [ 0.5883, -0.3594, 1.7557, 0.2545, -0.4152, -0.0611, 0.3353,
  5569. 0.3007],
  5570. [ 0.6164, -0.3956, 1.8249, -0.1150, -0.6000, -0.5076, 0.5836,
  5571. 0.1005],
  5572. [ 0.5778, -0.4389, 1.7107, 0.1192, -0.3921, 0.0815, 0.4741,
  5573. 0.0711],
  5574. [ 0.5960, -0.4102, 1.8018, -0.1612, -0.3344, 0.1159, 0.5490,
  5575. 0.2314],
  5576. [ 0.5664, -0.4321, 1.8249, -0.2074, -0.2651, 0.4162, 0.5663,
  5577. 0.2006],
  5578. [ 0.5614, -0.3834, 1.3688, -1.0619, -0.6520, -0.6693, 0.3069,
  5579. 0.0412],
  5580. [ 0.5991, -0.3803, 1.8018, -0.0534, -0.3460, 0.1852, 0.5374,
  5581. 0.1390]]], device='cuda:0')
  5582. loss_train_step before backward: tensor(0.0467, device='cuda:0', grad_fn=<MseLossBackward>)
  5583. loss_train_step after backward: tensor(0.0467, device='cuda:0', grad_fn=<MseLossBackward>)
  5584. loss_train: 2.8628441616892815
  5585. step: 32
  5586. running loss: 0.08946388005279005
  5587. Train Steps: 32/90 Loss: 0.0895 torch.Size([8, 600, 800])
  5588. torch.Size([8, 8])
  5589. tensor([[0.6222, 0.3937, 0.8350, 0.5617, 0.4138, 0.4600, 0.5800, 0.5233],
  5590. [0.6218, 0.4185, 0.7338, 0.2650, 0.4625, 0.1950, 0.5687, 0.5800],
  5591. [0.6277, 0.4103, 0.8087, 0.5717, 0.4188, 0.4750, 0.5663, 0.6083],
  5592. [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960],
  5593. [0.6325, 0.4165, 0.9000, 0.4617, 0.3813, 0.4900, 0.7485, 0.5447],
  5594. [0.6321, 0.4048, 0.8738, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927],
  5595. [0.6353, 0.4128, 0.9138, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991],
  5596. [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400]],
  5597. device='cuda:0', dtype=torch.float64)
  5598. predictions are: tensor([[ 0.3982, -0.5270, 1.6897, 0.0661, -0.4512, -0.1298, 0.4194, 0.2139],
  5599. [ 0.4820, -0.4519, 1.0651, -1.1793, -0.3780, -1.2608, 0.3199, 0.3024],
  5600. [ 0.7377, -0.2993, 1.5560, 0.0784, -0.4592, -0.0693, 0.3546, 0.2839],
  5601. [ 0.3434, -0.5464, 1.7874, -0.2388, -0.3171, -0.1256, 0.5441, 0.1799],
  5602. [ 0.4071, -0.4964, 1.8216, -0.1906, -0.4884, 0.0066, 0.5674, 0.1910],
  5603. [ 0.8802, -0.2046, 1.6890, 0.0257, -0.5662, -0.2507, 0.3828, 0.1599],
  5604. [ 0.1629, -0.7121, 1.8712, -0.8241, -0.1683, -0.9695, 0.9181, 0.1917],
  5605. [ 0.9275, -0.1944, 1.2945, -1.3044, -0.3726, -1.3874, 0.5662, 0.2288]],
  5606. device='cuda:0', grad_fn=<AddmmBackward>)
  5607. landmarks are: tensor([[[ 0.5880, -0.4676, 1.5709, 0.3084, -0.3748, -0.1612, 0.3931,
  5608. 0.1313],
  5609. [ 0.5862, -0.3530, 1.1032, -1.0619, -0.1497, -1.3852, 0.3411,
  5610. 0.3931],
  5611. [ 0.6134, -0.3910, 1.4497, 0.3546, -0.3517, -0.0919, 0.3296,
  5612. 0.5239],
  5613. [ 0.5205, -0.4577, 1.7326, -0.1304, -0.1862, 0.0447, 0.1301,
  5614. 0.0051],
  5615. [ 0.6355, -0.3623, 1.8711, -0.1535, -0.5249, -0.0226, 1.1715,
  5616. 0.2302],
  5617. [ 0.6335, -0.4162, 1.7499, 0.3084, -0.4961, -0.2459, 0.6524,
  5618. -0.0102],
  5619. [ 0.6487, -0.3792, 1.9346, -0.6539, -0.1208, -0.7848, 1.0143,
  5620. 0.4814],
  5621. [ 0.5842, -0.3848, 1.2938, -1.2159, -0.2132, -1.4160, 0.5778,
  5622. 0.2083]]], device='cuda:0')
  5623. loss_train_step before backward: tensor(0.0337, device='cuda:0', grad_fn=<MseLossBackward>)
  5624. loss_train_step after backward: tensor(0.0337, device='cuda:0', grad_fn=<MseLossBackward>)
  5625. loss_train: 2.8965813778340816
  5626. step: 33
  5627. running loss: 0.08777519326769945
  5628. Train Steps: 33/90 Loss: 0.0878 torch.Size([8, 600, 800])
  5629. torch.Size([8, 8])
  5630. tensor([[0.6200, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183],
  5631. [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617],
  5632. [0.6109, 0.4041, 0.6975, 0.3167, 0.3513, 0.3383, 0.5153, 0.5319],
  5633. [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482],
  5634. [0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945],
  5635. [0.6263, 0.4039, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804],
  5636. [0.6163, 0.4114, 0.7650, 0.2017, 0.3763, 0.2867, 0.5631, 0.5071],
  5637. [0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367]],
  5638. device='cuda:0', dtype=torch.float64)
  5639. predictions are: tensor([[ 7.4253e-01, -3.1317e-01, 1.4131e+00, -8.4927e-01, -4.4791e-01,
  5640. -8.2762e-01, 5.7978e-01, 2.1282e-01],
  5641. [ 4.5485e-01, -4.8451e-01, 1.7093e+00, 3.4834e-01, -4.3732e-01,
  5642. -1.3684e-02, 3.7540e-01, 2.9459e-01],
  5643. [ 3.7311e-01, -5.3037e-01, 1.1646e+00, -8.6670e-01, -5.0887e-01,
  5644. -7.3844e-01, 2.4022e-01, 2.4823e-01],
  5645. [ 7.2810e-01, -3.0743e-01, 1.7642e+00, -2.3315e-01, -4.7777e-01,
  5646. -3.8799e-01, 5.0412e-01, 2.1321e-01],
  5647. [ 2.1975e-01, -6.6211e-01, 1.9491e+00, -2.8295e-01, -1.4412e-01,
  5648. -5.3107e-01, 8.4005e-01, 2.2348e-01],
  5649. [ 3.7371e-01, -5.1912e-01, 1.7898e+00, -9.8038e-04, -4.0995e-01,
  5650. 1.5966e-01, 5.4400e-01, 1.4020e-01],
  5651. [ 8.9009e-01, -2.2979e-01, 1.2286e+00, -9.8838e-01, -5.1678e-01,
  5652. -9.6966e-01, 4.1535e-01, 2.4532e-01],
  5653. [ 8.0319e-01, -2.6611e-01, 1.7041e+00, -7.1392e-01, -3.1660e-01,
  5654. -9.2966e-01, 6.5947e-01, 2.2535e-01]], device='cuda:0',
  5655. grad_fn=<AddmmBackward>)
  5656. landmarks are: tensor([[[ 0.5781, -0.4129, 1.4142, -0.9057, -0.5115, -0.9937, 0.4621,
  5657. 0.1080],
  5658. [ 0.5897, -0.3527, 1.8018, 0.2545, -0.5307, -0.3229, 0.3122,
  5659. 0.3084],
  5660. [ 0.5359, -0.4193, 0.9358, -0.8232, -0.6635, -0.7232, 0.0943,
  5661. 0.1710],
  5662. [ 0.5791, -0.4127, 1.8442, -0.3985, -0.6031, -0.6154, 0.4473,
  5663. 0.2464],
  5664. [ 0.6471, -0.3719, 1.9866, -0.3921, -0.3806, -0.5538, 1.0070,
  5665. 0.4600],
  5666. [ 0.6072, -0.4206, 1.8711, -0.2536, -0.6115, -0.1304, 0.6812,
  5667. -0.0670],
  5668. [ 0.5606, -0.3856, 1.2476, -1.3544, -0.5480, -0.9618, 0.3152,
  5669. 0.0562],
  5670. [ 0.6174, -0.4201, 1.7309, -0.8784, -0.4735, -0.9524, 0.6242,
  5671. 0.1931]]], device='cuda:0')
  5672. loss_train_step before backward: tensor(0.0267, device='cuda:0', grad_fn=<MseLossBackward>)
  5673. loss_train_step after backward: tensor(0.0267, device='cuda:0', grad_fn=<MseLossBackward>)
  5674. loss_train: 2.92330539226532
  5675. step: 34
  5676. running loss: 0.0859795703607447
  5677.  
  5678. Train Steps: 34/90 Loss: 0.0860 torch.Size([8, 600, 800])
  5679. torch.Size([8, 8])
  5680. tensor([[0.6219, 0.4097, 0.8738, 0.3400, 0.3563, 0.4117, 0.5975, 0.5683],
  5681. [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208],
  5682. [0.6148, 0.4076, 0.8666, 0.4820, 0.4138, 0.5067, 0.5250, 0.5767],
  5683. [0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689],
  5684. [0.6196, 0.4088, 0.8888, 0.4583, 0.4500, 0.5683, 0.6138, 0.5883],
  5685. [0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835],
  5686. [0.6205, 0.4016, 0.8350, 0.2717, 0.3987, 0.2550, 0.5787, 0.5133],
  5687. [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390]],
  5688. device='cuda:0', dtype=torch.float64)
  5689. predictions are: tensor([[ 0.8903, -0.2314, 1.6684, -0.4563, -0.6001, -0.5134, 0.4911, 0.2531],
  5690. [ 0.6476, -0.3743, 1.3517, -1.0156, -0.4161, -1.1527, 0.6536, 0.2023],
  5691. [ 0.3698, -0.5161, 1.6889, 0.1680, -0.3293, 0.1163, 0.4273, 0.2552],
  5692. [ 0.4036, -0.4773, 1.3534, -0.6551, -0.5117, -0.5085, 0.4268, 0.2266],
  5693. [ 0.5026, -0.4508, 1.7764, 0.0095, -0.2172, 0.1691, 0.5443, 0.2641],
  5694. [ 0.8662, -0.2546, 1.6973, -0.5305, -0.3797, -1.0850, 0.5950, 0.2191],
  5695. [ 0.7662, -0.2932, 1.5427, -0.9673, -0.3746, -1.1312, 0.6072, 0.1884],
  5696. [ 0.2559, -0.5938, 1.7963, 0.2113, -0.3554, 0.0888, 0.5212, 0.1696]],
  5697. device='cuda:0', grad_fn=<AddmmBackward>)
  5698. landmarks are: tensor([[[ 0.5867, -0.3937, 1.7499, -0.7155, -0.6404, -0.3844, 0.4739,
  5699. 0.3392],
  5700. [ 0.5885, -0.4429, 1.4266, -0.9926, -0.4383, -1.2313, 0.4228,
  5701. 0.1195],
  5702. [ 0.5539, -0.4031, 1.7168, -0.0594, -0.3748, 0.0543, 0.1390,
  5703. 0.3777],
  5704. [ 0.5327, -0.4381, 1.2880, -0.6308, -0.6866, -0.5307, 0.2658,
  5705. 0.3417],
  5706. [ 0.5759, -0.3978, 1.8192, -0.1689, -0.2074, 0.3392, 0.5490,
  5707. 0.4316],
  5708. [ 0.6117, -0.3877, 1.8249, -0.5230, -0.3517, -1.2851, 0.4200,
  5709. -0.0529],
  5710. [ 0.5800, -0.4312, 1.5709, -1.0311, -0.4441, -1.1081, 0.3873,
  5711. 0.0851],
  5712. [ 0.5250, -0.4661, 1.7383, -0.0765, -0.4268, -0.0226, 0.2535,
  5713. 0.2035]]], device='cuda:0')
  5714. loss_train_step before backward: tensor(0.0207, device='cuda:0', grad_fn=<MseLossBackward>)
  5715. loss_train_step after backward: tensor(0.0207, device='cuda:0', grad_fn=<MseLossBackward>)
  5716. loss_train: 2.944044752046466
  5717. step: 35
  5718. running loss: 0.08411556434418474
  5719. Train Steps: 35/90 Loss: 0.0841 torch.Size([8, 600, 800])
  5720. torch.Size([8, 8])
  5721. tensor([[ nan, nan, 0.8300, 0.3150, 0.3588, 0.3383, 0.5208, 0.5194],
  5722. [0.6185, 0.4067, 0.8838, 0.4450, 0.4037, 0.4733, 0.5213, 0.5142],
  5723. [0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938],
  5724. [ nan, nan, 0.7225, 0.2167, 0.3987, 0.2283, 0.5427, 0.5181],
  5725. [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235],
  5726. [0.6230, 0.4113, 0.7213, 0.1983, 0.4325, 0.2367, 0.6262, 0.5400],
  5727. [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167],
  5728. [0.6038, 0.3946, 0.8413, 0.4883, 0.3563, 0.4550, 0.5266, 0.4693]],
  5729. device='cuda:0', dtype=torch.float64)
  5730. predictions are: tensor([[ 0.1132, -0.7020, 1.3995, -0.7614, -0.5315, -0.8818, 0.3390, 0.2145],
  5731. [ 0.6932, -0.3358, 1.7980, -0.1423, -0.4874, -0.1082, 0.4483, 0.2249],
  5732. [ 1.0330, -0.1652, 1.8622, 0.3639, -0.4635, 0.0057, 0.5467, 0.1839],
  5733. [-0.3976, -1.0464, 1.0914, -1.1965, -0.3316, -1.3292, 0.3589, 0.2467],
  5734. [ 0.4669, -0.4691, 1.8673, 0.1160, -0.2885, 0.2232, 0.6384, 0.2342],
  5735. [ 0.8902, -0.2152, 1.3809, -1.1008, -0.2806, -1.3165, 0.6639, 0.2483],
  5736. [ 1.1518, -0.0532, 1.6962, -0.6029, -0.5221, -0.6022, 0.6503, 0.2845],
  5737. [ 0.8269, -0.2339, 1.7614, 0.1012, -0.4519, -0.0218, 0.5402, 0.1910]],
  5738. device='cuda:0', grad_fn=<AddmmBackward>)
  5739. landmarks are: tensor([[[-2.2859, -2.2859, 1.5478, -0.8309, -0.6289, -0.7232, 0.1198,
  5740. 0.1133],
  5741. [ 0.5708, -0.4075, 1.7961, -0.2305, -0.4210, -0.0996, 0.1219,
  5742. 0.0893],
  5743. [ 0.6395, -0.4213, 1.7037, 0.3623, -0.4326, -0.1073, 0.6560,
  5744. -0.0049],
  5745. [-2.2859, -2.2859, 1.0513, -1.2851, -0.4441, -1.2313, 0.2206,
  5746. 0.1073],
  5747. [ 0.5762, -0.4706, 1.7754, -0.0984, -0.3680, 0.2380, 0.6277,
  5748. 0.1322],
  5749. [ 0.5917, -0.3861, 1.0455, -1.3698, -0.2882, -1.1928, 0.6067,
  5750. 0.2083],
  5751. [ 0.5768, -0.3899, 1.3861, -0.7771, -0.5885, -0.5461, 0.5028,
  5752. 0.5624],
  5753. [ 0.5030, -0.4631, 1.5998, -0.0303, -0.6404, -0.1843, 0.1465,
  5754. -0.1181]]], device='cuda:0')
  5755. loss_train_step before backward: tensor(0.2440, device='cuda:0', grad_fn=<MseLossBackward>)
  5756. loss_train_step after backward: tensor(0.2440, device='cuda:0', grad_fn=<MseLossBackward>)
  5757. loss_train: 3.1880801748484373
  5758. step: 36
  5759. running loss: 0.08855778263467881
  5760. Train Steps: 36/90 Loss: 0.0886 torch.Size([8, 600, 800])
  5761. torch.Size([8, 8])
  5762. tensor([[0.6266, 0.4067, 0.8588, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267],
  5763. [0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183],
  5764. [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235],
  5765. [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333],
  5766. [0.6197, 0.4090, 0.7825, 0.2500, 0.4200, 0.2483, 0.5988, 0.5667],
  5767. [0.6275, 0.4157, 0.8337, 0.5800, 0.3763, 0.4200, 0.5547, 0.6125],
  5768. [0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767],
  5769. [0.6255, 0.4017, 0.8688, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901]],
  5770. device='cuda:0', dtype=torch.float64)
  5771. predictions are: tensor([[ 0.6768, -0.3240, 1.6868, -1.0501, -0.2776, -1.1492, 0.7312, 0.1656],
  5772. [ 0.6052, -0.3957, 1.7534, -0.1127, -0.5399, -0.3517, 0.5323, 0.1732],
  5773. [ 0.3163, -0.5653, 1.7590, -0.0033, -0.3279, 0.0508, 0.4693, 0.2262],
  5774. [ 0.7730, -0.3122, 1.8314, -0.1909, -0.5424, -0.1122, 0.4518, 0.2226],
  5775. [ 0.7522, -0.3238, 1.3224, -1.1095, -0.3794, -1.1402, 0.5556, 0.2368],
  5776. [ 0.8689, -0.2526, 1.6390, 0.1715, -0.4881, -0.2192, 0.3752, 0.2783],
  5777. [ 0.3152, -0.6392, 1.7184, 0.0382, -0.4561, -0.1316, 0.4029, 0.2143],
  5778. [ 0.4995, -0.4912, 1.5503, -0.8207, -0.5154, -0.7694, 0.6317, 0.1574]],
  5779. device='cuda:0', grad_fn=<AddmmBackward>)
  5780. landmarks are: tensor([[[ 0.6084, -0.4076, 1.6806, -0.9618, -0.2998, -0.9695, 0.6356,
  5781. 0.1467],
  5782. [ 0.5944, -0.4543, 1.8018, 0.0082, -0.6058, -0.4306, 0.4162,
  5783. 0.1082],
  5784. [ 0.5762, -0.4706, 1.7754, -0.0984, -0.3680, 0.2380, 0.6277,
  5785. 0.1322],
  5786. [ 0.6003, -0.3747, 1.8885, -0.1997, -0.5018, -0.0149, 0.5490,
  5787. 0.1775],
  5788. [ 0.5764, -0.3969, 1.3284, -1.1312, -0.3460, -1.1389, 0.4797,
  5789. 0.3315],
  5790. [ 0.6124, -0.3658, 1.5651, 0.3931, -0.5480, -0.3460, 0.2761,
  5791. 0.5431],
  5792. [ 0.5610, -0.4381, 1.7730, 0.0390, -0.4326, -0.1458, 0.1794,
  5793. 0.3777],
  5794. [ 0.6031, -0.4307, 1.7268, -0.8002, -0.6058, -0.6462, 0.6442,
  5795. -0.0220]]], device='cuda:0')
  5796. loss_train_step before backward: tensor(0.0140, device='cuda:0', grad_fn=<MseLossBackward>)
  5797. loss_train_step after backward: tensor(0.0140, device='cuda:0', grad_fn=<MseLossBackward>)
  5798. loss_train: 3.2020612247288227
  5799. step: 37
  5800. running loss: 0.08654219526294116
  5801. Train Steps: 37/90 Loss: 0.0865 torch.Size([8, 600, 800])
  5802. torch.Size([8, 8])
  5803. tensor([[0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301],
  5804. [0.6197, 0.4091, 0.8800, 0.4783, 0.3538, 0.4767, 0.5950, 0.5550],
  5805. [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333],
  5806. [0.6346, 0.4144, 0.9088, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899],
  5807. [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367],
  5808. [0.6213, 0.4131, 0.8438, 0.3550, 0.3513, 0.4400, 0.5716, 0.5123],
  5809. [0.6034, 0.4011, 0.7350, 0.2533, 0.3438, 0.3367, 0.5516, 0.5084],
  5810. [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426]],
  5811. device='cuda:0', dtype=torch.float64)
  5812. predictions are: tensor([[ 0.5374, -0.4509, 1.7941, 0.2094, -0.3937, 0.1123, 0.4918, 0.2022],
  5813. [ 0.6597, -0.3745, 1.7606, 0.0741, -0.4634, 0.1028, 0.4925, 0.2353],
  5814. [ 0.3929, -0.5305, 1.1526, -1.1111, -0.3497, -1.3115, 0.4014, 0.2133],
  5815. [ 0.3941, -0.5861, 1.9533, 0.0592, -0.3937, -0.2195, 0.7192, 0.2054],
  5816. [ 0.7141, -0.3497, 1.5147, -0.8251, -0.4656, -0.9256, 0.4631, 0.2019],
  5817. [ 0.7035, -0.3585, 1.7677, -0.3576, -0.5942, -0.2835, 0.4756, 0.1937],
  5818. [ 0.6302, -0.3811, 1.2863, -0.9902, -0.5425, -0.8486, 0.4420, 0.1782],
  5819. [ 0.4674, -0.5172, 1.7270, -0.2313, -0.5116, -0.5873, 0.4607, 0.1997]],
  5820. device='cuda:0', grad_fn=<AddmmBackward>)
  5821. landmarks are: tensor([[[ 0.6010, -0.4562, 1.7198, -0.0090, -0.3464, 0.0108, 0.6294,
  5822. 0.1627],
  5823. [ 0.5763, -0.3963, 1.7788, -0.0765, -0.6520, -0.0842, 0.4624,
  5824. 0.2776],
  5825. [ 0.5853, -0.3920, 1.1090, -1.3313, -0.2882, -1.3390, 0.4624,
  5826. 0.1775],
  5827. [ 0.6454, -0.3719, 1.9115, -0.1304, -0.5076, -0.2844, 1.0033,
  5828. 0.4386],
  5829. [ 0.5697, -0.4442, 1.3529, -0.9515, -0.5774, -0.7801, 0.5253,
  5830. 0.1931],
  5831. [ 0.5841, -0.3778, 1.6113, -0.6462, -0.6635, -0.2536, 0.3542,
  5832. 0.0802],
  5833. [ 0.5009, -0.4333, 1.1090, -1.1158, -0.6982, -0.7309, 0.2617,
  5834. 0.0622],
  5835. [ 0.5788, -0.4135, 1.8214, -0.2408, -0.6039, -0.7854, 0.4115,
  5836. 0.2203]]], device='cuda:0')
  5837. loss_train_step before backward: tensor(0.0173, device='cuda:0', grad_fn=<MseLossBackward>)
  5838. loss_train_step after backward: tensor(0.0173, device='cuda:0', grad_fn=<MseLossBackward>)
  5839. loss_train: 3.2193711400032043
  5840. step: 38
  5841. running loss: 0.08472029315797906
  5842.  
  5843. Train Steps: 38/90 Loss: 0.0847 torch.Size([8, 600, 800])
  5844. torch.Size([8, 8])
  5845. tensor([[0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869],
  5846. [0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131],
  5847. [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869],
  5848. [0.6124, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485],
  5849. [0.6198, 0.3997, 0.8582, 0.5361, 0.4117, 0.5016, 0.5942, 0.5134],
  5850. [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667],
  5851. [0.6332, 0.4118, 0.9238, 0.4267, 0.4012, 0.4733, 0.7525, 0.5436],
  5852. [0.6268, 0.4052, 0.8175, 0.2250, 0.4688, 0.1917, 0.6375, 0.5267]],
  5853. device='cuda:0', dtype=torch.float64)
  5854. predictions are: tensor([[ 0.3966, -0.5936, 1.8039, 0.2802, -0.5525, -0.2142, 0.4026, 0.1941],
  5855. [ 0.7885, -0.2987, 1.7053, -0.3514, -0.5758, -0.1134, 0.5165, 0.1591],
  5856. [ 0.6118, -0.3937, 1.6096, -0.7545, -0.3470, -1.0085, 0.4864, 0.1848],
  5857. [ 0.2276, -0.6336, 1.0980, -0.9782, -0.5359, -0.9213, 0.2522, 0.2356],
  5858. [ 0.6053, -0.4369, 1.8010, 0.2543, -0.5330, 0.1212, 0.4244, 0.2200],
  5859. [ 0.6160, -0.4273, 1.6637, -0.5837, -0.6304, -0.6150, 0.4379, 0.2124],
  5860. [ 0.6644, -0.3704, 1.7846, -0.1783, -0.4724, 0.0512, 0.6495, 0.2111],
  5861. [ 0.3931, -0.5582, 1.4735, -1.0682, -0.2781, -1.2448, 0.7007, 0.1874]],
  5862. device='cuda:0', grad_fn=<AddmmBackward>)
  5863. landmarks are: tensor([[[ 0.5320, -0.4488, 1.6633, 0.3315, -0.5018, -0.2459, 0.0828,
  5864. -0.0370],
  5865. [ 0.5581, -0.4586, 1.5586, -0.3747, -0.6792, -0.2391, 0.4455,
  5866. 0.0840],
  5867. [ 0.6119, -0.3927, 1.6979, -0.7925, -0.2536, -1.3698, 0.4282,
  5868. -0.0368],
  5869. [ 0.5425, -0.3998, 0.9263, -0.8683, -0.6000, -1.0157, 0.0990,
  5870. 0.2476],
  5871. [ 0.5767, -0.4396, 1.6782, 0.1905, -0.3844, 0.0308, 0.4588,
  5872. 0.0855],
  5873. [ 0.5768, -0.3857, 1.5305, -0.7694, -0.6462, -0.6308, 0.3988,
  5874. 0.3315],
  5875. [ 0.6388, -0.3840, 1.9808, -0.3152, -0.4326, -0.0996, 1.1897,
  5876. 0.2249],
  5877. [ 0.6092, -0.4143, 1.4901, -1.2467, -0.1208, -1.4006, 0.6587,
  5878. 0.1467]]], device='cuda:0')
  5879. loss_train_step before backward: tensor(0.0228, device='cuda:0', grad_fn=<MseLossBackward>)
  5880. loss_train_step after backward: tensor(0.0228, device='cuda:0', grad_fn=<MseLossBackward>)
  5881. loss_train: 3.242176616564393
  5882. step: 39
  5883. running loss: 0.08313273375806136
  5884. Train Steps: 39/90 Loss: 0.0831 torch.Size([8, 600, 800])
  5885. torch.Size([8, 8])
  5886. tensor([[0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050],
  5887. [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383],
  5888. [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600],
  5889. [0.6230, 0.4113, 0.7213, 0.1983, 0.4325, 0.2367, 0.6262, 0.5400],
  5890. [0.6250, 0.4103, 0.8950, 0.4400, 0.3912, 0.5650, 0.6050, 0.5133],
  5891. [0.6161, 0.4076, 0.8900, 0.4667, 0.4125, 0.5917, 0.6262, 0.5367],
  5892. [0.6201, 0.4017, 0.8871, 0.4621, 0.3517, 0.4675, 0.5999, 0.5106],
  5893. [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334]],
  5894. device='cuda:0', dtype=torch.float64)
  5895. predictions are: tensor([[ 3.5106e-01, -6.2456e-01, 1.8187e+00, -1.4204e-01, -7.0050e-01,
  5896. -4.5938e-01, 4.0855e-01, 1.4211e-01],
  5897. [ 3.8717e-01, -5.6815e-01, 1.1987e+00, -1.1633e+00, -4.1027e-01,
  5898. -1.3726e+00, 3.9880e-01, 2.0288e-01],
  5899. [ 5.8238e-01, -4.1615e-01, 1.8140e+00, -5.7646e-02, -3.4776e-01,
  5900. -5.9229e-03, 4.6542e-01, 2.3383e-01],
  5901. [ 1.1506e-01, -7.4752e-01, 1.2902e+00, -1.1127e+00, -3.6575e-01,
  5902. -1.3568e+00, 4.9368e-01, 2.0120e-01],
  5903. [ 9.3898e-01, -1.8494e-01, 1.8908e+00, 1.1370e-03, -5.2647e-01,
  5904. 2.3662e-01, 5.5223e-01, 1.8537e-01],
  5905. [ 7.6586e-01, -3.0031e-01, 1.8964e+00, 9.3340e-02, -4.1862e-01,
  5906. 3.0666e-01, 5.2751e-01, 2.0029e-01],
  5907. [ 6.4795e-01, -4.0853e-01, 1.7987e+00, -6.2643e-03, -6.3486e-01,
  5908. -1.4099e-01, 4.4951e-01, 1.5382e-01],
  5909. [ 4.4453e-01, -5.2647e-01, 1.3447e+00, -1.0519e+00, -6.0382e-01,
  5910. -9.8575e-01, 5.0882e-01, 1.7739e-01]], device='cuda:0',
  5911. grad_fn=<AddmmBackward>)
  5912. landmarks are: tensor([[[ 0.5844, -0.4466, 1.8423, -0.1997, -0.5942, -0.3998, 0.4219,
  5913. 0.0467],
  5914. [ 0.5911, -0.3984, 1.1956, -1.0850, -0.0804, -1.5392, 0.4393,
  5915. 0.2006],
  5916. [ 0.5652, -0.4325, 1.7388, -0.1429, -0.0688, 0.3469, 0.4970,
  5917. 0.3007],
  5918. [ 0.5917, -0.3861, 1.0455, -1.3698, -0.2882, -1.1928, 0.6067,
  5919. 0.2083],
  5920. [ 0.6010, -0.3909, 1.8480, -0.2536, -0.4788, 0.3238, 0.5085,
  5921. 0.0851],
  5922. [ 0.5598, -0.4032, 1.8249, -0.1304, -0.3806, 0.4470, 0.6067,
  5923. 0.1929],
  5924. [ 0.5783, -0.4306, 1.8114, -0.1515, -0.6617, -0.1268, 0.4851,
  5925. 0.0727],
  5926. [ 0.5730, -0.4562, 1.2195, -1.2440, -0.5497, -0.7711, 0.5704,
  5927. 0.1779]]], device='cuda:0')
  5928. loss_train_step before backward: tensor(0.0254, device='cuda:0', grad_fn=<MseLossBackward>)
  5929. loss_train_step after backward: tensor(0.0254, device='cuda:0', grad_fn=<MseLossBackward>)
  5930. loss_train: 3.2675318624824286
  5931. step: 40
  5932. running loss: 0.08168829656206071
  5933. Train Steps: 40/90 Loss: 0.0817 torch.Size([8, 600, 800])
  5934. torch.Size([8, 8])
  5935. tensor([[0.6218, 0.4098, 0.7238, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350],
  5936. [0.6120, 0.4014, 0.6863, 0.2817, 0.3700, 0.2783, 0.5513, 0.5667],
  5937. [0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938],
  5938. [0.6148, 0.4076, 0.8666, 0.4820, 0.4138, 0.5067, 0.5250, 0.5767],
  5939. [0.6350, 0.4043, 0.8738, 0.5650, 0.3850, 0.4750, 0.6401, 0.4950],
  5940. [0.6225, 0.4116, 0.8662, 0.3517, 0.3663, 0.3233, 0.5837, 0.5317],
  5941. [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446],
  5942. [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351]],
  5943. device='cuda:0', dtype=torch.float64)
  5944. predictions are: tensor([[ 0.4367, -0.5225, 1.2960, -1.1648, -0.3793, -1.1430, 0.5874, 0.1746],
  5945. [ 0.1962, -0.6501, 1.1282, -1.0401, -0.5617, -0.9396, 0.2712, 0.2072],
  5946. [ 0.6875, -0.3992, 1.8857, 0.2476, -0.6058, -0.0681, 0.4812, 0.1275],
  5947. [ 0.7143, -0.3665, 1.8605, 0.0773, -0.5132, 0.1704, 0.4499, 0.2356],
  5948. [ 0.6062, -0.4491, 1.9378, 0.2578, -0.6167, -0.0194, 0.5542, 0.1477],
  5949. [ 0.5631, -0.4239, 1.6553, -0.6901, -0.5287, -0.7452, 0.4552, 0.1585],
  5950. [ 0.6069, -0.3986, 1.7356, -0.1607, -0.5416, 0.0315, 0.4627, 0.1705],
  5951. [ 0.2619, -0.6570, 1.2979, -1.1250, -0.3744, -1.2685, 0.4204, 0.1447]],
  5952. device='cuda:0', grad_fn=<AddmmBackward>)
  5953. landmarks are: tensor([[[ 5.8614e-01, -3.9292e-01, 1.0570e+00, -1.4314e+00, -3.2864e-01,
  5954. -1.1235e+00, 6.1824e-01, 1.8522e-01],
  5955. [ 5.4099e-01, -4.3210e-01, 8.8383e-01, -9.8491e-01, -5.7691e-01,
  5956. -1.0003e+00, 2.6028e-01, 3.3149e-01],
  5957. [ 6.3949e-01, -4.2132e-01, 1.7037e+00, 3.6228e-01, -4.3256e-01,
  5958. -1.0731e-01, 6.5602e-01, -4.8817e-03],
  5959. [ 5.5393e-01, -4.0308e-01, 1.7168e+00, -5.9398e-02, -3.7483e-01,
  5960. 5.4350e-02, 1.3903e-01, 3.7768e-01],
  5961. [ 6.4706e-01, -4.1832e-01, 1.7499e+00, 3.2379e-01, -5.0762e-01,
  5962. -9.1917e-02, 6.7064e-01, 4.6189e-04],
  5963. [ 5.8932e-01, -3.8468e-01, 1.7152e+00, -6.6159e-01, -5.9423e-01,
  5964. -7.9246e-01, 4.1039e-01, 1.6982e-01],
  5965. [ 5.4440e-01, -3.8460e-01, 1.6171e+00, -1.6890e-01, -5.8845e-01,
  5966. -3.8029e-02, 1.7915e-01, 2.2961e-01],
  5967. [ 5.6637e-01, -4.3212e-01, 1.2862e+00, -1.0003e+00, -2.1894e-01,
  5968. -1.4608e+00, 3.8827e-01, 1.8549e-01]]], device='cuda:0')
  5969. loss_train_step before backward: tensor(0.0188, device='cuda:0', grad_fn=<MseLossBackward>)
  5970. loss_train_step after backward: tensor(0.0188, device='cuda:0', grad_fn=<MseLossBackward>)
  5971. loss_train: 3.2863815408200026
  5972. step: 41
  5973. running loss: 0.08015564733707323
  5974. Train Steps: 41/90 Loss: 0.0802 torch.Size([8, 600, 800])
  5975. torch.Size([8, 8])
  5976. tensor([[0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5988, 0.5667],
  5977. [0.6250, 0.4146, 0.8838, 0.3933, 0.3588, 0.4283, 0.6162, 0.5367],
  5978. [0.6185, 0.4067, 0.8838, 0.4450, 0.4037, 0.4733, 0.5213, 0.5142],
  5979. [ nan, nan, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621],
  5980. [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600],
  5981. [0.6262, 0.4052, 0.8888, 0.4700, 0.3675, 0.5117, 0.6350, 0.5233],
  5982. [0.6126, 0.3954, 0.8538, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350],
  5983. [0.6164, 0.4119, 0.7913, 0.2650, 0.3538, 0.3500, 0.5614, 0.5038]],
  5984. device='cuda:0', dtype=torch.float64)
  5985. predictions are: tensor([[ 0.6545, -0.3618, 1.5680, -0.6445, -0.5170, -0.7510, 0.4842, 0.1577],
  5986. [ 0.4752, -0.4913, 1.6541, -0.3822, -0.6434, -0.4164, 0.4551, 0.2095],
  5987. [ 0.3454, -0.5692, 1.5968, -0.1963, -0.6108, -0.2757, 0.2590, 0.1625],
  5988. [-0.3859, -1.0327, 1.4048, -1.1683, -0.1083, -1.1432, 0.6973, 0.1848],
  5989. [ 0.7585, -0.2932, 1.6858, -0.1326, -0.3363, 0.0059, 0.4644, 0.1856],
  5990. [ 0.8664, -0.2708, 1.8608, -0.0368, -0.6589, -0.0560, 0.5618, 0.1448],
  5991. [ 0.7278, -0.3654, 1.6793, 0.0226, -0.5101, -0.1967, 0.3999, 0.1454],
  5992. [ 0.5154, -0.4584, 1.2732, -1.0080, -0.6305, -0.8144, 0.3943, 0.1073]],
  5993. device='cuda:0', grad_fn=<AddmmBackward>)
  5994. landmarks are: tensor([[[ 0.5783, -0.3933, 1.6748, -0.6154, -0.5769, -0.6462, 0.4797,
  5995. 0.3315],
  5996. [ 0.6009, -0.3710, 1.7961, -0.4691, -0.6289, -0.3075, 0.5605,
  5997. 0.1929],
  5998. [ 0.5708, -0.4075, 1.7961, -0.2305, -0.4210, -0.0996, 0.1219,
  5999. 0.0893],
  6000. [-2.2859, -2.2859, 1.7210, -0.9772, 0.1852, -1.3698, 0.7986,
  6001. 0.3104],
  6002. [ 0.5652, -0.4325, 1.7388, -0.1429, -0.0688, 0.3469, 0.4970,
  6003. 0.3007],
  6004. [ 0.6063, -0.4142, 1.8192, -0.1150, -0.5885, 0.0774, 0.6471,
  6005. 0.1313],
  6006. [ 0.5435, -0.4597, 1.6575, 0.0159, -0.3229, -0.1150, 0.1877,
  6007. 0.1854],
  6008. [ 0.5614, -0.3834, 1.3688, -1.0619, -0.6520, -0.6693, 0.3069,
  6009. 0.0412]]], device='cuda:0')
  6010. loss_train_step before backward: tensor(0.1001, device='cuda:0', grad_fn=<MseLossBackward>)
  6011. loss_train_step after backward: tensor(0.1001, device='cuda:0', grad_fn=<MseLossBackward>)
  6012. loss_train: 3.386511316522956
  6013. step: 42
  6014. running loss: 0.08063122182197514
  6015.  
  6016. Train Steps: 42/90 Loss: 0.0806 torch.Size([8, 600, 800])
  6017. torch.Size([8, 8])
  6018. tensor([[0.6151, 0.4085, 0.8750, 0.4367, 0.3887, 0.4367, 0.5066, 0.5846],
  6019. [0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058],
  6020. [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167],
  6021. [0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367],
  6022. [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795],
  6023. [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450],
  6024. [ nan, nan, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737],
  6025. [0.6250, 0.4146, 0.8838, 0.3933, 0.3588, 0.4283, 0.6162, 0.5367]],
  6026. device='cuda:0', dtype=torch.float64)
  6027. predictions are: tensor([[ 0.3529, -0.5591, 1.5007, -0.3760, -0.6113, -0.3282, 0.2479, 0.2410],
  6028. [ 0.8080, -0.2771, 1.6025, -0.3582, -0.6192, -0.1207, 0.3971, 0.0940],
  6029. [ 0.5119, -0.4641, 1.4563, -0.7292, -0.5492, -0.8653, 0.3301, 0.0829],
  6030. [ 0.2396, -0.6196, 1.5460, -0.4810, -0.4276, -0.7962, 0.3526, 0.1782],
  6031. [ 0.9781, -0.1953, 1.7784, 0.0812, -0.5292, 0.1354, 0.6026, 0.1393],
  6032. [ 0.7572, -0.3118, 1.6693, -0.2089, -0.5883, -0.2834, 0.4285, 0.1437],
  6033. [-0.4364, -1.0733, 1.4616, -1.2481, -0.0045, -1.1192, 0.8184, 0.1947],
  6034. [ 0.5306, -0.4469, 1.6497, -0.4279, -0.6176, -0.3333, 0.4536, 0.1986]],
  6035. device='cuda:0', grad_fn=<AddmmBackward>)
  6036. landmarks are: tensor([[[ 0.5553, -0.3992, 1.7557, -0.2690, -0.4903, -0.2690, 0.0542,
  6037. 0.4145],
  6038. [ 0.5245, -0.4361, 1.5940, -0.2921, -0.5480, -0.0919, 0.2432,
  6039. 0.0502],
  6040. [ 0.5483, -0.4249, 1.5305, -0.7386, -0.6115, -0.8694, 0.3353,
  6041. 0.1005],
  6042. [ 0.5900, -0.3932, 1.8307, -0.3921, -0.4268, -1.1851, 0.3758,
  6043. 0.1929],
  6044. [ 0.6355, -0.4080, 1.6113, 0.1852, -0.4730, 0.1467, 0.9996,
  6045. 0.3905],
  6046. [ 0.5785, -0.4252, 1.7676, -0.1602, -0.5845, -0.3446, 0.4566,
  6047. 0.2314],
  6048. [-2.2859, -2.2859, 1.6979, -1.1081, 0.2487, -1.2697, 1.1020,
  6049. 0.3638],
  6050. [ 0.6009, -0.3710, 1.7961, -0.4691, -0.6289, -0.3075, 0.5605,
  6051. 0.1929]]], device='cuda:0')
  6052. loss_train_step before backward: tensor(0.1016, device='cuda:0', grad_fn=<MseLossBackward>)
  6053. loss_train_step after backward: tensor(0.1016, device='cuda:0', grad_fn=<MseLossBackward>)
  6054. loss_train: 3.488085037097335
  6055. step: 43
  6056. running loss: 0.08111825667668221
  6057. Train Steps: 43/90 Loss: 0.0811 torch.Size([8, 600, 800])
  6058. torch.Size([8, 8])
  6059. tensor([[0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274],
  6060. [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183],
  6061. [0.6151, 0.4125, 0.8738, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483],
  6062. [0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507],
  6063. [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583],
  6064. [0.6201, 0.4064, 0.8688, 0.5050, 0.4225, 0.5100, 0.6138, 0.5500],
  6065. [ nan, nan, 0.8363, 0.3317, 0.3563, 0.3367, 0.5329, 0.5142],
  6066. [0.6257, 0.4167, 0.8775, 0.3433, 0.3563, 0.4133, 0.6200, 0.5667]],
  6067. device='cuda:0', dtype=torch.float64)
  6068. predictions are: tensor([[ 0.5349, -0.4455, 1.7570, -0.2074, -0.5198, -0.4080, 0.4787, 0.1820],
  6069. [ 0.8963, -0.2477, 1.9448, -0.1026, -0.4083, 0.2798, 0.6832, 0.1334],
  6070. [ 0.3211, -0.5580, 1.6122, -0.2500, -0.5693, -0.4049, 0.2903, 0.1702],
  6071. [ 0.3067, -0.5936, 1.0340, -1.2498, -0.3961, -1.2777, 0.3361, 0.1416],
  6072. [ 0.9243, -0.1990, 1.6087, -0.5908, -0.5360, -0.4200, 0.5517, 0.1284],
  6073. [ 0.6387, -0.3762, 1.7011, 0.0358, -0.4203, 0.0398, 0.3977, 0.1838],
  6074. [-0.7036, -1.2559, 1.3289, -0.8699, -0.4771, -0.9723, 0.2688, 0.1476],
  6075. [ 0.5620, -0.4357, 1.7383, -0.5753, -0.5294, -0.5052, 0.5535, 0.1751]],
  6076. device='cuda:0', grad_fn=<AddmmBackward>)
  6077. landmarks are: tensor([[[ 0.6108, -0.4008, 1.9088, -0.0253, -0.5769, -0.3075, 0.7905,
  6078. 0.1499],
  6079. [ 0.5934, -0.4276, 1.8192, -0.1458, -0.4499, 0.3777, 0.6702,
  6080. 0.1082],
  6081. [ 0.5551, -0.3807, 1.7499, -0.2459, -0.6346, -0.5384, 0.0871,
  6082. 0.2468],
  6083. [ 0.5185, -0.4252, 0.9647, -1.2928, -0.4788, -1.2390, 0.2617,
  6084. 0.2576],
  6085. [ 0.5761, -0.3966, 1.6171, -0.4845, -0.6346, -0.4691, 0.4739,
  6086. 0.2930],
  6087. [ 0.5784, -0.4088, 1.7268, 0.0467, -0.3344, 0.0697, 0.5490,
  6088. 0.2545],
  6089. [-2.2859, -2.2859, 1.5767, -0.7540, -0.6404, -0.7309, 0.1753,
  6090. 0.0893],
  6091. [ 0.6040, -0.3614, 1.7672, -0.7001, -0.6404, -0.3768, 0.5778,
  6092. 0.3315]]], device='cuda:0')
  6093. loss_train_step before backward: tensor(0.0732, device='cuda:0', grad_fn=<MseLossBackward>)
  6094. loss_train_step after backward: tensor(0.0732, device='cuda:0', grad_fn=<MseLossBackward>)
  6095. loss_train: 3.561272194609046
  6096. step: 44
  6097. running loss: 0.08093800442293286
  6098. Train Steps: 44/90 Loss: 0.0809 torch.Size([8, 600, 800])
  6099. torch.Size([8, 8])
  6100. tensor([[0.6233, 0.4091, 0.8100, 0.2950, 0.3563, 0.3883, 0.6013, 0.5200],
  6101. [0.6212, 0.4171, 0.7875, 0.3633, 0.3813, 0.2933, 0.5675, 0.5700],
  6102. [0.6102, 0.4005, 0.8688, 0.5100, 0.4813, 0.5400, 0.5404, 0.5064],
  6103. [ nan, nan, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702],
  6104. [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235],
  6105. [0.6268, 0.4061, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300],
  6106. [0.6212, 0.4033, 0.8938, 0.4167, 0.3813, 0.4267, 0.5613, 0.5583],
  6107. [ nan, nan, 0.9088, 0.3783, 0.4562, 0.2617, 0.6741, 0.5575]],
  6108. device='cuda:0', dtype=torch.float64)
  6109. predictions are: tensor([[ 0.5818, -0.3966, 1.3973, -0.8580, -0.6440, -0.5126, 0.4475, 0.1186],
  6110. [ 0.4153, -0.4874, 1.2256, -0.7259, -0.5466, -0.6686, 0.3026, 0.2101],
  6111. [ 0.7514, -0.3059, 1.7006, 0.0987, -0.3783, 0.1060, 0.2963, 0.1680],
  6112. [-0.2648, -0.9652, 1.7160, -0.6842, -0.1627, -0.8070, 0.7399, 0.2166],
  6113. [ 0.5834, -0.4091, 1.7065, -0.0113, -0.5106, 0.1586, 0.4210, 0.1692],
  6114. [ 0.2055, -0.6349, 1.4568, -0.9875, -0.3721, -1.0002, 0.5307, 0.1490],
  6115. [ 0.5841, -0.4346, 1.6850, -0.2511, -0.6876, -0.1978, 0.3221, 0.1884],
  6116. [ 0.0765, -0.7286, 1.7289, -0.5439, -0.2815, -0.7442, 0.6015, 0.1698]],
  6117. device='cuda:0', grad_fn=<AddmmBackward>)
  6118. landmarks are: tensor([[[ 0.5932, -0.3962, 1.4554, -0.9233, -0.6404, -0.4922, 0.4912,
  6119. 0.1159],
  6120. [ 0.5832, -0.3593, 1.3515, -0.6077, -0.5249, -0.9310, 0.3353,
  6121. 0.3469],
  6122. [ 0.5328, -0.4361, 1.7268, 0.0697, -0.0630, 0.2083, 0.2103,
  6123. 0.0532],
  6124. [-2.2859, -2.2859, 1.8942, -0.6693, 0.0871, -1.2236, 1.1130,
  6125. 0.3478],
  6126. [ 0.5762, -0.4706, 1.7754, -0.0984, -0.3680, 0.2380, 0.6277,
  6127. 0.1322],
  6128. [ 0.6093, -0.4104, 1.5709, -1.1620, -0.1727, -1.2313, 0.6471,
  6129. 0.1621],
  6130. [ 0.5832, -0.4231, 1.8423, -0.3614, -0.5249, -0.3152, 0.3065,
  6131. 0.2930],
  6132. [-2.2859, -2.2859, 1.9115, -0.5384, -0.1785, -1.0773, 0.8278,
  6133. 0.2890]]], device='cuda:0')
  6134. loss_train_step before backward: tensor(0.2405, device='cuda:0', grad_fn=<MseLossBackward>)
  6135. loss_train_step after backward: tensor(0.2405, device='cuda:0', grad_fn=<MseLossBackward>)
  6136. loss_train: 3.8017283137887716
  6137. step: 45
  6138. running loss: 0.08448285141752826
  6139. Train Steps: 45/90 Loss: 0.0845 torch.Size([8, 600, 800])
  6140. torch.Size([8, 8])
  6141. tensor([[0.6204, 0.4055, 0.8438, 0.5733, 0.4574, 0.4801, 0.5487, 0.5617],
  6142. [0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133],
  6143. [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6188, 0.5283],
  6144. [0.6109, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117],
  6145. [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617],
  6146. [0.6124, 0.4075, 0.7696, 0.4153, 0.3475, 0.3767, 0.5157, 0.5427],
  6147. [0.6127, 0.4066, 0.8550, 0.5567, 0.4662, 0.5141, 0.5070, 0.5412],
  6148. [0.6276, 0.4120, 0.8738, 0.3133, 0.4225, 0.2217, 0.6203, 0.4892]],
  6149. device='cuda:0', dtype=torch.float64)
  6150. predictions are: tensor([[ 0.2672, -0.5984, 1.5340, -0.0941, -0.3328, -0.1551, 0.3480, 0.2435],
  6151. [ 0.4768, -0.4793, 1.7141, -0.1435, -0.5566, -0.1459, 0.5068, 0.1785],
  6152. [ 0.1151, -0.7118, 1.6953, -0.8581, -0.3410, -0.9446, 0.6515, 0.1427],
  6153. [ 0.2253, -0.6507, 1.7210, -0.3347, -0.5786, -0.4340, 0.3847, 0.1178],
  6154. [ 0.5995, -0.4095, 1.6505, -0.4597, -0.5058, -0.2441, 0.3787, 0.1836],
  6155. [ 0.4865, -0.4288, 1.3854, -0.7569, -0.5593, -0.5213, 0.3683, 0.1623],
  6156. [ 0.2983, -0.6022, 1.6223, 0.0036, -0.2758, -0.0715, 0.3743, 0.2261],
  6157. [-0.1031, -0.8570, 1.6374, -0.9917, -0.2689, -1.0909, 0.6932, 0.1572]],
  6158. device='cuda:0', grad_fn=<AddmmBackward>)
  6159. landmarks are: tensor([[[ 0.5795, -0.4129, 1.6113, 0.3623, -0.1733, -0.0684, 0.2487,
  6160. 0.3084],
  6161. [ 0.6148, -0.4130, 1.6864, 0.1698, -0.5307, -0.1150, 0.6125,
  6162. 0.0851],
  6163. [ 0.6062, -0.3778, 1.8423, -0.6462, -0.4383, -1.0465, 0.5721,
  6164. 0.1544],
  6165. [ 0.5356, -0.4343, 1.7441, -0.1150, -0.6462, -0.3614, 0.3238,
  6166. 0.0774],
  6167. [ 0.5488, -0.4221, 1.8018, -0.2459, -0.4845, -0.3075, 0.2309,
  6168. 0.3087],
  6169. [ 0.5427, -0.4035, 1.2688, -0.3675, -0.6808, -0.5461, 0.0959,
  6170. 0.2206],
  6171. [ 0.5443, -0.4079, 1.6633, 0.2853, -0.1325, 0.0888, 0.0558,
  6172. 0.2138],
  6173. [ 0.6128, -0.3828, 1.7499, -0.8386, -0.3344, -1.2620, 0.5792,
  6174. -0.0263]]], device='cuda:0')
  6175. loss_train_step before backward: tensor(0.0429, device='cuda:0', grad_fn=<MseLossBackward>)
  6176. loss_train_step after backward: tensor(0.0429, device='cuda:0', grad_fn=<MseLossBackward>)
  6177. loss_train: 3.8446308206766844
  6178. step: 46
  6179. running loss: 0.08357893088427575
  6180.  
  6181. Train Steps: 46/90 Loss: 0.0836 torch.Size([8, 600, 800])
  6182. torch.Size([8, 8])
  6183. tensor([[0.6364, 0.4165, 0.9088, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297],
  6184. [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575],
  6185. [0.6265, 0.4088, 0.8025, 0.1850, 0.4163, 0.2500, 0.6290, 0.4947],
  6186. [0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131],
  6187. [0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689],
  6188. [ nan, nan, 0.7412, 0.2200, 0.4450, 0.1517, 0.5312, 0.4983],
  6189. [0.6159, 0.4085, 0.6900, 0.2283, 0.4088, 0.1950, 0.5123, 0.5397],
  6190. [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220]],
  6191. device='cuda:0', dtype=torch.float64)
  6192. predictions are: tensor([[ 5.1204e-01, -4.0808e-01, 2.0241e+00, -6.4815e-04, -3.0531e-01,
  6193. -2.6063e-01, 6.0652e-01, 2.0491e-01],
  6194. [ 7.8274e-01, -2.8036e-01, 1.8523e+00, -4.5895e-01, -4.4856e-01,
  6195. 9.5163e-03, 7.9724e-01, 1.7100e-01],
  6196. [ 1.6438e-02, -7.3354e-01, 1.6209e+00, -8.2435e-01, -2.8295e-01,
  6197. -8.0990e-01, 6.1467e-01, 1.8458e-01],
  6198. [ 1.1100e+00, -7.7610e-02, 1.8684e+00, -1.1496e-01, -5.9891e-01,
  6199. 1.6606e-01, 4.7673e-01, 1.2712e-01],
  6200. [ 7.8423e-01, -2.5731e-01, 1.6101e+00, -3.6818e-01, -6.1474e-01,
  6201. -1.4655e-01, 3.7293e-01, 1.9924e-01],
  6202. [-1.0913e+00, -1.4849e+00, 1.1627e+00, -9.4262e-01, -2.5918e-01,
  6203. -1.1330e+00, 3.0192e-01, 2.3340e-01],
  6204. [-3.3309e-01, -9.9671e-01, 1.1426e+00, -8.3217e-01, -3.9860e-01,
  6205. -1.0915e+00, 2.8002e-01, 2.2096e-01],
  6206. [-2.7431e-01, -9.2364e-01, 1.1610e+00, -8.5328e-01, -3.1564e-01,
  6207. -1.1378e+00, 2.8685e-01, 2.2268e-01]], device='cuda:0',
  6208. grad_fn=<AddmmBackward>)
  6209. landmarks are: tensor([[[ 6.5365e-01, -3.6231e-01, 1.9115e+00, -2.6898e-01, -4.0370e-01,
  6210. -8.3095e-01, 6.9257e-01, 1.6077e-01],
  6211. [ 6.1577e-01, -4.2490e-01, 1.8654e+00, -9.0023e-01, -3.2286e-01,
  6212. -3.5366e-01, 9.6675e-01, 2.8902e-01],
  6213. [ 6.0785e-01, -3.9761e-01, 1.4208e+00, -1.4314e+00, -3.6328e-01,
  6214. -1.1312e+00, 6.1950e-01, -9.2270e-04],
  6215. [ 5.5813e-01, -4.5860e-01, 1.5586e+00, -3.7466e-01, -6.7920e-01,
  6216. -2.3907e-01, 4.4552e-01, 8.4044e-02],
  6217. [ 5.3274e-01, -4.3811e-01, 1.2880e+00, -6.3079e-01, -6.8661e-01,
  6218. -5.3072e-01, 2.6581e-01, 3.4174e-01],
  6219. [-2.2859e+00, -2.2859e+00, 1.1379e+00, -1.2697e+00, -2.3048e-01,
  6220. -1.5854e+00, 1.6790e-01, 1.5858e-02],
  6221. [ 5.5912e-01, -3.9900e-01, 9.0115e-01, -1.2313e+00, -3.9792e-01,
  6222. -1.3852e+00, 8.0445e-02, 2.0706e-01],
  6223. [ 5.5664e-01, -4.1601e-01, 9.9353e-01, -1.3313e+00, -2.8245e-01,
  6224. -1.5161e+00, 2.1441e-01, 1.2532e-01]]], device='cuda:0')
  6225. loss_train_step before backward: tensor(0.1320, device='cuda:0', grad_fn=<MseLossBackward>)
  6226. loss_train_step after backward: tensor(0.1320, device='cuda:0', grad_fn=<MseLossBackward>)
  6227. loss_train: 3.9766319822520018
  6228. step: 47
  6229. running loss: 0.08460919111174472
  6230. Train Steps: 47/90 Loss: 0.0846 torch.Size([8, 600, 800])
  6231. torch.Size([8, 8])
  6232. tensor([[0.6179, 0.4008, 0.7505, 0.2678, 0.4368, 0.1891, 0.5831, 0.5263],
  6233. [ nan, nan, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819],
  6234. [0.6250, 0.4054, 0.8770, 0.4723, 0.4662, 0.5367, 0.6162, 0.5433],
  6235. [0.6125, 0.3999, 0.8750, 0.4883, 0.4750, 0.4700, 0.5533, 0.5617],
  6236. [ nan, nan, 0.7335, 0.2569, 0.3788, 0.2667, 0.5066, 0.5578],
  6237. [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383],
  6238. [0.6263, 0.4039, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804],
  6239. [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5463, 0.5800]],
  6240. device='cuda:0', dtype=torch.float64)
  6241. predictions are: tensor([[ 0.6122, -0.3670, 1.3447, -0.9112, -0.3654, -1.0974, 0.4136, 0.1788],
  6242. [-0.8619, -1.2983, 1.1162, -0.8752, -0.3778, -1.0457, 0.2680, 0.2437],
  6243. [ 0.9006, -0.2098, 2.0488, 0.0939, -0.2921, 0.4136, 0.6445, 0.1981],
  6244. [ 0.8501, -0.2358, 1.9843, 0.0972, -0.2533, 0.0895, 0.5619, 0.2259],
  6245. [-1.4110, -1.6712, 1.2796, -0.7929, -0.3579, -0.9350, 0.3379, 0.2312],
  6246. [ 0.4885, -0.4263, 1.3314, -1.0387, -0.2906, -1.1694, 0.4948, 0.2056],
  6247. [ 0.9188, -0.1726, 2.0675, -0.0769, -0.5989, 0.1320, 0.6938, 0.0644],
  6248. [-0.4029, -0.9729, 1.3120, -0.6700, -0.4773, -0.8498, 0.2555, 0.2297]],
  6249. device='cuda:0', grad_fn=<AddmmBackward>)
  6250. landmarks are: tensor([[[ 0.5680, -0.4345, 1.1806, -1.0490, -0.2683, -1.4127, 0.4074,
  6251. 0.1449],
  6252. [-2.2859, -2.2859, 0.9438, -0.9967, -0.4614, -1.1851, 0.2468,
  6253. 0.4019],
  6254. [ 0.6009, -0.4135, 1.7651, -0.1043, -0.1323, 0.1929, 0.5605,
  6255. 0.2237],
  6256. [ 0.5432, -0.4388, 1.7557, -0.0303, -0.0919, -0.1150, 0.2699,
  6257. 0.3087],
  6258. [-2.2859, -2.2859, 1.1020, -1.0994, -0.5365, -1.0542, 0.0542,
  6259. 0.2905],
  6260. [ 0.5911, -0.3984, 1.1956, -1.0850, -0.0804, -1.5392, 0.4393,
  6261. 0.2006],
  6262. [ 0.6072, -0.4206, 1.8711, -0.2536, -0.6115, -0.1304, 0.6812,
  6263. -0.0670],
  6264. [ 0.5548, -0.3936, 1.1634, -0.8105, -0.5192, -1.0696, 0.2372,
  6265. 0.3931]]], device='cuda:0')
  6266. loss_train_step before backward: tensor(0.1134, device='cuda:0', grad_fn=<MseLossBackward>)
  6267. loss_train_step after backward: tensor(0.1134, device='cuda:0', grad_fn=<MseLossBackward>)
  6268. loss_train: 4.090075312182307
  6269. step: 48
  6270. running loss: 0.0852099023371314
  6271. Train Steps: 48/90 Loss: 0.0852 torch.Size([8, 600, 800])
  6272. torch.Size([8, 8])
  6273. tensor([[0.6175, 0.3957, 0.8700, 0.4817, 0.4662, 0.5133, 0.5800, 0.5517],
  6274. [0.6250, 0.4146, 0.8838, 0.3933, 0.3588, 0.4283, 0.6162, 0.5367],
  6275. [0.6128, 0.4022, 0.8738, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064],
  6276. [0.6143, 0.4040, 0.8237, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973],
  6277. [0.6164, 0.4102, 0.8850, 0.4867, 0.4213, 0.5983, 0.5609, 0.5038],
  6278. [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927],
  6279. [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092],
  6280. [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760]],
  6281. device='cuda:0', dtype=torch.float64)
  6282. predictions are: tensor([[-0.0881, -0.8622, 1.7180, -0.3380, -0.1900, -0.3595, 0.5118, 0.1946],
  6283. [ 0.3039, -0.5632, 1.6433, -0.7080, -0.5210, -0.6785, 0.4957, 0.2307],
  6284. [ 0.1802, -0.7043, 1.6741, -0.2930, -0.1235, -0.5044, 0.3889, 0.2316],
  6285. [-0.6809, -1.1939, 1.2687, -1.0639, -0.3153, -1.4568, 0.3238, 0.1441],
  6286. [ 0.3311, -0.5428, 1.7258, -0.2202, -0.2688, -0.0183, 0.4605, 0.2348],
  6287. [ 0.1304, -0.6649, 1.6293, -0.6603, -0.5442, -0.5658, 0.5650, 0.1014],
  6288. [ 0.3655, -0.5611, 1.7344, -0.1714, -0.4312, -0.2769, 0.4720, 0.1755],
  6289. [ 0.4916, -0.4244, 1.5928, -0.1345, -0.4315, -0.3216, 0.4903, 0.2304]],
  6290. device='cuda:0', grad_fn=<AddmmBackward>)
  6291. landmarks are: tensor([[[ 0.5662, -0.4581, 1.7326, -0.0611, -0.1323, 0.0851, 0.3931,
  6292. 0.2622],
  6293. [ 0.6009, -0.3710, 1.7961, -0.4691, -0.6289, -0.3075, 0.5605,
  6294. 0.1929],
  6295. [ 0.5446, -0.4280, 1.7499, 0.0543, 0.0156, 0.1301, 0.1918,
  6296. 0.0532],
  6297. [ 0.5515, -0.4201, 1.5189, -0.7463, -0.4037, -1.3082, 0.0869,
  6298. 0.0111],
  6299. [ 0.5614, -0.3912, 1.8018, -0.0380, -0.3402, 0.4778, 0.3049,
  6300. 0.0412],
  6301. [ 0.6010, -0.4345, 1.8480, -0.3537, -0.6462, -0.2613, 0.6524,
  6302. -0.0099],
  6303. [ 0.5783, -0.4363, 1.7724, 0.0432, -0.5153, 0.0871, 0.4840,
  6304. 0.0663],
  6305. [ 0.6421, -0.4008, 1.6055, 0.2160, -0.5076, -0.0534, 1.1020,
  6306. 0.3745]]], device='cuda:0')
  6307. loss_train_step before backward: tensor(0.0986, device='cuda:0', grad_fn=<MseLossBackward>)
  6308. loss_train_step after backward: tensor(0.0986, device='cuda:0', grad_fn=<MseLossBackward>)
  6309. loss_train: 4.188635913655162
  6310. step: 49
  6311. running loss: 0.08548236558479923
  6312. Train Steps: 49/90 Loss: 0.0855 torch.Size([8, 600, 800])
  6313. torch.Size([8, 8])
  6314. tensor([[0.6124, 0.4075, 0.7696, 0.4153, 0.3475, 0.3767, 0.5157, 0.5427],
  6315. [0.6126, 0.4073, 0.8750, 0.5133, 0.3800, 0.4333, 0.4986, 0.5378],
  6316. [0.6230, 0.4152, 0.7588, 0.2283, 0.4012, 0.2883, 0.6200, 0.5767],
  6317. [0.6022, 0.3994, 0.8025, 0.3350, 0.3350, 0.4400, 0.5565, 0.5025],
  6318. [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6038, 0.4833],
  6319. [0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400],
  6320. [0.6138, 0.4101, 0.8800, 0.5083, 0.4637, 0.5950, 0.5587, 0.5077],
  6321. [0.6163, 0.4114, 0.7650, 0.2017, 0.3763, 0.2867, 0.5631, 0.5071]],
  6322. device='cuda:0', dtype=torch.float64)
  6323. predictions are: tensor([[ 0.3454, -0.4971, 1.4509, -0.5696, -0.5079, -0.5487, 0.3331, 0.1901],
  6324. [ 0.3420, -0.5645, 1.7734, 0.0995, -0.4219, -0.2674, 0.3567, 0.2133],
  6325. [ 0.2438, -0.6110, 1.4486, -0.8155, -0.3154, -1.0944, 0.4986, 0.2272],
  6326. [-0.4310, -1.0179, 1.4841, -0.7465, -0.4986, -0.4605, 0.4811, 0.1436],
  6327. [ 0.0408, -0.7762, 1.8346, -0.0525, -0.3941, -0.1446, 0.5553, 0.1564],
  6328. [-0.1729, -0.8700, 1.3802, -0.9201, -0.1231, -1.1452, 0.4031, 0.2139],
  6329. [ 0.3208, -0.5864, 1.8330, -0.0026, -0.1255, 0.0109, 0.5079, 0.2647],
  6330. [ 0.1492, -0.6696, 1.2125, -1.0467, -0.4508, -1.1560, 0.3488, 0.1378]],
  6331. device='cuda:0', grad_fn=<AddmmBackward>)
  6332. landmarks are: tensor([[[ 0.5427, -0.4035, 1.2688, -0.3675, -0.6808, -0.5461, 0.0959,
  6333. 0.2206],
  6334. [ 0.5435, -0.4045, 1.7557, 0.0851, -0.5307, -0.2844, 0.0172,
  6335. 0.1980],
  6336. [ 0.5915, -0.3682, 1.2187, -1.2313, -0.4326, -0.9541, 0.5778,
  6337. 0.3777],
  6338. [ 0.4956, -0.4413, 1.4208, -0.7386, -0.7386, -0.2536, 0.2843,
  6339. 0.0351],
  6340. [ 0.5726, -0.4795, 1.7788, -0.0149, -0.5711, -0.0688, 0.5028,
  6341. -0.0534],
  6342. [ 0.5911, -0.3888, 1.4727, -0.9541, -0.0919, -1.4930, 0.3988,
  6343. 0.2083],
  6344. [ 0.5491, -0.3918, 1.7788, 0.0620, -0.1439, 0.4624, 0.2946,
  6345. 0.0592],
  6346. [ 0.5606, -0.3856, 1.2476, -1.3544, -0.5480, -0.9618, 0.3152,
  6347. 0.0562]]], device='cuda:0')
  6348. loss_train_step before backward: tensor(0.0689, device='cuda:0', grad_fn=<MseLossBackward>)
  6349.  
  6350. loss_train_step after backward: tensor(0.0689, device='cuda:0', grad_fn=<MseLossBackward>)
  6351. loss_train: 4.257585039362311
  6352. step: 50
  6353. running loss: 0.08515170078724622
  6354. Train Steps: 50/90 Loss: 0.0852 torch.Size([8, 600, 800])
  6355. torch.Size([8, 8])
  6356. tensor([[ nan, nan, 0.7648, 0.2722, 0.3962, 0.2183, 0.5060, 0.5422],
  6357. [0.6266, 0.4070, 0.8712, 0.5600, 0.3713, 0.4783, 0.5775, 0.6100],
  6358. [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212],
  6359. [0.6205, 0.4062, 0.8337, 0.2683, 0.3675, 0.4283, 0.6338, 0.5250],
  6360. [0.6251, 0.4163, 0.8662, 0.4467, 0.3625, 0.3567, 0.6038, 0.5533],
  6361. [0.6262, 0.4052, 0.8888, 0.4700, 0.3675, 0.5117, 0.6350, 0.5233],
  6362. [0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373],
  6363. [0.6159, 0.4085, 0.6900, 0.2283, 0.4088, 0.1950, 0.5123, 0.5397]],
  6364. device='cuda:0', dtype=torch.float64)
  6365. predictions are: tensor([[-1.4041, -1.6769, 1.1390, -0.9196, -0.2693, -1.1782, 0.2023, 0.2169],
  6366. [ 0.6574, -0.3363, 1.7427, 0.2147, -0.4140, 0.0338, 0.3964, 0.2300],
  6367. [ 0.1212, -0.6699, 1.4240, -1.0007, -0.3117, -1.0039, 0.5159, 0.1357],
  6368. [ 0.3221, -0.5317, 1.4546, -0.9382, -0.4403, -0.4426, 0.5357, 0.1844],
  6369. [ 0.6383, -0.3453, 1.7350, -0.3782, -0.4291, -0.7050, 0.3973, 0.1939],
  6370. [ 0.6196, -0.4027, 1.9948, -0.1199, -0.3932, 0.0846, 0.6095, 0.1669],
  6371. [ 0.4998, -0.4693, 1.7146, 0.2622, -0.3820, -0.1580, 0.3382, 0.1745],
  6372. [-0.3319, -0.9927, 1.0219, -1.1324, -0.3122, -1.3981, 0.2109, 0.1964]],
  6373. device='cuda:0', grad_fn=<AddmmBackward>)
  6374. landmarks are: tensor([[[-2.2859, -2.2859, 1.2469, -1.0288, -0.4557, -1.2774, 0.0511,
  6375. 0.2183],
  6376. [ 0.6081, -0.4059, 1.7383, 0.3007, -0.5711, -0.0765, 0.3815,
  6377. 0.5316],
  6378. [ 0.6125, -0.4345, 1.4308, -1.1384, -0.4213, -1.0031, 0.7190,
  6379. 0.1214],
  6380. [ 0.5799, -0.4099, 1.5651, -1.0465, -0.5885, -0.3075, 0.6413,
  6381. 0.1390],
  6382. [ 0.6016, -0.3633, 1.7152, -0.2228, -0.6115, -0.6385, 0.5028,
  6383. 0.2699],
  6384. [ 0.6063, -0.4142, 1.8192, -0.1150, -0.5885, 0.0774, 0.6471,
  6385. 0.1313],
  6386. [ 0.5680, -0.4562, 1.5697, 0.4947, -0.4904, -0.1503, 0.3536,
  6387. 0.1956],
  6388. [ 0.5591, -0.3990, 0.9012, -1.2313, -0.3979, -1.3852, 0.0804,
  6389. 0.2071]]], device='cuda:0')
  6390. loss_train_step before backward: tensor(0.0524, device='cuda:0', grad_fn=<MseLossBackward>)
  6391. loss_train_step after backward: tensor(0.0524, device='cuda:0', grad_fn=<MseLossBackward>)
  6392. loss_train: 4.309938298538327
  6393. step: 51
  6394. running loss: 0.08450859408898681
  6395. Train Steps: 51/90 Loss: 0.0845 torch.Size([8, 600, 800])
  6396. torch.Size([8, 8])
  6397. tensor([[0.6204, 0.4007, 0.7838, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167],
  6398. [0.6236, 0.4084, 0.7738, 0.2133, 0.3663, 0.3233, 0.5813, 0.5567],
  6399. [0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938],
  6400. [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153],
  6401. [0.6250, 0.4131, 0.8688, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367],
  6402. [0.6204, 0.4110, 0.7913, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367],
  6403. [ nan, nan, 0.6793, 0.2110, 0.4012, 0.2167, 0.5112, 0.5583],
  6404. [0.6127, 0.4066, 0.8550, 0.5567, 0.4662, 0.5141, 0.5070, 0.5412]],
  6405. device='cuda:0', dtype=torch.float64)
  6406. predictions are: tensor([[ 0.1343, -0.6719, 1.1848, -0.9330, -0.2622, -1.1929, 0.2911, 0.1972],
  6407. [ 0.1454, -0.6445, 1.3932, -0.7919, -0.4341, -0.7242, 0.3274, 0.2350],
  6408. [ 0.6946, -0.3061, 1.8101, 0.3960, -0.4587, 0.0493, 0.4624, 0.1293],
  6409. [ 0.3389, -0.5140, 1.4280, -0.7080, -0.5626, -0.5696, 0.3538, 0.1557],
  6410. [ 0.1701, -0.6353, 1.6288, -0.7426, -0.2407, -0.7913, 0.5828, 0.1906],
  6411. [ 0.6218, -0.3358, 1.3809, -0.8998, -0.4000, -0.9331, 0.4869, 0.1414],
  6412. [-1.3165, -1.6054, 1.0127, -1.0492, -0.3465, -1.1461, 0.2269, 0.2403],
  6413. [ 0.4412, -0.4863, 1.7145, 0.4136, -0.2517, 0.1877, 0.3192, 0.2372]],
  6414. device='cuda:0', grad_fn=<AddmmBackward>)
  6415. landmarks are: tensor([[[ 0.5796, -0.4351, 1.3342, -1.3159, -0.2189, -1.4853, 0.4046,
  6416. 0.1005],
  6417. [ 0.5946, -0.3995, 1.2880, -1.3005, -0.5942, -0.7925, 0.3988,
  6418. 0.2853],
  6419. [ 0.6395, -0.4213, 1.7037, 0.3623, -0.4326, -0.1073, 0.6560,
  6420. -0.0049],
  6421. [ 0.5614, -0.4514, 1.3252, -0.7240, -0.6237, -0.7215, 0.4299,
  6422. 0.0942],
  6423. [ 0.6011, -0.3778, 1.7268, -0.9079, -0.3113, -1.1928, 0.5605,
  6424. 0.1929],
  6425. [ 0.5796, -0.3878, 1.3688, -1.0542, -0.4095, -1.1312, 0.5894,
  6426. 0.1929],
  6427. [-2.2859, -2.2859, 0.8516, -1.3112, -0.4326, -1.2851, 0.0755,
  6428. 0.2930],
  6429. [ 0.5443, -0.4079, 1.6633, 0.2853, -0.1325, 0.0888, 0.0558,
  6430. 0.2138]]], device='cuda:0')
  6431. loss_train_step before backward: tensor(0.0551, device='cuda:0', grad_fn=<MseLossBackward>)
  6432. loss_train_step after backward: tensor(0.0551, device='cuda:0', grad_fn=<MseLossBackward>)
  6433. loss_train: 4.365078965201974
  6434. step: 52
  6435. running loss: 0.08394382625388411
  6436. Train Steps: 52/90 Loss: 0.0839 torch.Size([8, 600, 800])
  6437. torch.Size([8, 8])
  6438. tensor([[0.6152, 0.4131, 0.6863, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305],
  6439. [0.6364, 0.4154, 0.8938, 0.3717, 0.4500, 0.2583, 0.6448, 0.5285],
  6440. [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773],
  6441. [ nan, nan, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733],
  6442. [0.6164, 0.4102, 0.8850, 0.4867, 0.4213, 0.5983, 0.5609, 0.5038],
  6443. [0.6299, 0.4008, 0.8450, 0.5350, 0.4213, 0.5000, 0.6350, 0.5100],
  6444. [0.6284, 0.4093, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167],
  6445. [0.6307, 0.4029, 0.8988, 0.4817, 0.3937, 0.3500, 0.7311, 0.5378]],
  6446. device='cuda:0', dtype=torch.float64)
  6447. predictions are: tensor([[ 0.7222, -0.2672, 0.9825, -1.1078, -0.5705, -0.9164, 0.1739, 0.2092],
  6448. [ 0.0665, -0.7308, 1.5910, -0.8096, -0.2316, -1.0253, 0.5274, 0.2099],
  6449. [ 0.3037, -0.5983, 1.5746, -0.1910, -0.2327, -0.1118, 0.2617, 0.2293],
  6450. [-1.2385, -1.5823, 0.9541, -1.2474, -0.2575, -1.4829, 0.2528, 0.2454],
  6451. [ 0.3653, -0.5106, 1.7113, -0.0647, -0.3102, 0.2423, 0.4094, 0.1881],
  6452. [ 0.4853, -0.4565, 1.5947, 0.0082, -0.4378, -0.0843, 0.3777, 0.1922],
  6453. [ 0.8494, -0.2006, 1.7019, -0.1805, -0.6135, -0.5201, 0.4251, 0.0953],
  6454. [ 0.1864, -0.6458, 1.7229, -0.2324, -0.4180, -0.7384, 0.4583, 0.1592]],
  6455. device='cuda:0', grad_fn=<AddmmBackward>)
  6456. landmarks are: tensor([[[ 0.5557, -0.3779, 0.8838, -1.1004, -0.6115, -0.7617, 0.3769,
  6457. 0.1644],
  6458. [ 0.6537, -0.3671, 1.8423, -0.5692, -0.2074, -1.0927, 0.6926,
  6459. 0.1554],
  6460. [ 0.5428, -0.4244, 1.7095, -0.0380, 0.0232, 0.0725, 0.0866,
  6461. 0.3806],
  6462. [-2.2859, -2.2859, 0.7106, -1.4468, -0.2882, -1.4237, 0.2430,
  6463. 0.3623],
  6464. [ 0.5614, -0.3912, 1.8018, -0.0380, -0.3402, 0.4778, 0.3049,
  6465. 0.0412],
  6466. [ 0.6236, -0.4344, 1.6171, 0.1852, -0.3402, 0.0236, 0.6471,
  6467. 0.0697],
  6468. [ 0.6164, -0.3956, 1.8249, -0.1150, -0.6000, -0.5076, 0.5836,
  6469. 0.1005],
  6470. [ 0.6273, -0.4249, 1.8654, -0.0611, -0.4672, -0.6693, 1.0910,
  6471. 0.1982]]], device='cuda:0')
  6472. loss_train_step before backward: tensor(0.0608, device='cuda:0', grad_fn=<MseLossBackward>)
  6473. loss_train_step after backward: tensor(0.0608, device='cuda:0', grad_fn=<MseLossBackward>)
  6474. loss_train: 4.425908403471112
  6475. step: 53
  6476. running loss: 0.08350770572587005
  6477. Train Steps: 53/90 Loss: 0.0835 torch.Size([8, 600, 800])
  6478. torch.Size([8, 8])
  6479. tensor([[0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284],
  6480. [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600],
  6481. [0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400],
  6482. [0.6176, 0.4030, 0.8850, 0.4850, 0.3688, 0.4050, 0.5312, 0.5783],
  6483. [0.6264, 0.4049, 0.8988, 0.4633, 0.3813, 0.4983, 0.6326, 0.4843],
  6484. [0.6109, 0.4015, 0.7668, 0.3639, 0.3513, 0.3667, 0.5200, 0.5641],
  6485. [ nan, nan, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729],
  6486. [0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6138, 0.5450]],
  6487. device='cuda:0', dtype=torch.float64)
  6488. predictions are: tensor([[ 0.5744, -0.4358, 1.1659, -1.2056, -0.2736, -1.3638, 0.3496, 0.1837],
  6489. [ 0.7267, -0.3071, 1.6715, 0.2827, -0.4446, -0.0200, 0.4095, 0.1985],
  6490. [ 0.3578, -0.5564, 1.3084, -1.0304, -0.2044, -1.2064, 0.3487, 0.2020],
  6491. [ 0.2623, -0.5900, 1.6264, 0.0475, -0.5414, -0.3029, 0.2797, 0.1594],
  6492. [ 0.3787, -0.5423, 1.8219, -0.1248, -0.5105, -0.0737, 0.4833, 0.1497],
  6493. [ 0.3219, -0.5242, 1.3093, -0.7075, -0.5767, -0.5406, 0.2827, 0.1780],
  6494. [-0.7890, -1.2903, 0.9590, -1.2208, -0.3849, -1.4650, 0.2070, 0.1993],
  6495. [ 0.2771, -0.6158, 1.7014, -0.1671, -0.3850, 0.1569, 0.4584, 0.2011]],
  6496. device='cuda:0', grad_fn=<AddmmBackward>)
  6497. landmarks are: tensor([[[ 0.5761, -0.4070, 1.2452, -1.2541, -0.1725, -1.4835, 0.4511,
  6498. 0.1545],
  6499. [ 0.5883, -0.3594, 1.7557, 0.2545, -0.4152, -0.0611, 0.3353,
  6500. 0.3007],
  6501. [ 0.5911, -0.3888, 1.4727, -0.9541, -0.0919, -1.4930, 0.3988,
  6502. 0.2083],
  6503. [ 0.5667, -0.4246, 1.8018, -0.0457, -0.5827, -0.4152, 0.1679,
  6504. 0.3854],
  6505. [ 0.6075, -0.4159, 1.8654, -0.1458, -0.5249, 0.0159, 0.6359,
  6506. -0.0490],
  6507. [ 0.5357, -0.4315, 1.2557, -0.6051, -0.6635, -0.5923, 0.1159,
  6508. 0.3198],
  6509. [-2.2859, -2.2859, 0.8824, -1.2727, -0.3691, -1.4237, 0.0943,
  6510. 0.3604],
  6511. [ 0.5960, -0.4102, 1.8018, -0.1612, -0.3344, 0.1159, 0.5490,
  6512. 0.2314]]], device='cuda:0')
  6513. loss_train_step before backward: tensor(0.0666, device='cuda:0', grad_fn=<MseLossBackward>)
  6514. loss_train_step after backward: tensor(0.0666, device='cuda:0', grad_fn=<MseLossBackward>)
  6515. loss_train: 4.492464052513242
  6516. step: 54
  6517. running loss: 0.08319377875024522
  6518.  
  6519. Train Steps: 54/90 Loss: 0.0832 torch.Size([8, 600, 800])
  6520. torch.Size([8, 8])
  6521. tensor([[0.6214, 0.4112, 0.7838, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083],
  6522. [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667],
  6523. [0.6212, 0.4159, 0.8675, 0.5783, 0.4088, 0.4317, 0.5613, 0.5917],
  6524. [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617],
  6525. [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378],
  6526. [ nan, nan, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737],
  6527. [0.6185, 0.4067, 0.8838, 0.4450, 0.4037, 0.4733, 0.5213, 0.5142],
  6528. [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5513, 0.5750]],
  6529. device='cuda:0', dtype=torch.float64)
  6530. predictions are: tensor([[ 0.9089, -0.1808, 1.2733, -1.0337, -0.5515, -1.1081, 0.3434, 0.1587],
  6531. [ 0.8128, -0.2403, 1.4522, -0.6860, -0.6329, -0.7484, 0.2570, 0.1964],
  6532. [ 1.2244, 0.0209, 1.4750, 0.2158, -0.4895, -0.2091, 0.2452, 0.2366],
  6533. [ 0.7003, -0.3044, 1.5976, -0.0632, -0.4107, 0.0459, 0.3074, 0.2059],
  6534. [ 0.8016, -0.2368, 1.7200, -0.0157, -0.4262, 0.1961, 0.5880, 0.1784],
  6535. [-2.0187, -2.1144, 1.5037, -1.0512, -0.0099, -1.3094, 0.7017, 0.2278],
  6536. [ 0.6363, -0.3475, 1.5620, -0.2526, -0.5205, -0.2222, 0.1970, 0.1762],
  6537. [-0.8353, -1.2941, 0.8914, -1.2110, -0.3824, -1.3947, 0.1707, 0.2402]],
  6538. device='cuda:0', grad_fn=<AddmmBackward>)
  6539. landmarks are: tensor([[[ 0.5845, -0.3864, 1.3342, -1.3082, -0.6000, -0.8386, 0.3353,
  6540. 0.0620],
  6541. [ 0.5768, -0.3857, 1.5305, -0.7694, -0.6462, -0.6308, 0.3988,
  6542. 0.3315],
  6543. [ 0.5836, -0.3649, 1.7210, 0.3854, -0.3979, -0.2921, 0.3065,
  6544. 0.4470],
  6545. [ 0.5637, -0.4143, 1.7519, -0.0787, -0.3055, -0.0149, 0.3758,
  6546. 0.3084],
  6547. [ 0.6421, -0.3864, 1.7961, 0.0543, -0.4383, 0.2237, 1.2007,
  6548. 0.1982],
  6549. [-2.2859, -2.2859, 1.6979, -1.1081, 0.2487, -1.2697, 1.1020,
  6550. 0.3638],
  6551. [ 0.5708, -0.4075, 1.7961, -0.2305, -0.4210, -0.0996, 0.1219,
  6552. 0.0893],
  6553. [ 0.5663, -0.4396, 0.8261, -1.1312, -0.2940, -1.3929, 0.2603,
  6554. 0.3700]]], device='cuda:0')
  6555. loss_train_step before backward: tensor(0.0768, device='cuda:0', grad_fn=<MseLossBackward>)
  6556. loss_train_step after backward: tensor(0.0768, device='cuda:0', grad_fn=<MseLossBackward>)
  6557. loss_train: 4.569240244105458
  6558. step: 55
  6559. running loss: 0.08307709534737197
  6560. Train Steps: 55/90 Loss: 0.0831 torch.Size([8, 600, 800])
  6561. torch.Size([8, 8])
  6562. tensor([[0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6262, 0.5167],
  6563. [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329],
  6564. [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283],
  6565. [ nan, nan, 0.7512, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617],
  6566. [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810],
  6567. [0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933],
  6568. [0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938],
  6569. [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672]],
  6570. device='cuda:0', dtype=torch.float64)
  6571. predictions are: tensor([[ 0.4381, -0.5051, 1.6501, -0.5635, -0.4440, -0.7391, 0.4992, 0.2375],
  6572. [ 0.8613, -0.2273, 1.4765, -0.8301, -0.5814, -0.6689, 0.3431, 0.1861],
  6573. [ 0.5253, -0.4384, 1.5418, -0.5003, -0.5351, -0.5880, 0.3159, 0.2336],
  6574. [-1.3941, -1.6921, 0.9840, -1.2563, -0.2956, -1.3670, 0.2403, 0.2752],
  6575. [ 0.2707, -0.6181, 1.3609, -1.0859, -0.4104, -1.0531, 0.4897, 0.1458],
  6576. [ 0.4641, -0.4778, 1.4754, 0.0333, -0.1887, -0.2335, 0.3023, 0.2703],
  6577. [ 0.3964, -0.5099, 1.5877, 0.1125, -0.4331, -0.1267, 0.4035, 0.1513],
  6578. [ 1.0365, -0.1044, 1.5411, -0.0426, -0.5323, -0.1892, 0.3891, 0.1199]],
  6579. device='cuda:0', grad_fn=<AddmmBackward>)
  6580. landmarks are: tensor([[[ 0.6094, -0.3947, 1.8885, -0.2998, -0.5769, -0.6770, 0.6067,
  6581. 0.1005],
  6582. [ 0.5680, -0.4393, 1.5920, -0.6672, -0.6453, -0.5457, 0.5149,
  6583. 0.1753],
  6584. [ 0.5880, -0.3887, 1.8423, -0.3306, -0.6231, -0.5230, 0.4046,
  6585. 0.1544],
  6586. [-2.2859, -2.2859, 1.1841, -1.3082, -0.3055, -1.3621, 0.3007,
  6587. 0.3084],
  6588. [ 0.6064, -0.3990, 1.6113, -0.8309, -0.4268, -1.0696, 0.6421,
  6589. -0.0640],
  6590. [ 0.5959, -0.3579, 1.6055, 0.3623, -0.0573, -0.2074, 0.3122,
  6591. 0.4547],
  6592. [ 0.6395, -0.4213, 1.7037, 0.3623, -0.4326, -0.1073, 0.6560,
  6593. -0.0049],
  6594. [ 0.6128, -0.4375, 1.7788, 0.2699, -0.6346, -0.2536, 0.5463,
  6595. -0.1278]]], device='cuda:0')
  6596. loss_train_step before backward: tensor(0.0489, device='cuda:0', grad_fn=<MseLossBackward>)
  6597. loss_train_step after backward: tensor(0.0489, device='cuda:0', grad_fn=<MseLossBackward>)
  6598. loss_train: 4.618122825399041
  6599. step: 56
  6600. running loss: 0.08246647902498287
  6601. Train Steps: 56/90 Loss: 0.0825 torch.Size([8, 600, 800])
  6602. torch.Size([8, 8])
  6603. tensor([[0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090],
  6604. [0.6038, 0.3946, 0.8413, 0.4883, 0.3563, 0.4550, 0.5266, 0.4693],
  6605. [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167],
  6606. [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869],
  6607. [0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804],
  6608. [0.6166, 0.4008, 0.8563, 0.5667, 0.4388, 0.4933, 0.5575, 0.5567],
  6609. [0.6212, 0.4159, 0.8675, 0.5783, 0.4088, 0.4317, 0.5613, 0.5917],
  6610. [0.6339, 0.4102, 0.9088, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390]],
  6611. device='cuda:0', dtype=torch.float64)
  6612. predictions are: tensor([[ 0.1561, -0.7037, 1.4751, -0.8651, -0.4711, -1.0469, 0.5206, 0.1341],
  6613. [ 0.5377, -0.4234, 1.5176, -0.2102, -0.5475, -0.2934, 0.2895, 0.1455],
  6614. [ 0.2270, -0.6552, 1.1592, -1.1152, -0.4722, -1.0442, 0.3479, 0.2583],
  6615. [ 0.4311, -0.5644, 1.5158, -1.1601, -0.2336, -1.4870, 0.4191, 0.1780],
  6616. [ 0.3855, -0.5671, 1.5510, -0.7643, -0.5894, -0.6581, 0.5060, 0.1097],
  6617. [ 0.1805, -0.6948, 1.5721, 0.1061, -0.2787, -0.1696, 0.3229, 0.2565],
  6618. [ 0.5569, -0.4505, 1.4991, 0.0849, -0.4491, -0.2875, 0.2419, 0.2883],
  6619. [ 0.1861, -0.6641, 1.6742, -0.1751, -0.4272, 0.1524, 0.5527, 0.2295]],
  6620. device='cuda:0', grad_fn=<AddmmBackward>)
  6621. landmarks are: tensor([[[ 0.6165, -0.4249, 1.7961, -0.5384, -0.4499, -0.9695, 0.6401,
  6622. 0.0652],
  6623. [ 0.5030, -0.4631, 1.5998, -0.0303, -0.6404, -0.1843, 0.1465,
  6624. -0.1181],
  6625. [ 0.5631, -0.4129, 1.2129, -0.9233, -0.4152, -1.0311, 0.4566,
  6626. 0.5624],
  6627. [ 0.6119, -0.3927, 1.6979, -0.7925, -0.2536, -1.3698, 0.4282,
  6628. -0.0368],
  6629. [ 0.6071, -0.4119, 1.7788, -0.5153, -0.6000, -0.5692, 0.6586,
  6630. -0.0670],
  6631. [ 0.5620, -0.4346, 1.6691, 0.3315, -0.2594, -0.0072, 0.2891,
  6632. 0.2853],
  6633. [ 0.5836, -0.3649, 1.7210, 0.3854, -0.3979, -0.2921, 0.3065,
  6634. 0.4470],
  6635. [ 0.6421, -0.3912, 1.9115, -0.0842, -0.4730, 0.1544, 1.1824,
  6636. 0.2035]]], device='cuda:0')
  6637. loss_train_step before backward: tensor(0.0432, device='cuda:0', grad_fn=<MseLossBackward>)
  6638. loss_train_step after backward: tensor(0.0432, device='cuda:0', grad_fn=<MseLossBackward>)
  6639. loss_train: 4.661323884502053
  6640. step: 57
  6641. running loss: 0.08177761200880795
  6642. Train Steps: 57/90 Loss: 0.0818 torch.Size([8, 600, 800])
  6643. torch.Size([8, 8])
  6644. tensor([[0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550],
  6645. [0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367],
  6646. [0.6201, 0.4017, 0.8871, 0.4621, 0.3517, 0.4675, 0.5999, 0.5106],
  6647. [0.6131, 0.4064, 0.8638, 0.5200, 0.4788, 0.4783, 0.5258, 0.5867],
  6648. [0.6179, 0.4040, 0.7412, 0.1850, 0.3825, 0.2783, 0.5837, 0.5600],
  6649. [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967],
  6650. [0.6332, 0.4118, 0.9238, 0.4267, 0.4012, 0.4733, 0.7525, 0.5436],
  6651. [0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433]],
  6652. device='cuda:0', dtype=torch.float64)
  6653. predictions are: tensor([[ 0.0544, -0.7835, 1.6998, -0.2782, -0.1604, -0.2922, 0.4850, 0.1874],
  6654. [ 0.4879, -0.4952, 1.6149, -0.6312, -0.3887, -1.1637, 0.4164, 0.2082],
  6655. [ 0.2649, -0.6053, 1.6451, -0.2102, -0.5783, -0.1651, 0.4630, 0.1315],
  6656. [ 0.2480, -0.6386, 1.6026, 0.0719, -0.2249, -0.1325, 0.3098, 0.2759],
  6657. [ 0.1634, -0.7327, 1.2079, -1.2229, -0.4614, -1.3717, 0.3322, 0.2116],
  6658. [ 1.0088, -0.1769, 1.6499, -0.2045, -0.4546, -0.9317, 0.4750, 0.1453],
  6659. [ 0.2639, -0.6029, 1.6201, -0.3808, -0.5056, -0.0960, 0.6122, 0.1866],
  6660. [ 0.3316, -0.5745, 1.2158, -1.0627, -0.6265, -0.7776, 0.3601, 0.1813]],
  6661. device='cuda:0', grad_fn=<AddmmBackward>)
  6662. landmarks are: tensor([[[ 0.5351, -0.4321, 1.7326, -0.1381, 0.0463, 0.1049, 0.4162,
  6663. 0.2776],
  6664. [ 0.5900, -0.3932, 1.8307, -0.3921, -0.4268, -1.1851, 0.3758,
  6665. 0.1929],
  6666. [ 0.5783, -0.4306, 1.8114, -0.1515, -0.6617, -0.1268, 0.4851,
  6667. 0.0727],
  6668. [ 0.5462, -0.4090, 1.7037, 0.1159, -0.0746, -0.0765, 0.1427,
  6669. 0.4239],
  6670. [ 0.5680, -0.4201, 1.1379, -1.4314, -0.5192, -1.0003, 0.4104,
  6671. 0.3007],
  6672. [ 0.6054, -0.3767, 1.7788, 0.0774, -0.5711, -0.7694, 0.5887,
  6673. 0.0081],
  6674. [ 0.6388, -0.3840, 1.9808, -0.3152, -0.4326, -0.0996, 1.1897,
  6675. 0.2249],
  6676. [ 0.5665, -0.3963, 1.3457, -0.9926, -0.6058, -0.6308, 0.5721,
  6677. 0.2237]]], device='cuda:0')
  6678. loss_train_step before backward: tensor(0.0437, device='cuda:0', grad_fn=<MseLossBackward>)
  6679. loss_train_step after backward: tensor(0.0437, device='cuda:0', grad_fn=<MseLossBackward>)
  6680. loss_train: 4.704992236569524
  6681. step: 58
  6682. running loss: 0.08112055580292282
  6683.  
  6684. Train Steps: 58/90 Loss: 0.0811 torch.Size([8, 600, 800])
  6685. torch.Size([8, 8])
  6686. tensor([[0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389],
  6687. [0.6179, 0.4040, 0.7412, 0.1850, 0.3825, 0.2783, 0.5837, 0.5600],
  6688. [ nan, nan, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533],
  6689. [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5637, 0.5633],
  6690. [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364],
  6691. [ nan, nan, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702],
  6692. [ nan, nan, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621],
  6693. [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332]],
  6694. device='cuda:0', dtype=torch.float64)
  6695. predictions are: tensor([[ 0.6725, -0.3250, 1.3117, -0.8996, -0.4375, -0.9283, 0.3662, 0.2144],
  6696. [ 0.8203, -0.2586, 1.2539, -0.8671, -0.4885, -0.8802, 0.2916, 0.2076],
  6697. [-0.6151, -1.1069, 1.2332, -0.7149, -0.5765, -0.8060, 0.1702, 0.2368],
  6698. [ 1.7519, 0.3353, 1.6793, 0.2937, -0.6515, -0.0736, 0.3457, 0.1722],
  6699. [ 0.9992, -0.1174, 1.3909, -0.8545, -0.3347, -0.9903, 0.4672, 0.1633],
  6700. [-0.7490, -1.2308, 1.8226, -0.4618, -0.1372, -0.7754, 0.7525, 0.2115],
  6701. [-1.7876, -1.9232, 1.5286, -0.8708, -0.1098, -0.9816, 0.6602, 0.2375],
  6702. [ 1.5744, 0.1914, 1.7823, 0.1854, -0.5948, 0.1148, 0.5642, 0.1143]],
  6703. device='cuda:0', grad_fn=<AddmmBackward>)
  6704. landmarks are: tensor([[[ 0.5697, -0.4466, 1.1973, -1.1871, -0.4571, -0.9965, 0.5219,
  6705. 0.2032],
  6706. [ 0.5680, -0.4201, 1.1379, -1.4314, -0.5192, -1.0003, 0.4104,
  6707. 0.3007],
  6708. [-2.2859, -2.2859, 1.5074, -1.0388, -0.5423, -0.9849, 0.2199,
  6709. 0.2699],
  6710. [ 0.5902, -0.3493, 1.7961, -0.0072, -0.5942, -0.5615, 0.3180,
  6711. 0.3161],
  6712. [ 0.6158, -0.3960, 1.4092, -1.2774, -0.2074, -1.1851, 0.8491,
  6713. 0.1917],
  6714. [-2.2859, -2.2859, 1.8942, -0.6693, 0.0871, -1.2236, 1.1130,
  6715. 0.3478],
  6716. [-2.2859, -2.2859, 1.7210, -0.9772, 0.1852, -1.3698, 0.7986,
  6717. 0.3104],
  6718. [ 0.6520, -0.4056, 1.9173, -0.0765, -0.5596, -0.4537, 0.7949,
  6719. 0.1768]]], device='cuda:0')
  6720. loss_train_step before backward: tensor(0.2201, device='cuda:0', grad_fn=<MseLossBackward>)
  6721. loss_train_step after backward: tensor(0.2201, device='cuda:0', grad_fn=<MseLossBackward>)
  6722. loss_train: 4.925117092207074
  6723. step: 59
  6724. running loss: 0.08347656088486566
  6725. Train Steps: 59/90 Loss: 0.0835 torch.Size([8, 600, 800])
  6726. torch.Size([8, 8])
  6727. tensor([[0.6265, 0.4091, 0.8950, 0.3533, 0.3600, 0.3967, 0.6295, 0.4901],
  6728. [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791],
  6729. [0.6271, 0.4040, 0.9138, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413],
  6730. [0.6219, 0.4114, 0.8175, 0.2817, 0.3925, 0.2783, 0.5900, 0.5350],
  6731. [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649],
  6732. [0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183],
  6733. [0.6134, 0.4090, 0.6926, 0.2819, 0.3538, 0.3233, 0.5563, 0.5667],
  6734. [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750]],
  6735. device='cuda:0', dtype=torch.float64)
  6736. predictions are: tensor([[ 0.4033, -0.5310, 1.8569, -0.6054, -0.5117, -0.6812, 0.6229, 0.1313],
  6737. [ 0.4407, -0.5129, 1.8577, 0.1828, -0.4270, 0.1355, 0.5361, 0.1932],
  6738. [-0.1968, -0.9619, 1.9427, -0.6664, -0.1138, -1.0862, 0.8384, 0.1491],
  6739. [-0.0191, -0.8111, 1.3862, -1.2234, -0.2565, -1.2246, 0.4617, 0.2555],
  6740. [ 0.8325, -0.2923, 1.7349, 0.3951, -0.4775, -0.0507, 0.5048, 0.1812],
  6741. [ 0.3294, -0.5901, 1.8064, 0.0931, -0.5397, -0.2080, 0.5408, 0.1618],
  6742. [ 0.2896, -0.6054, 1.0851, -0.9982, -0.5647, -0.8826, 0.1846, 0.2427],
  6743. [ 0.8394, -0.2579, 1.1610, -0.9770, -0.4822, -1.0040, 0.3513, 0.2346]],
  6744. device='cuda:0', grad_fn=<AddmmBackward>)
  6745. landmarks are: tensor([[[ 0.6077, -0.3965, 1.8480, -0.6539, -0.6231, -0.4537, 0.6216,
  6746. -0.0220],
  6747. [ 0.6074, -0.4223, 1.8192, -0.0303, -0.4152, 0.1236, 0.6524,
  6748. -0.0731],
  6749. [ 0.6108, -0.4201, 1.9346, -0.5538, -0.1497, -1.0773, 1.0545,
  6750. 0.2142],
  6751. [ 0.5868, -0.3858, 1.4901, -0.9849, -0.4730, -1.0003, 0.4393,
  6752. 0.1852],
  6753. [ 0.6105, -0.4360, 1.6171, 0.5162, -0.5076, 0.0159, 0.5171,
  6754. -0.1385],
  6755. [ 0.5944, -0.4543, 1.8018, 0.0082, -0.6058, -0.4306, 0.4162,
  6756. 0.1082],
  6757. [ 0.5473, -0.3966, 0.9131, -0.9838, -0.6520, -0.7925, 0.2834,
  6758. 0.3315],
  6759. [ 0.5805, -0.3818, 1.0282, -1.1774, -0.4903, -0.9310, 0.5894,
  6760. 0.3700]]], device='cuda:0')
  6761. loss_train_step before backward: tensor(0.0440, device='cuda:0', grad_fn=<MseLossBackward>)
  6762. loss_train_step after backward: tensor(0.0440, device='cuda:0', grad_fn=<MseLossBackward>)
  6763. loss_train: 4.969099497422576
  6764. step: 60
  6765. running loss: 0.08281832495704293
  6766. Train Steps: 60/90 Loss: 0.0828 torch.Size([8, 600, 800])
  6767. torch.Size([8, 8])
  6768. tensor([[0.6229, 0.4198, 0.7662, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783],
  6769. [0.6153, 0.4117, 0.8688, 0.5167, 0.4895, 0.5647, 0.5524, 0.5136],
  6770. [0.6219, 0.4114, 0.8175, 0.2817, 0.3925, 0.2783, 0.5900, 0.5350],
  6771. [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567],
  6772. [0.6192, 0.4128, 0.8513, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633],
  6773. [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650],
  6774. [0.6129, 0.4069, 0.8750, 0.5067, 0.3875, 0.4233, 0.5235, 0.5881],
  6775. [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726]],
  6776. device='cuda:0', dtype=torch.float64)
  6777. predictions are: tensor([[-0.1576, -0.9122, 1.2972, -1.2263, -0.1811, -1.4133, 0.4684, 0.2480],
  6778. [ 0.2890, -0.6472, 1.9153, 0.1004, -0.1274, -0.1649, 0.6380, 0.1997],
  6779. [-0.0181, -0.8165, 1.4741, -1.3042, -0.2795, -1.3269, 0.5811, 0.2282],
  6780. [ 0.4858, -0.4523, 1.6892, -0.2371, -0.5700, -0.3723, 0.6010, 0.1366],
  6781. [ 0.8972, -0.2306, 1.8460, 0.3335, -0.4067, 0.0752, 0.6305, 0.1847],
  6782. [ 0.3262, -0.6062, 1.8215, -0.3808, -0.6322, -0.4538, 0.5082, 0.1785],
  6783. [ 0.2143, -0.6786, 1.8186, 0.2153, -0.5068, -0.2337, 0.4954, 0.1994],
  6784. [ 0.7243, -0.3422, 1.3246, -1.1767, -0.5365, -1.1800, 0.4903, 0.1229]],
  6785. device='cuda:0', grad_fn=<AddmmBackward>)
  6786. landmarks are: tensor([[[ 0.5912, -0.3467, 1.2533, -1.0388, -0.1150, -1.3005, 0.3353,
  6787. 0.3854],
  6788. [ 0.5564, -0.3842, 1.7268, 0.1005, -0.0250, 0.3225, 0.2658,
  6789. 0.0862],
  6790. [ 0.5868, -0.3858, 1.4901, -0.9849, -0.4730, -1.0003, 0.4393,
  6791. 0.1852],
  6792. [ 0.5432, -0.4336, 1.7095, -0.1766, -0.5942, -0.4845, 0.3007,
  6793. 0.2853],
  6794. [ 0.5742, -0.3792, 1.6460, 0.3084, -0.3460, 0.1467, 0.4162,
  6795. 0.3161],
  6796. [ 0.5757, -0.3917, 1.7095, -0.4768, -0.6346, -0.4229, 0.3931,
  6797. 0.3238],
  6798. [ 0.5452, -0.4067, 1.7557, 0.0543, -0.4961, -0.3306, 0.1323,
  6799. 0.4306],
  6800. [ 0.4997, -0.4446, 1.1610, -0.9772, -0.6058, -1.0311, 0.1404,
  6801. -0.1031]]], device='cuda:0')
  6802. loss_train_step before backward: tensor(0.0569, device='cuda:0', grad_fn=<MseLossBackward>)
  6803. loss_train_step after backward: tensor(0.0569, device='cuda:0', grad_fn=<MseLossBackward>)
  6804. loss_train: 5.025980396196246
  6805. step: 61
  6806. running loss: 0.0823931212491188
  6807. Train Steps: 61/90 Loss: 0.0824 torch.Size([8, 600, 800])
  6808. torch.Size([8, 8])
  6809. tensor([[0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058],
  6810. [0.6204, 0.4013, 0.8075, 0.2400, 0.4313, 0.2050, 0.5800, 0.5150],
  6811. [0.6289, 0.4032, 0.8419, 0.5446, 0.4075, 0.5017, 0.6312, 0.5117],
  6812. [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383],
  6813. [0.6229, 0.4066, 0.8513, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350],
  6814. [0.6159, 0.4085, 0.6900, 0.2283, 0.4088, 0.1950, 0.5123, 0.5397],
  6815. [ nan, nan, 0.8463, 0.2550, 0.5850, 0.2133, 0.7129, 0.6072],
  6816. [0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400]],
  6817. device='cuda:0', dtype=torch.float64)
  6818. predictions are: tensor([[ 0.6747, -0.3336, 1.7131, -0.3043, -0.6373, -0.1198, 0.5301, 0.1279],
  6819. [ 0.3088, -0.5878, 1.4396, -1.0615, -0.3396, -1.2393, 0.4457, 0.2140],
  6820. [ 0.5520, -0.4229, 1.8363, 0.2268, -0.5093, -0.0411, 0.6594, 0.1800],
  6821. [ 0.2341, -0.6258, 1.8258, 0.1728, -0.3747, -0.0613, 0.4904, 0.1651],
  6822. [ 0.6160, -0.3853, 1.7532, 0.3894, -0.3895, -0.0292, 0.4597, 0.1998],
  6823. [-0.1318, -0.8842, 1.0308, -1.2271, -0.5247, -1.3953, 0.2541, 0.2315],
  6824. [-0.7442, -1.3088, 1.8895, -0.8294, -0.0512, -1.0553, 1.0225, 0.2369],
  6825. [ 0.9457, -0.1848, 1.5723, -1.0821, -0.2259, -1.0955, 0.6738, 0.1960]],
  6826. device='cuda:0', grad_fn=<AddmmBackward>)
  6827. landmarks are: tensor([[[ 5.2448e-01, -4.3610e-01, 1.5940e+00, -2.9207e-01, -5.4804e-01,
  6828. -9.1917e-02, 2.4319e-01, 5.0176e-02],
  6829. [ 5.7962e-01, -4.3256e-01, 1.4439e+00, -1.1774e+00, -2.9400e-01,
  6830. -1.3390e+00, 3.9307e-01, 9.2841e-02],
  6831. [ 6.1888e-01, -4.2379e-01, 1.6026e+00, 2.2948e-01, -4.0370e-01,
  6832. 3.1255e-02, 6.2979e-01, 7.7444e-02],
  6833. [ 5.3603e-01, -4.6490e-01, 1.6517e+00, 4.6189e-04, -2.8245e-01,
  6834. -6.8822e-02, 2.3086e-01, 2.0046e-01],
  6835. [ 5.9107e-01, -4.0805e-01, 1.6460e+00, 3.5458e-01, -2.0739e-01,
  6836. 4.6651e-02, 4.9700e-01, 1.8522e-01],
  6837. [ 5.5912e-01, -3.9900e-01, 9.0115e-01, -1.2313e+00, -3.9792e-01,
  6838. -1.3852e+00, 8.0445e-02, 2.0706e-01],
  6839. [-2.2859e+00, -2.2859e+00, 1.6229e+00, -1.1081e+00, 4.1617e-01,
  6840. -1.3005e+00, 1.0070e+00, 5.1879e-01],
  6841. [ 5.9579e-01, -3.8176e-01, 1.5536e+00, -1.1081e+00, -2.0739e-01,
  6842. -1.3390e+00, 5.6628e-01, 2.0831e-01]]], device='cuda:0')
  6843. loss_train_step before backward: tensor(0.0864, device='cuda:0', grad_fn=<MseLossBackward>)
  6844. loss_train_step after backward: tensor(0.0864, device='cuda:0', grad_fn=<MseLossBackward>)
  6845. loss_train: 5.112371591851115
  6846. step: 62
  6847. running loss: 0.08245760632017927
  6848.  
  6849. Train Steps: 62/90 Loss: 0.0825 torch.Size([8, 600, 800])
  6850. torch.Size([8, 8])
  6851. tensor([[0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683],
  6852. [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167],
  6853. [0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183],
  6854. [0.6357, 0.4118, 0.8400, 0.2500, 0.5413, 0.1633, 0.6725, 0.5586],
  6855. [0.6078, 0.4033, 0.8019, 0.3055, 0.3450, 0.4200, 0.6025, 0.5550],
  6856. [0.6271, 0.4040, 0.9138, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413],
  6857. [0.6261, 0.3987, 0.8688, 0.4917, 0.4300, 0.5333, 0.7010, 0.5309],
  6858. [0.6254, 0.3993, 0.8988, 0.4767, 0.3987, 0.5517, 0.6955, 0.5285]],
  6859. device='cuda:0', dtype=torch.float64)
  6860. predictions are: tensor([[ 1.0652, -0.0872, 1.6244, 0.1411, -0.5858, -0.4915, 0.3320, 0.1692],
  6861. [ 0.4764, -0.4995, 1.2932, -0.9137, -0.4796, -0.9494, 0.4467, 0.2394],
  6862. [ 0.5212, -0.4979, 1.4810, -0.8740, -0.4198, -0.9972, 0.4141, 0.2848],
  6863. [-0.5558, -1.2075, 1.6232, -1.1434, 0.0361, -1.3837, 0.8139, 0.2174],
  6864. [ 0.6402, -0.3467, 1.4580, -0.7266, -0.6276, -0.4378, 0.4899, 0.1890],
  6865. [ 0.1559, -0.7151, 1.9275, -0.6109, -0.1726, -1.0339, 0.7443, 0.1235],
  6866. [ 0.2836, -0.6025, 1.9319, 0.1667, -0.3236, 0.1527, 0.6902, 0.1907],
  6867. [ 0.1549, -0.6862, 1.9986, 0.2127, -0.4233, 0.2488, 0.7887, 0.1478]],
  6868. device='cuda:0', grad_fn=<AddmmBackward>)
  6869. landmarks are: tensor([[[ 0.5777, -0.3915, 1.7961, 0.1698, -0.5192, -0.5307, 0.2141,
  6870. 0.3392],
  6871. [ 0.5631, -0.4129, 1.2129, -0.9233, -0.4152, -1.0311, 0.4566,
  6872. 0.5624],
  6873. [ 0.5759, -0.3948, 1.2072, -0.9849, -0.4672, -0.8156, 0.4912,
  6874. 0.5701],
  6875. [ 0.6504, -0.3840, 1.5940, -1.1312, 0.2141, -1.5315, 0.8205,
  6876. 0.2944],
  6877. [ 0.5215, -0.4232, 1.4182, -0.8747, -0.6924, -0.3460, 0.4970,
  6878. 0.2776],
  6879. [ 0.6108, -0.4201, 1.9346, -0.5538, -0.1497, -1.0773, 1.0545,
  6880. 0.2142],
  6881. [ 0.6059, -0.4442, 1.7268, -0.0149, -0.2998, 0.1775, 0.9521,
  6882. 0.1661],
  6883. [ 0.6026, -0.4417, 1.8654, -0.0842, -0.4441, 0.2622, 0.9265,
  6884. 0.1554]]], device='cuda:0')
  6885. loss_train_step before backward: tensor(0.0639, device='cuda:0', grad_fn=<MseLossBackward>)
  6886. loss_train_step after backward: tensor(0.0639, device='cuda:0', grad_fn=<MseLossBackward>)
  6887. loss_train: 5.176298273727298
  6888. step: 63
  6889. running loss: 0.08216346466233806
  6890. Train Steps: 63/90 Loss: 0.0822 torch.Size([8, 600, 800])
  6891. torch.Size([8, 8])
  6892. tensor([[0.6221, 0.4107, 0.7788, 0.3033, 0.3950, 0.2817, 0.6075, 0.5517],
  6893. [0.6250, 0.4110, 0.7238, 0.2067, 0.4263, 0.1883, 0.5625, 0.5633],
  6894. [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609],
  6895. [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334],
  6896. [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123],
  6897. [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650],
  6898. [0.6078, 0.4033, 0.8019, 0.3055, 0.3450, 0.4200, 0.6025, 0.5550],
  6899. [0.6108, 0.4011, 0.8037, 0.3400, 0.3700, 0.2933, 0.5658, 0.5617]],
  6900. device='cuda:0', dtype=torch.float64)
  6901. predictions are: tensor([[ 1.0356, -0.1234, 1.7094, -0.6482, -0.3317, -0.7847, 0.5696, 0.2211],
  6902. [ 0.3286, -0.6007, 1.2848, -0.8995, -0.3070, -1.1710, 0.4285, 0.2396],
  6903. [-0.8547, -1.3968, 1.9650, -0.7491, 0.1321, -1.1016, 1.1164, 0.2244],
  6904. [ 0.4023, -0.5407, 1.4807, -0.8273, -0.4771, -0.7097, 0.6189, 0.1762],
  6905. [ 0.4463, -0.5124, 1.2960, -0.9030, -0.4845, -0.9337, 0.3351, 0.2077],
  6906. [ 0.2804, -0.6267, 1.8965, 0.6625, -0.3471, 0.0380, 0.4486, 0.2180],
  6907. [ 0.5654, -0.3786, 1.6728, -0.4492, -0.5768, -0.1508, 0.6567, 0.1743],
  6908. [ 0.4768, -0.4687, 1.5823, -0.5242, -0.4902, -0.6029, 0.5122, 0.1763]],
  6909. device='cuda:0', grad_fn=<AddmmBackward>)
  6910. landmarks are: tensor([[[ 0.5875, -0.3888, 1.3111, -0.8848, -0.4614, -0.9849, 0.5201,
  6911. 0.2622],
  6912. [ 0.6010, -0.3875, 1.0570, -1.3313, -0.3171, -1.4160, 0.3122,
  6913. 0.3161],
  6914. [ 0.6372, -0.3575, 1.6633, -1.2082, 0.2199, -1.2467, 1.1313,
  6915. 0.3050],
  6916. [ 0.5730, -0.4562, 1.2195, -1.2440, -0.5497, -0.7711, 0.5704,
  6917. 0.1779],
  6918. [ 0.5342, -0.4280, 1.0051, -1.0619, -0.5711, -1.0388, 0.2226,
  6919. 0.0802],
  6920. [ 0.5770, -0.4036, 1.7095, 0.3084, -0.3691, -0.2690, 0.2314,
  6921. 0.3238],
  6922. [ 0.5215, -0.4232, 1.4182, -0.8747, -0.6924, -0.3460, 0.4970,
  6923. 0.2776],
  6924. [ 0.5355, -0.4335, 1.4266, -0.7155, -0.5769, -0.9310, 0.3275,
  6925. 0.3087]]], device='cuda:0')
  6926. loss_train_step before backward: tensor(0.0955, device='cuda:0', grad_fn=<MseLossBackward>)
  6927. loss_train_step after backward: tensor(0.0955, device='cuda:0', grad_fn=<MseLossBackward>)
  6928. loss_train: 5.271774662658572
  6929. step: 64
  6930. running loss: 0.08237147910404019
  6931. Train Steps: 64/90 Loss: 0.0824 torch.Size([8, 600, 800])
  6932. torch.Size([8, 8])
  6933. tensor([[0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950],
  6934. [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672],
  6935. [0.6126, 0.3954, 0.8538, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350],
  6936. [0.6147, 0.4107, 0.8137, 0.3333, 0.3750, 0.2683, 0.5006, 0.5412],
  6937. [0.6059, 0.4002, 0.7562, 0.2767, 0.3538, 0.3033, 0.5529, 0.5455],
  6938. [0.6034, 0.4011, 0.7350, 0.2533, 0.3438, 0.3367, 0.5516, 0.5084],
  6939. [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5837, 0.5500],
  6940. [0.6064, 0.3953, 0.8738, 0.4417, 0.3663, 0.4683, 0.5511, 0.5416]],
  6941. device='cuda:0', dtype=torch.float64)
  6942. predictions are: tensor([[-0.1920, -0.9985, 1.9063, 0.1103, -0.1318, -0.3085, 0.7094, 0.2498],
  6943. [ 0.7767, -0.3053, 1.7186, -0.3475, -0.4719, -0.7444, 0.6525, 0.1397],
  6944. [ 0.1857, -0.7125, 1.8274, -0.0031, -0.2454, -0.2910, 0.5246, 0.2424],
  6945. [ 0.6512, -0.3863, 1.4643, -0.8827, -0.4335, -1.0586, 0.3989, 0.2542],
  6946. [ 0.4884, -0.4620, 1.2575, -1.1185, -0.4550, -1.0417, 0.4923, 0.2755],
  6947. [ 0.5038, -0.4734, 1.2670, -1.2477, -0.5148, -1.0009, 0.4584, 0.2556],
  6948. [ 0.4376, -0.5049, 1.8492, -0.1855, -0.2348, -0.1066, 0.7327, 0.2066],
  6949. [ 0.5826, -0.4122, 1.8609, -0.2930, -0.4397, -0.3699, 0.7610, 0.1792]],
  6950. device='cuda:0', grad_fn=<AddmmBackward>)
  6951. landmarks are: tensor([[[ 5.8199e-01, -4.7544e-01, 1.7095e+00, 1.9292e-01, -2.5358e-01,
  6952. 5.4350e-02, 4.9700e-01, 4.6189e-04],
  6953. [ 6.1322e-01, -4.3241e-01, 1.8192e+00, -8.4219e-02, -6.2309e-01,
  6954. -6.3849e-01, 5.5366e-01, -1.2778e-01],
  6955. [ 5.4348e-01, -4.5974e-01, 1.6575e+00, 1.5858e-02, -3.2286e-01,
  6956. -1.1501e-01, 1.8767e-01, 1.8544e-01],
  6957. [ 5.5319e-01, -3.8879e-01, 1.4727e+00, -7.4627e-01, -5.5381e-01,
  6958. -1.0465e+00, 2.6467e-02, 2.1383e-01],
  6959. [ 5.1288e-01, -4.3741e-01, 1.2072e+00, -1.0080e+00, -6.5196e-01,
  6960. -8.8483e-01, 2.6787e-01, 2.3353e-01],
  6961. [ 5.0092e-01, -4.3333e-01, 1.1090e+00, -1.1158e+00, -6.9815e-01,
  6962. -7.3087e-01, 2.6170e-01, 6.2199e-02],
  6963. [ 5.6039e-01, -4.6197e-01, 1.7961e+00, -1.9969e-01, -3.2286e-01,
  6964. 1.0824e-01, 4.1039e-01, 2.5450e-01],
  6965. [ 5.1490e-01, -4.6028e-01, 1.7499e+00, -2.4588e-01, -5.9423e-01,
  6966. -1.2271e-01, 2.5964e-01, 2.1549e-01]]], device='cuda:0')
  6967. loss_train_step before backward: tensor(0.0448, device='cuda:0', grad_fn=<MseLossBackward>)
  6968. loss_train_step after backward: tensor(0.0448, device='cuda:0', grad_fn=<MseLossBackward>)
  6969. loss_train: 5.316538447514176
  6970. step: 65
  6971. running loss: 0.08179289919252579
  6972. Train Steps: 65/90 Loss: 0.0818 torch.Size([8, 600, 800])
  6973. torch.Size([8, 8])
  6974. tensor([[0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383],
  6975. [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217],
  6976. [0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153],
  6977. [0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058],
  6978. [0.6185, 0.4079, 0.8838, 0.4617, 0.4838, 0.5650, 0.6175, 0.5850],
  6979. [0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5737, 0.5350],
  6980. [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364],
  6981. [0.6284, 0.4093, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167]],
  6982. device='cuda:0', dtype=torch.float64)
  6983. predictions are: tensor([[ 0.1470, -0.7047, 1.6310, -0.2381, -0.2549, -0.3992, 0.4727, 0.2435],
  6984. [ 0.4058, -0.5714, 1.8483, -0.1800, -0.3816, -0.1532, 0.7788, 0.2107],
  6985. [ 0.4472, -0.5076, 1.6044, -0.1745, -0.2147, -0.2088, 0.4506, 0.2606],
  6986. [ 0.2987, -0.5943, 1.5517, -0.6556, -0.5261, -0.4434, 0.5252, 0.2261],
  6987. [ 0.4258, -0.5116, 1.7422, -0.3681, -0.1858, -0.1330, 0.6773, 0.2530],
  6988. [ 0.9321, -0.1669, 1.5974, -0.3834, -0.4631, -0.9787, 0.4675, 0.2501],
  6989. [ 0.9310, -0.1953, 1.3346, -1.2929, -0.2924, -1.4419, 0.5972, 0.2600],
  6990. [ 0.1639, -0.7135, 1.7702, -0.3880, -0.5578, -0.8406, 0.6680, 0.1920]],
  6991. device='cuda:0', grad_fn=<AddmmBackward>)
  6992. landmarks are: tensor([[[ 5.3603e-01, -4.6490e-01, 1.6517e+00, 4.6189e-04, -2.8245e-01,
  6993. -6.8822e-02, 2.3086e-01, 2.0046e-01],
  6994. [ 6.0468e-01, -4.2094e-01, 1.7557e+00, -3.0331e-02, -4.8453e-01,
  6995. 2.5450e-01, 6.5866e-01, 1.2363e-01],
  6996. [ 5.5484e-01, -4.6823e-01, 1.7309e+00, 9.6578e-02, -1.3942e-01,
  6997. 1.7573e-01, 5.0451e-01, 9.4188e-02],
  6998. [ 5.2448e-01, -4.3610e-01, 1.5940e+00, -2.9207e-01, -5.4804e-01,
  6999. -9.1917e-02, 2.4319e-01, 5.0176e-02],
  7000. [ 5.7067e-01, -4.0169e-01, 1.7961e+00, -1.5350e-01, -5.1501e-02,
  7001. 3.2379e-01, 5.6628e-01, 4.1617e-01],
  7002. [ 5.9007e-01, -4.0000e-01, 1.8423e+00, -6.8822e-02, -5.3072e-01,
  7003. -9.2333e-01, 3.6420e-01, 1.8522e-01],
  7004. [ 6.1577e-01, -3.9601e-01, 1.4092e+00, -1.2774e+00, -2.0739e-01,
  7005. -1.1851e+00, 8.4910e-01, 1.9173e-01],
  7006. [ 6.1640e-01, -3.9561e-01, 1.8249e+00, -1.1501e-01, -6.0000e-01,
  7007. -5.0762e-01, 5.8360e-01, 1.0054e-01]]], device='cuda:0')
  7008. loss_train_step before backward: tensor(0.0462, device='cuda:0', grad_fn=<MseLossBackward>)
  7009. loss_train_step after backward: tensor(0.0462, device='cuda:0', grad_fn=<MseLossBackward>)
  7010. loss_train: 5.362761614844203
  7011. step: 66
  7012. running loss: 0.0812539638612758
  7013.  
  7014. Train Steps: 66/90 Loss: 0.0813 torch.Size([8, 600, 800])
  7015. torch.Size([8, 8])
  7016. tensor([[0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411],
  7017. [ nan, nan, 0.7648, 0.2722, 0.3962, 0.2183, 0.5060, 0.5422],
  7018. [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309],
  7019. [ nan, nan, 0.7225, 0.2167, 0.3987, 0.2283, 0.5427, 0.5181],
  7020. [0.6202, 0.4064, 0.7879, 0.2179, 0.4567, 0.1725, 0.5955, 0.5478],
  7021. [0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533],
  7022. [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100],
  7023. [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320]],
  7024. device='cuda:0', dtype=torch.float64)
  7025. predictions are: tensor([[ 1.1903, 0.0091, 1.5244, -0.6896, -0.3186, -0.7837, 0.4619, 0.2637],
  7026. [-1.1955, -1.5569, 1.2581, -0.9294, -0.3222, -1.0647, 0.3363, 0.3174],
  7027. [ 1.1021, -0.0334, 1.8661, 0.1296, -0.4194, -0.2605, 0.6152, 0.1854],
  7028. [-1.0808, -1.4725, 1.1546, -1.0496, -0.2942, -1.1167, 0.2463, 0.3367],
  7029. [ 1.2006, -0.0122, 1.3387, -0.9424, -0.2428, -0.9820, 0.4809, 0.2814],
  7030. [ 0.8916, -0.2122, 1.2474, -0.9546, -0.4703, -0.7094, 0.5053, 0.2677],
  7031. [ 0.9518, -0.2066, 1.7941, -0.2452, -0.5559, -0.2133, 0.7859, 0.1155],
  7032. [ 0.6461, -0.3687, 1.8293, 0.2481, -0.3678, 0.1926, 0.6810, 0.1674]],
  7033. device='cuda:0', grad_fn=<AddmmBackward>)
  7034. landmarks are: tensor([[[ 0.5713, -0.4321, 1.4975, -0.8134, -0.3094, -1.3345, 0.3779,
  7035. 0.2134],
  7036. [-2.2859, -2.2859, 1.2469, -1.0288, -0.4557, -1.2774, 0.0511,
  7037. 0.2183],
  7038. [ 0.6520, -0.3623, 1.8885, 0.0313, -0.5538, -0.5384, 0.6926,
  7039. 0.1661],
  7040. [-2.2859, -2.2859, 1.0513, -1.2851, -0.4441, -1.2313, 0.2206,
  7041. 0.1073],
  7042. [ 0.5787, -0.4086, 1.3535, -1.2794, -0.1764, -1.4891, 0.4645,
  7043. 0.2442],
  7044. [ 0.5878, -0.4052, 1.0229, -1.2855, -0.5596, -0.8232, 0.5316,
  7045. 0.2699],
  7046. [ 0.6220, -0.4436, 1.8711, -0.3691, -0.6173, -0.4999, 0.6702,
  7047. 0.0697],
  7048. [ 0.6273, -0.4393, 1.6402, 0.1313, -0.5076, 0.0467, 1.1532,
  7049. 0.1715]]], device='cuda:0')
  7050. loss_train_step before backward: tensor(0.1144, device='cuda:0', grad_fn=<MseLossBackward>)
  7051. loss_train_step after backward: tensor(0.1144, device='cuda:0', grad_fn=<MseLossBackward>)
  7052. loss_train: 5.477159423753619
  7053. step: 67
  7054. running loss: 0.08174864811572566
  7055. Train Steps: 67/90 Loss: 0.0817 torch.Size([8, 600, 800])
  7056. torch.Size([8, 8])
  7057. tensor([[0.6250, 0.4106, 0.8700, 0.3717, 0.3588, 0.4967, 0.6038, 0.5167],
  7058. [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150],
  7059. [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285],
  7060. [0.6250, 0.4236, 0.8638, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799],
  7061. [ nan, nan, 0.7268, 0.2333, 0.4125, 0.1933, 0.5112, 0.5383],
  7062. [0.6243, 0.4128, 0.7762, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417],
  7063. [0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733],
  7064. [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795]],
  7065. device='cuda:0', dtype=torch.float64)
  7066. predictions are: tensor([[ 0.7726, -0.2872, 1.5579, -0.6419, -0.4527, -0.1764, 0.6722, 0.1945],
  7067. [ 0.7038, -0.3455, 1.7372, -0.0100, -0.4515, -0.4409, 0.4419, 0.2044],
  7068. [ 0.3806, -0.5373, 1.5902, -0.6628, -0.4155, -0.2866, 0.7564, 0.1726],
  7069. [ 0.3435, -0.5409, 1.6290, -0.5285, -0.3532, -0.7808, 0.4576, 0.3081],
  7070. [-0.3292, -1.0527, 0.9807, -1.2696, -0.3676, -1.4329, 0.1426, 0.3147],
  7071. [ 0.8057, -0.2529, 1.3971, -1.0243, -0.3280, -1.0554, 0.4332, 0.2968],
  7072. [ 0.4678, -0.4743, 1.6984, 0.0821, -0.4409, -0.3279, 0.4225, 0.2544],
  7073. [ 0.9089, -0.2301, 1.7610, 0.1076, -0.3777, 0.0481, 0.7147, 0.1679]],
  7074. device='cuda:0', grad_fn=<AddmmBackward>)
  7075. landmarks are: tensor([[[ 0.6010, -0.3896, 1.7326, -0.5692, -0.6289, 0.0082, 0.5028,
  7076. 0.1005],
  7077. [ 0.6096, -0.4046, 1.8249, -0.0072, -0.6115, -0.3537, 0.6182,
  7078. 0.0928],
  7079. [ 0.6108, -0.4273, 1.8711, -0.6616, -0.5769, -0.1997, 0.9156,
  7080. 0.1554],
  7081. [ 0.6008, -0.3293, 1.7037, -0.5461, -0.4152, -0.8309, 0.3234,
  7082. 0.3928],
  7083. [-2.2859, -2.2859, 1.0712, -1.2085, -0.3806, -1.3929, 0.0755,
  7084. 0.2006],
  7085. [ 0.5977, -0.3792, 1.2995, -1.0311, -0.5192, -0.8386, 0.5836,
  7086. 0.2160],
  7087. [ 0.6121, -0.3138, 1.8423, 0.0082, -0.6462, -0.3075, 0.3469,
  7088. 0.3623],
  7089. [ 0.6355, -0.4080, 1.6113, 0.1852, -0.4730, 0.1467, 0.9996,
  7090. 0.3905]]], device='cuda:0')
  7091. loss_train_step before backward: tensor(0.1032, device='cuda:0', grad_fn=<MseLossBackward>)
  7092. loss_train_step after backward: tensor(0.1032, device='cuda:0', grad_fn=<MseLossBackward>)
  7093. loss_train: 5.580386197194457
  7094. step: 68
  7095. running loss: 0.08206450289991848
  7096. Train Steps: 68/90 Loss: 0.0821 torch.Size([8, 600, 800])
  7097. torch.Size([8, 8])
  7098. tensor([[0.6277, 0.4083, 0.8350, 0.2717, 0.4562, 0.1800, 0.5918, 0.4878],
  7099. [0.6059, 0.4002, 0.7562, 0.2767, 0.3538, 0.3033, 0.5529, 0.5455],
  7100. [0.6142, 0.4127, 0.7575, 0.3067, 0.3438, 0.4383, 0.5778, 0.5207],
  7101. [0.6189, 0.4049, 0.8888, 0.4417, 0.4213, 0.5200, 0.5988, 0.5633],
  7102. [0.6357, 0.4118, 0.8400, 0.2500, 0.5413, 0.1633, 0.6725, 0.5586],
  7103. [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550],
  7104. [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367],
  7105. [0.6212, 0.4171, 0.7875, 0.3633, 0.3813, 0.2933, 0.5675, 0.5700]],
  7106. device='cuda:0', dtype=torch.float64)
  7107. predictions are: tensor([[ 0.0113, -0.7472, 1.4550, -0.8894, -0.2580, -0.9753, 0.4123, 0.2772],
  7108. [ 0.4000, -0.5421, 1.2346, -0.7698, -0.5962, -0.8030, 0.3929, 0.2342],
  7109. [ 0.3309, -0.5806, 1.3000, -0.6666, -0.6106, -0.3164, 0.3896, 0.2519],
  7110. [ 0.7557, -0.3389, 1.8301, 0.0937, -0.4879, 0.2119, 0.5420, 0.1904],
  7111. [ 0.4026, -0.5461, 1.5781, -1.0008, -0.0243, -1.1922, 0.8809, 0.2303],
  7112. [ 0.6067, -0.4049, 1.7667, 0.0835, -0.2052, -0.0215, 0.5257, 0.1727],
  7113. [ 0.5976, -0.4035, 1.4973, -0.7421, -0.5810, -0.9208, 0.4500, 0.2530],
  7114. [ 0.6578, -0.3394, 1.3067, -0.5173, -0.5163, -0.6640, 0.3923, 0.2634]],
  7115. device='cuda:0', grad_fn=<AddmmBackward>)
  7116. landmarks are: tensor([[[ 0.6132, -0.4002, 1.5709, -1.0311, -0.1785, -1.4545, 0.4474,
  7117. -0.0328],
  7118. [ 0.5129, -0.4374, 1.2072, -1.0080, -0.6520, -0.8848, 0.2679,
  7119. 0.2335],
  7120. [ 0.5509, -0.3798, 1.2129, -0.8694, -0.6982, -0.2613, 0.3830,
  7121. 0.1193],
  7122. [ 0.5726, -0.4159, 1.8192, -0.2459, -0.3402, 0.1159, 0.4797,
  7123. 0.3161],
  7124. [ 0.6504, -0.3840, 1.5940, -1.1312, 0.2141, -1.5315, 0.8205,
  7125. 0.2944],
  7126. [ 0.5351, -0.4321, 1.7326, -0.1381, 0.0463, 0.1049, 0.4162,
  7127. 0.2776],
  7128. [ 0.5697, -0.4442, 1.3529, -0.9515, -0.5774, -0.7801, 0.5253,
  7129. 0.1931],
  7130. [ 0.5832, -0.3593, 1.3515, -0.6077, -0.5249, -0.9310, 0.3353,
  7131. 0.3469]]], device='cuda:0')
  7132. loss_train_step before backward: tensor(0.0307, device='cuda:0', grad_fn=<MseLossBackward>)
  7133. loss_train_step after backward: tensor(0.0307, device='cuda:0', grad_fn=<MseLossBackward>)
  7134. loss_train: 5.611077474430203
  7135. step: 69
  7136. running loss: 0.08131996339753918
  7137. Train Steps: 69/90 Loss: 0.0813 torch.Size([8, 600, 800])
  7138. torch.Size([8, 8])
  7139. tensor([[0.6185, 0.4067, 0.8838, 0.4450, 0.4037, 0.4733, 0.5213, 0.5142],
  7140. [0.6149, 0.4054, 0.6713, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695],
  7141. [0.6175, 0.3957, 0.8700, 0.4817, 0.4662, 0.5133, 0.5800, 0.5517],
  7142. [0.6268, 0.4061, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300],
  7143. [0.6165, 0.4106, 0.7575, 0.1733, 0.3838, 0.2650, 0.5680, 0.5116],
  7144. [0.6218, 0.4137, 0.7263, 0.2233, 0.4075, 0.2650, 0.6212, 0.5783],
  7145. [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517],
  7146. [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609]],
  7147. device='cuda:0', dtype=torch.float64)
  7148. predictions are: tensor([[ 0.3484, -0.5802, 1.6820, -0.0106, -0.6042, -0.0191, 0.2717, 0.2010],
  7149. [-0.0371, -0.8269, 1.0462, -1.0266, -0.5768, -1.1250, 0.1318, 0.3113],
  7150. [ 0.2505, -0.6367, 1.7123, 0.0646, -0.3241, 0.1082, 0.4883, 0.1706],
  7151. [ 0.7437, -0.2872, 1.5725, -0.9677, -0.3447, -1.0602, 0.6722, 0.2208],
  7152. [ 0.6053, -0.4160, 1.2632, -0.9159, -0.6230, -0.9971, 0.3718, 0.2289],
  7153. [ 0.5458, -0.4351, 1.2529, -0.7791, -0.5063, -0.8925, 0.3689, 0.3058],
  7154. [ 0.9709, -0.1950, 1.7543, 0.1357, -0.3991, 0.3494, 0.5204, 0.2254],
  7155. [ 0.2469, -0.6231, 1.6615, -1.0366, -0.0289, -1.1851, 1.0284, 0.2094]],
  7156. device='cuda:0', grad_fn=<AddmmBackward>)
  7157. landmarks are: tensor([[[ 0.5708, -0.4075, 1.7961, -0.2305, -0.4210, -0.0996, 0.1219,
  7158. 0.0893],
  7159. [ 0.5544, -0.4133, 0.8145, -1.2082, -0.4268, -1.3544, 0.1221,
  7160. 0.3446],
  7161. [ 0.5662, -0.4581, 1.7326, -0.0611, -0.1323, 0.0851, 0.3931,
  7162. 0.2622],
  7163. [ 0.6093, -0.4104, 1.5709, -1.1620, -0.1727, -1.2313, 0.6471,
  7164. 0.1621],
  7165. [ 0.5618, -0.3895, 1.2129, -1.4853, -0.5134, -1.0619, 0.3378,
  7166. 0.0772],
  7167. [ 0.5863, -0.3749, 1.0686, -1.2543, -0.4037, -1.0619, 0.5836,
  7168. 0.3854],
  7169. [ 0.5897, -0.3804, 1.7788, -0.0226, -0.2016, 0.3007, 0.6125,
  7170. 0.2622],
  7171. [ 0.6372, -0.3575, 1.6633, -1.2082, 0.2199, -1.2467, 1.1313,
  7172. 0.3050]]], device='cuda:0')
  7173. loss_train_step before backward: tensor(0.0399, device='cuda:0', grad_fn=<MseLossBackward>)
  7174. loss_train_step after backward: tensor(0.0399, device='cuda:0', grad_fn=<MseLossBackward>)
  7175. loss_train: 5.6510243806988
  7176. step: 70
  7177. running loss: 0.08072891972426857
  7178.  
  7179. Train Steps: 70/90 Loss: 0.0807 torch.Size([8, 600, 800])
  7180. torch.Size([8, 8])
  7181. tensor([[0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167],
  7182. [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100],
  7183. [ nan, nan, 0.7425, 0.2117, 0.3937, 0.2433, 0.5438, 0.5567],
  7184. [0.6064, 0.3953, 0.8738, 0.4417, 0.3663, 0.4683, 0.5511, 0.5416],
  7185. [0.6275, 0.4024, 0.8600, 0.2283, 0.5350, 0.1800, 0.7074, 0.5413],
  7186. [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650],
  7187. [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450],
  7188. [0.6185, 0.4098, 0.8838, 0.4900, 0.4537, 0.5800, 0.6288, 0.5400]],
  7189. device='cuda:0', dtype=torch.float64)
  7190. predictions are: tensor([[ 0.6906, -0.3369, 1.5061, -0.8076, -0.6131, -0.6987, 0.4593, 0.2867],
  7191. [ 1.1220, -0.1122, 1.6284, 0.0842, -0.5748, -0.0547, 0.4497, 0.1736],
  7192. [-1.1066, -1.4997, 1.2091, -1.2279, -0.3574, -1.2843, 0.2825, 0.3239],
  7193. [ 0.8739, -0.2342, 1.6656, -0.2235, -0.6230, -0.1619, 0.4995, 0.1426],
  7194. [ 0.1316, -0.6897, 1.6279, -1.2489, 0.0040, -1.3089, 1.0166, 0.2204],
  7195. [-0.0195, -0.7876, 1.2561, -1.1085, -0.4725, -1.0848, 0.1783, 0.2908],
  7196. [ 1.0028, -0.1599, 1.6230, -0.1222, -0.6060, -0.2397, 0.3403, 0.1853],
  7197. [ 1.0546, -0.1132, 1.6749, -0.0623, -0.2407, 0.1899, 0.5196, 0.2179]],
  7198. device='cuda:0', grad_fn=<AddmmBackward>)
  7199. landmarks are: tensor([[[ 0.5768, -0.3899, 1.3861, -0.7771, -0.5885, -0.5461, 0.5028,
  7200. 0.5624],
  7201. [ 0.6125, -0.4273, 1.6402, 0.2006, -0.4788, -0.0303, 0.6182,
  7202. 0.0697],
  7203. [-2.2859, -2.2859, 1.1436, -1.3082, -0.4672, -1.1620, 0.2256,
  7204. 0.2853],
  7205. [ 0.5149, -0.4603, 1.7499, -0.2459, -0.5942, -0.1227, 0.2596,
  7206. 0.2155],
  7207. [ 0.6125, -0.4273, 1.6864, -1.2313, 0.1852, -1.4545, 0.9814,
  7208. 0.2142],
  7209. [ 0.5320, -0.4189, 1.3053, -1.0773, -0.5711, -0.9849, 0.2267,
  7210. 0.3237],
  7211. [ 0.5785, -0.4252, 1.7676, -0.1602, -0.5845, -0.3446, 0.4566,
  7212. 0.2314],
  7213. [ 0.5709, -0.3933, 1.7961, -0.0226, -0.1901, 0.3931, 0.6182,
  7214. 0.2083]]], device='cuda:0')
  7215. loss_train_step before backward: tensor(0.0682, device='cuda:0', grad_fn=<MseLossBackward>)
  7216. loss_train_step after backward: tensor(0.0682, device='cuda:0', grad_fn=<MseLossBackward>)
  7217. loss_train: 5.719259122386575
  7218. step: 71
  7219. running loss: 0.08055294538572641
  7220. Train Steps: 71/90 Loss: 0.0806 torch.Size([8, 600, 800])
  7221. torch.Size([8, 8])
  7222. tensor([[0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367],
  7223. [0.6100, 0.4016, 0.8600, 0.5067, 0.4612, 0.5233, 0.5086, 0.5519],
  7224. [0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133],
  7225. [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500],
  7226. [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879],
  7227. [0.6192, 0.4128, 0.8513, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633],
  7228. [0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373],
  7229. [0.6353, 0.4128, 0.9138, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991]],
  7230. device='cuda:0', dtype=torch.float64)
  7231. predictions are: tensor([[ 0.3293, -0.5473, 1.5970, -0.7552, -0.4687, -1.1431, 0.3605, 0.2590],
  7232. [ 0.5273, -0.4548, 1.5444, -0.1336, -0.3016, -0.1362, 0.2556, 0.2522],
  7233. [ 0.5754, -0.4633, 1.3858, -1.3929, -0.2541, -1.4203, 0.6970, 0.2267],
  7234. [ 0.3987, -0.5329, 1.1977, -1.1102, -0.6819, -0.5298, 0.3855, 0.2635],
  7235. [ 0.7827, -0.3052, 1.5950, -0.4448, -0.5119, 0.0966, 0.5064, 0.1696],
  7236. [ 0.7617, -0.3088, 1.5791, 0.0172, -0.4719, 0.0445, 0.3501, 0.2548],
  7237. [ 0.5747, -0.4432, 1.4921, -0.0964, -0.5980, -0.3201, 0.3020, 0.2039],
  7238. [-0.0334, -0.8246, 1.8454, -0.9392, -0.1398, -1.1044, 0.9246, 0.2017]],
  7239. device='cuda:0', grad_fn=<AddmmBackward>)
  7240. landmarks are: tensor([[[ 0.5900, -0.3932, 1.8307, -0.3921, -0.4268, -1.1851, 0.3758,
  7241. 0.1929],
  7242. [ 0.5318, -0.4310, 1.6864, 0.0543, -0.1554, 0.1313, 0.0635,
  7243. 0.2634],
  7244. [ 0.6085, -0.4084, 1.5536, -1.1466, -0.0746, -1.4853, 0.6298,
  7245. 0.0851],
  7246. [ 0.5763, -0.4147, 1.3226, -1.0619, -0.6635, -0.4152, 0.5374,
  7247. 0.2545],
  7248. [ 0.5676, -0.4327, 1.8252, -0.2095, -0.4886, 0.3084, 0.4727,
  7249. -0.0322],
  7250. [ 0.5742, -0.3792, 1.6460, 0.3084, -0.3460, 0.1467, 0.4162,
  7251. 0.3161],
  7252. [ 0.5680, -0.4562, 1.5697, 0.4947, -0.4904, -0.1503, 0.3536,
  7253. 0.1956],
  7254. [ 0.6487, -0.3792, 1.9346, -0.6539, -0.1208, -0.7848, 1.0143,
  7255. 0.4814]]], device='cuda:0')
  7256. loss_train_step before backward: tensor(0.0389, device='cuda:0', grad_fn=<MseLossBackward>)
  7257. loss_train_step after backward: tensor(0.0389, device='cuda:0', grad_fn=<MseLossBackward>)
  7258. loss_train: 5.758151488378644
  7259. step: 72
  7260. running loss: 0.07997432622748117
  7261. Train Steps: 72/90 Loss: 0.0800 torch.Size([8, 600, 800])
  7262. torch.Size([8, 8])
  7263. tensor([[0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167],
  7264. [0.6230, 0.4113, 0.7213, 0.1983, 0.4325, 0.2367, 0.6262, 0.5400],
  7265. [0.6204, 0.4055, 0.8438, 0.5733, 0.4574, 0.4801, 0.5487, 0.5617],
  7266. [0.6042, 0.3990, 0.6831, 0.2875, 0.3500, 0.3133, 0.5143, 0.5510],
  7267. [0.6143, 0.4034, 0.8800, 0.4833, 0.4512, 0.5367, 0.5289, 0.5097],
  7268. [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450],
  7269. [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901],
  7270. [0.6152, 0.4131, 0.6863, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305]],
  7271. device='cuda:0', dtype=torch.float64)
  7272. predictions are: tensor([[ 0.7339, -0.3342, 1.6868, -0.8073, -0.5287, -0.6222, 0.6321, 0.2477],
  7273. [ 0.2298, -0.6195, 1.3810, -1.2579, -0.2440, -1.3400, 0.6141, 0.2401],
  7274. [ 0.6787, -0.3341, 1.6653, 0.1779, -0.2311, -0.0670, 0.4532, 0.2377],
  7275. [ 0.1206, -0.7352, 1.1463, -1.0930, -0.5663, -0.8224, 0.2956, 0.2897],
  7276. [ 0.6156, -0.3924, 1.8099, -0.0031, -0.2276, 0.0718, 0.4336, 0.1735],
  7277. [ 0.9364, -0.2195, 1.8013, -0.1140, -0.5267, -0.1311, 0.5028, 0.1453],
  7278. [-0.1940, -0.9203, 1.4506, -1.0456, -0.4763, -1.1710, 0.3304, 0.2000],
  7279. [ 0.4077, -0.5119, 1.1473, -1.1172, -0.5598, -0.7982, 0.3682, 0.2957]],
  7280. device='cuda:0', grad_fn=<AddmmBackward>)
  7281. landmarks are: tensor([[[ 0.5768, -0.3899, 1.3861, -0.7771, -0.5885, -0.5461, 0.5028,
  7282. 0.5624],
  7283. [ 0.5917, -0.3861, 1.0455, -1.3698, -0.2882, -1.1928, 0.6067,
  7284. 0.2083],
  7285. [ 0.5795, -0.4129, 1.6113, 0.3623, -0.1733, -0.0684, 0.2487,
  7286. 0.3084],
  7287. [ 0.5049, -0.4428, 0.8692, -0.9581, -0.6693, -0.8386, 0.0897,
  7288. 0.2589],
  7289. [ 0.5515, -0.4225, 1.7788, -0.0534, -0.2016, 0.1929, 0.1568,
  7290. 0.0682],
  7291. [ 0.5785, -0.4252, 1.7676, -0.1602, -0.5845, -0.3446, 0.4566,
  7292. 0.2314],
  7293. [ 0.5515, -0.4129, 1.4785, -1.0080, -0.5192, -1.1004, 0.1034,
  7294. -0.0220],
  7295. [ 0.5557, -0.3779, 0.8838, -1.1004, -0.6115, -0.7617, 0.3769,
  7296. 0.1644]]], device='cuda:0')
  7297. loss_train_step before backward: tensor(0.0385, device='cuda:0', grad_fn=<MseLossBackward>)
  7298. loss_train_step after backward: tensor(0.0385, device='cuda:0', grad_fn=<MseLossBackward>)
  7299. loss_train: 5.796630242839456
  7300. step: 73
  7301. running loss: 0.0794058937375268
  7302. Train Steps: 73/90 Loss: 0.0794 torch.Size([8, 600, 800])
  7303. torch.Size([8, 8])
  7304. tensor([[0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466],
  7305. [0.6261, 0.3987, 0.8688, 0.4917, 0.4300, 0.5333, 0.7010, 0.5309],
  7306. [0.6136, 0.4085, 0.6688, 0.2317, 0.3862, 0.2367, 0.5517, 0.5783],
  7307. [0.6261, 0.4066, 0.8325, 0.2150, 0.4763, 0.2667, 0.7002, 0.5633],
  7308. [0.6273, 0.4100, 0.7137, 0.2133, 0.4000, 0.2650, 0.6075, 0.5633],
  7309. [0.6185, 0.4079, 0.8838, 0.4617, 0.4838, 0.5650, 0.6175, 0.5850],
  7310. [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283],
  7311. [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285]],
  7312. device='cuda:0', dtype=torch.float64)
  7313. predictions are: tensor([[ 0.2847, -0.5792, 1.3632, -0.7519, -0.6296, -0.4213, 0.2320, 0.2479],
  7314. [ 0.9402, -0.1907, 1.7925, -0.0052, -0.3164, 0.1891, 0.5462, 0.1902],
  7315. [-0.1207, -0.9044, 1.0360, -1.2763, -0.4952, -1.2274, 0.1613, 0.3052],
  7316. [ 0.5824, -0.4396, 1.6488, -1.2598, -0.1587, -1.4267, 0.8792, 0.1771],
  7317. [-0.1792, -0.8918, 1.1804, -1.4088, -0.4414, -1.3778, 0.3201, 0.2816],
  7318. [ 0.9521, -0.1891, 1.8129, -0.1594, -0.1669, 0.0568, 0.5431, 0.2250],
  7319. [ 0.5457, -0.4516, 1.6558, 0.1544, -0.5624, -0.3605, 0.2724, 0.1758],
  7320. [ 0.7240, -0.3286, 1.6456, -0.6170, -0.5346, -0.1263, 0.6180, 0.1891]],
  7321. device='cuda:0', grad_fn=<AddmmBackward>)
  7322. landmarks are: tensor([[[ 0.5076, -0.4443, 1.2337, -0.5023, -0.6808, -0.3614, 0.0866,
  7323. 0.2386],
  7324. [ 0.6059, -0.4442, 1.7268, -0.0149, -0.2998, 0.1775, 0.9521,
  7325. 0.1661],
  7326. [ 0.5483, -0.3991, 0.8030, -1.2159, -0.5018, -1.1928, 0.2624,
  7327. 0.3852],
  7328. [ 0.6059, -0.4080, 1.5594, -1.2928, -0.0861, -1.0542, 0.9485,
  7329. 0.3157],
  7330. [ 0.6115, -0.3924, 1.0109, -1.3005, -0.4383, -1.0619, 0.5201,
  7331. 0.3161],
  7332. [ 0.5707, -0.4017, 1.7961, -0.1535, -0.0515, 0.3238, 0.5663,
  7333. 0.4162],
  7334. [ 0.5900, -0.4377, 1.6113, 0.3623, -0.5018, -0.3229, 0.3700,
  7335. 0.1544],
  7336. [ 0.6108, -0.4273, 1.8711, -0.6616, -0.5769, -0.1997, 0.9156,
  7337. 0.1554]]], device='cuda:0')
  7338. loss_train_step before backward: tensor(0.0481, device='cuda:0', grad_fn=<MseLossBackward>)
  7339. loss_train_step after backward: tensor(0.0481, device='cuda:0', grad_fn=<MseLossBackward>)
  7340. loss_train: 5.844778308644891
  7341. step: 74
  7342. running loss: 0.07898349065736339
  7343.  
  7344. Train Steps: 74/90 Loss: 0.0790 torch.Size([8, 600, 800])
  7345. torch.Size([8, 8])
  7346. tensor([[0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656],
  7347. [0.6353, 0.4128, 0.8488, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540],
  7348. [0.6236, 0.4084, 0.7738, 0.2133, 0.3663, 0.3233, 0.5813, 0.5567],
  7349. [ nan, nan, 0.7515, 0.2708, 0.3987, 0.2267, 0.5162, 0.5567],
  7350. [0.6064, 0.4019, 0.8650, 0.4517, 0.4037, 0.5367, 0.5703, 0.5609],
  7351. [0.6218, 0.4137, 0.7263, 0.2233, 0.4075, 0.2650, 0.6212, 0.5783],
  7352. [0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266],
  7353. [0.6153, 0.4119, 0.8463, 0.3833, 0.3600, 0.3200, 0.5106, 0.5563]],
  7354. device='cuda:0', dtype=torch.float64)
  7355. predictions are: tensor([[ 0.6634, -0.3417, 1.6444, -0.0885, -0.5261, 0.1245, 0.4036, 0.2044],
  7356. [ 0.0769, -0.6972, 1.6346, -1.2106, 0.0788, -1.1002, 0.9462, 0.1777],
  7357. [ 0.7241, -0.3058, 1.3758, -0.9954, -0.4782, -0.8373, 0.4019, 0.2229],
  7358. [-0.9479, -1.3831, 1.1639, -1.1251, -0.3941, -1.1239, 0.1696, 0.2574],
  7359. [ 0.7407, -0.2983, 1.7178, 0.0877, -0.3567, 0.4966, 0.5218, 0.1839],
  7360. [ 0.7174, -0.3030, 1.2691, -0.8878, -0.4343, -0.9197, 0.4034, 0.2610],
  7361. [ 0.7942, -0.2631, 1.2833, -0.8850, -0.3832, -0.8959, 0.4042, 0.2976],
  7362. [ 0.4527, -0.4678, 1.5730, -0.3388, -0.6123, -0.5245, 0.2143, 0.1994]],
  7363. device='cuda:0', grad_fn=<AddmmBackward>)
  7364. landmarks are: tensor([[[ 0.5283, -0.4429, 1.5940, -0.2844, -0.5827, -0.1458, 0.2823,
  7365. 0.3267],
  7366. [ 0.6487, -0.3792, 1.6344, -1.0850, 0.2659, -1.5397, 0.8059,
  7367. 0.2730],
  7368. [ 0.5946, -0.3995, 1.2880, -1.3005, -0.5942, -0.7925, 0.3988,
  7369. 0.2853],
  7370. [-2.2859, -2.2859, 1.1854, -1.0352, -0.4441, -1.2390, 0.0986,
  7371. 0.2853],
  7372. [ 0.5151, -0.4296, 1.7095, -0.1997, -0.4210, 0.1929, 0.3484,
  7373. 0.3047],
  7374. [ 0.5863, -0.3749, 1.0686, -1.2543, -0.4037, -1.0619, 0.5836,
  7375. 0.3854],
  7376. [ 0.5746, -0.4021, 1.0801, -1.1312, -0.3229, -1.1081, 0.4803,
  7377. 0.6084],
  7378. [ 0.5561, -0.3834, 1.6229, -0.5153, -0.6231, -0.8079, 0.0727,
  7379. 0.2837]]], device='cuda:0')
  7380. loss_train_step before backward: tensor(0.0734, device='cuda:0', grad_fn=<MseLossBackward>)
  7381. loss_train_step after backward: tensor(0.0734, device='cuda:0', grad_fn=<MseLossBackward>)
  7382. loss_train: 5.9181370083242655
  7383. step: 75
  7384. running loss: 0.07890849344432355
  7385. Train Steps: 75/90 Loss: 0.0789 torch.Size([8, 600, 800])
  7386. torch.Size([8, 8])
  7387. tensor([[0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892],
  7388. [0.6185, 0.4129, 0.8900, 0.4567, 0.3937, 0.5417, 0.5734, 0.5110],
  7389. [0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183],
  7390. [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895],
  7391. [0.6219, 0.3934, 0.8688, 0.5267, 0.4313, 0.4967, 0.5988, 0.4983],
  7392. [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650],
  7393. [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290],
  7394. [ nan, nan, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733]],
  7395. device='cuda:0', dtype=torch.float64)
  7396. predictions are: tensor([[ 0.4961, -0.4331, 1.5760, -0.2348, -0.5834, -0.4121, 0.2109, 0.2769],
  7397. [ 0.7604, -0.2908, 1.7494, -0.1816, -0.3328, 0.1616, 0.4631, 0.2463],
  7398. [ 0.6928, -0.3400, 1.6568, -0.1040, -0.4998, -0.2940, 0.4876, 0.1865],
  7399. [ 0.5090, -0.4994, 1.2263, -1.5922, -0.1983, -1.5729, 0.5735, 0.2009],
  7400. [ 0.4650, -0.4823, 1.6588, -0.0199, -0.2635, -0.0374, 0.3917, 0.1881],
  7401. [ 0.7070, -0.3193, 1.6073, -0.4942, -0.5277, -0.3145, 0.2996, 0.2586],
  7402. [ 0.6871, -0.3422, 1.4430, -0.9179, -0.5518, -0.4318, 0.6278, 0.2200],
  7403. [-0.7194, -1.2560, 1.0205, -1.5431, -0.1226, -1.5735, 0.3658, 0.3317]],
  7404. device='cuda:0', grad_fn=<AddmmBackward>)
  7405. landmarks are: tensor([[[ 0.5454, -0.4053, 1.6633, -0.1766, -0.6058, -0.5923, 0.1577,
  7406. 0.4357],
  7407. [ 0.5711, -0.3788, 1.8249, -0.1766, -0.4672, 0.2160, 0.3625,
  7408. 0.0742],
  7409. [ 0.5944, -0.4543, 1.8018, 0.0082, -0.6058, -0.4306, 0.4162,
  7410. 0.1082],
  7411. [ 0.5828, -0.4417, 1.2476, -1.3929, -0.1727, -1.5700, 0.4694,
  7412. -0.0248],
  7413. [ 0.5864, -0.4690, 1.7268, 0.1467, -0.2940, 0.0082, 0.4797,
  7414. 0.0159],
  7415. [ 0.5757, -0.3917, 1.7095, -0.4768, -0.6346, -0.4229, 0.3931,
  7416. 0.3238],
  7417. [ 0.5697, -0.4538, 1.5308, -0.8703, -0.6572, -0.3639, 0.5739,
  7418. 0.1576],
  7419. [-2.2859, -2.2859, 0.7106, -1.4468, -0.2882, -1.4237, 0.2430,
  7420. 0.3623]]], device='cuda:0')
  7421. loss_train_step before backward: tensor(0.0671, device='cuda:0', grad_fn=<MseLossBackward>)
  7422. loss_train_step after backward: tensor(0.0671, device='cuda:0', grad_fn=<MseLossBackward>)
  7423. loss_train: 5.985258931294084
  7424. step: 76
  7425. running loss: 0.07875340699071162
  7426. Train Steps: 76/90 Loss: 0.0788 torch.Size([8, 600, 800])
  7427. torch.Size([8, 8])
  7428. tensor([[0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092],
  7429. [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329],
  7430. [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609],
  7431. [0.6111, 0.3995, 0.8788, 0.4567, 0.3813, 0.4833, 0.5450, 0.5700],
  7432. [0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835],
  7433. [0.6261, 0.4066, 0.8325, 0.2150, 0.4763, 0.2667, 0.7002, 0.5633],
  7434. [0.6200, 0.3999, 0.8653, 0.5207, 0.4100, 0.5125, 0.5975, 0.5103],
  7435. [0.6090, 0.4010, 0.7838, 0.3483, 0.3538, 0.3783, 0.5462, 0.5077]],
  7436. device='cuda:0', dtype=torch.float64)
  7437. predictions are: tensor([[ 0.2429, -0.6192, 1.5609, 0.0381, -0.3987, 0.1600, 0.2758, 0.2319],
  7438. [ 0.6422, -0.3465, 1.4123, -0.8676, -0.6071, -0.7099, 0.3021, 0.2425],
  7439. [ 0.4155, -0.4701, 1.5756, -1.0316, -0.1855, -0.9949, 0.7275, 0.2318],
  7440. [ 0.2456, -0.6119, 1.5407, -0.1703, -0.4710, 0.0538, 0.3149, 0.2587],
  7441. [ 0.4306, -0.5004, 1.4961, -0.8191, -0.3195, -1.1378, 0.4181, 0.2407],
  7442. [ 0.6043, -0.4138, 1.4270, -1.2980, -0.1460, -1.2675, 0.8263, 0.2275],
  7443. [ 0.3764, -0.5286, 1.5590, 0.0285, -0.3335, 0.1087, 0.2864, 0.2451],
  7444. [ 0.3725, -0.5430, 1.1990, -0.9545, -0.6115, -0.5813, 0.2128, 0.2504]],
  7445. device='cuda:0', grad_fn=<AddmmBackward>)
  7446. landmarks are: tensor([[[ 0.5783, -0.4363, 1.7724, 0.0432, -0.5153, 0.0871, 0.4840,
  7447. 0.0663],
  7448. [ 0.5680, -0.4393, 1.5920, -0.6672, -0.6453, -0.5457, 0.5149,
  7449. 0.1753],
  7450. [ 0.6026, -0.4032, 1.7326, -0.7771, -0.3691, -0.8617, 0.9704,
  7451. 0.3050],
  7452. [ 0.5368, -0.4406, 1.7730, -0.1766, -0.5249, -0.0534, 0.2314,
  7453. 0.3469],
  7454. [ 0.6117, -0.3877, 1.8249, -0.5230, -0.3517, -1.2851, 0.4200,
  7455. -0.0529],
  7456. [ 0.6059, -0.4080, 1.5594, -1.2928, -0.0861, -1.0542, 0.9485,
  7457. 0.3157],
  7458. [ 0.5778, -0.4389, 1.7107, 0.1192, -0.3921, 0.0815, 0.4741,
  7459. 0.0711],
  7460. [ 0.5270, -0.4336, 1.3342, -0.6770, -0.6520, -0.5384, 0.2370,
  7461. 0.0592]]], device='cuda:0')
  7462. loss_train_step before backward: tensor(0.0258, device='cuda:0', grad_fn=<MseLossBackward>)
  7463. loss_train_step after backward: tensor(0.0258, device='cuda:0', grad_fn=<MseLossBackward>)
  7464. loss_train: 6.011011159047484
  7465. step: 77
  7466. running loss: 0.07806507998762967
  7467. Train Steps: 77/90 Loss: 0.0781 torch.Size([8, 600, 800])
  7468. torch.Size([8, 8])
  7469. tensor([[0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285],
  7470. [0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124],
  7471. [0.6289, 0.4019, 0.8113, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332],
  7472. [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391],
  7473. [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795],
  7474. [0.6205, 0.4062, 0.8337, 0.2683, 0.3675, 0.4283, 0.6338, 0.5250],
  7475. [0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461],
  7476. [0.6207, 0.4110, 0.8738, 0.5000, 0.4800, 0.5633, 0.6300, 0.5433]],
  7477. device='cuda:0', dtype=torch.float64)
  7478. predictions are: tensor([[ 0.4532, -0.4790, 1.3550, -0.6564, -0.5215, -0.4282, 0.2318, 0.2345],
  7479. [ 0.3515, -0.5440, 1.4068, -1.3205, -0.1814, -1.4695, 0.5088, 0.2038],
  7480. [ 0.1470, -0.6419, 1.4864, -0.0764, -0.4137, -0.0734, 0.2511, 0.2779],
  7481. [ 0.4911, -0.4676, 1.3029, -1.1856, -0.4020, -1.2807, 0.5175, 0.2134],
  7482. [ 0.4871, -0.4626, 1.7018, 0.0512, -0.3980, 0.0687, 0.4746, 0.2630],
  7483. [ 0.3120, -0.5893, 1.3329, -0.9873, -0.5125, -0.4713, 0.4975, 0.2488],
  7484. [ 0.5257, -0.4181, 1.4685, -0.6656, -0.6150, -0.5175, 0.4017, 0.2125],
  7485. [ 0.6028, -0.3793, 1.7381, -0.1580, -0.0723, -0.1443, 0.4997, 0.2906]],
  7486. device='cuda:0', grad_fn=<AddmmBackward>)
  7487. landmarks are: tensor([[[ 0.5253, -0.4296, 1.3084, -0.3098, -0.6115, -0.2767, 0.0928,
  7488. 0.1552],
  7489. [ 0.5790, -0.4079, 1.5929, -1.0630, -0.4729, -1.0725, 0.4137,
  7490. 0.0807],
  7491. [ 0.6191, -0.4297, 1.4612, 0.2391, -0.4961, 0.0313, 1.1166,
  7492. 0.1768],
  7493. [ 0.5785, -0.4259, 1.4228, -1.0261, -0.4190, -1.2189, 0.4763,
  7494. 0.2043],
  7495. [ 0.6355, -0.4080, 1.6113, 0.1852, -0.4730, 0.1467, 0.9996,
  7496. 0.3905],
  7497. [ 0.5799, -0.4099, 1.5651, -1.0465, -0.5885, -0.3075, 0.6413,
  7498. 0.1390],
  7499. [ 0.5124, -0.4483, 1.5420, -0.6231, -0.7155, -0.4152, 0.2679,
  7500. 0.2365],
  7501. [ 0.5811, -0.3878, 1.7499, 0.0236, -0.0688, 0.3161, 0.6240,
  7502. 0.2237]]], device='cuda:0')
  7503. loss_train_step before backward: tensor(0.0425, device='cuda:0', grad_fn=<MseLossBackward>)
  7504. loss_train_step after backward: tensor(0.0425, device='cuda:0', grad_fn=<MseLossBackward>)
  7505. loss_train: 6.053500270470977
  7506. step: 78
  7507. running loss: 0.07760897782655099
  7508.  
  7509. Train Steps: 78/90 Loss: 0.0776 torch.Size([8, 600, 800])
  7510. torch.Size([8, 8])
  7511. tensor([[0.6273, 0.4143, 0.8750, 0.5700, 0.3987, 0.4717, 0.6013, 0.5467],
  7512. [0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317],
  7513. [0.6109, 0.4041, 0.6975, 0.3167, 0.3513, 0.3383, 0.5153, 0.5319],
  7514. [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377],
  7515. [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5837, 0.5500],
  7516. [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333],
  7517. [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550],
  7518. [0.6110, 0.3984, 0.8750, 0.4933, 0.4625, 0.4950, 0.5578, 0.5676]],
  7519. device='cuda:0', dtype=torch.float64)
  7520. predictions are: tensor([[ 0.4413, -0.4662, 1.6231, -0.0098, -0.5385, -0.2463, 0.3870, 0.2384],
  7521. [-0.1857, -0.8836, 1.1719, -1.4335, -0.3002, -1.3386, 0.4504, 0.2599],
  7522. [ 0.3408, -0.5526, 1.1232, -1.1451, -0.6713, -0.8068, 0.2262, 0.2412],
  7523. [ 0.4423, -0.5052, 1.7007, 0.0671, -0.4582, -0.0958, 0.4439, 0.1918],
  7524. [ 0.5977, -0.3991, 1.6609, -0.2335, -0.4152, -0.0203, 0.4947, 0.2128],
  7525. [ 0.4420, -0.4872, 1.0551, -1.5288, -0.4055, -1.4832, 0.4326, 0.2587],
  7526. [ 0.4850, -0.4208, 1.7230, -0.2698, -0.1252, -0.2177, 0.5447, 0.2066],
  7527. [ 0.7261, -0.3160, 1.7007, -0.2241, -0.2409, -0.1268, 0.4922, 0.2211]],
  7528. device='cuda:0', grad_fn=<AddmmBackward>)
  7529. landmarks are: tensor([[[ 6.1149e-01, -3.7244e-01, 1.7557e+00, 3.4688e-01, -4.4411e-01,
  7530. -1.0731e-01, 4.9122e-01, 2.3911e-01],
  7531. [ 5.8487e-01, -3.8360e-01, 1.2649e+00, -1.2236e+00, -3.4596e-01,
  7532. -1.2313e+00, 4.5081e-01, 1.6982e-01],
  7533. [ 5.3591e-01, -4.1932e-01, 9.3580e-01, -8.2325e-01, -6.6351e-01,
  7534. -7.2317e-01, 9.4325e-02, 1.7099e-01],
  7535. [ 5.7748e-01, -4.6066e-01, 1.6741e+00, 1.9623e-01, -4.0362e-01,
  7536. -1.2115e-01, 4.5876e-01, 1.9786e-01],
  7537. [ 5.6039e-01, -4.6197e-01, 1.7961e+00, -1.9969e-01, -3.2286e-01,
  7538. 1.0824e-01, 4.1039e-01, 2.5450e-01],
  7539. [ 5.8528e-01, -3.9199e-01, 1.1090e+00, -1.3313e+00, -2.8822e-01,
  7540. -1.3390e+00, 4.6236e-01, 1.7752e-01],
  7541. [ 5.3508e-01, -4.3212e-01, 1.7326e+00, -1.3811e-01, 4.6329e-02,
  7542. 1.0491e-01, 4.1617e-01, 2.7760e-01],
  7543. [ 5.3637e-01, -4.4573e-01, 1.7557e+00, -7.2363e-03, -1.4965e-01,
  7544. 4.6189e-04, 2.9049e-01, 3.3573e-01]]], device='cuda:0')
  7545. loss_train_step before backward: tensor(0.0302, device='cuda:0', grad_fn=<MseLossBackward>)
  7546. loss_train_step after backward: tensor(0.0302, device='cuda:0', grad_fn=<MseLossBackward>)
  7547. loss_train: 6.083718063309789
  7548. step: 79
  7549. running loss: 0.07700908940898467
  7550. Train Steps: 79/90 Loss: 0.0770 torch.Size([8, 600, 800])
  7551. torch.Size([8, 8])
  7552. tensor([[0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837],
  7553. [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973],
  7554. [0.6296, 0.4045, 0.9138, 0.4100, 0.4232, 0.4242, 0.7422, 0.5297],
  7555. [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261],
  7556. [0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835],
  7557. [0.6160, 0.4093, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808],
  7558. [0.6125, 0.3974, 0.7725, 0.2517, 0.3538, 0.3317, 0.5887, 0.5500],
  7559. [0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742]],
  7560. device='cuda:0', dtype=torch.float64)
  7561. predictions are: tensor([[ 0.1627, -0.6941, 0.9786, -1.1865, -0.5091, -0.9566, 0.1641, 0.3193],
  7562. [ 0.4703, -0.4958, 1.7074, 0.0998, -0.4157, -0.0632, 0.4524, 0.1856],
  7563. [ 0.6864, -0.3392, 1.5980, -0.5553, -0.4134, -0.3026, 0.6323, 0.1914],
  7564. [ 0.4189, -0.4609, 1.6604, -0.2478, -0.3532, -0.0982, 0.3636, 0.2113],
  7565. [ 0.6215, -0.3834, 1.6149, -0.8199, -0.2564, -1.1783, 0.5599, 0.1797],
  7566. [ 0.3355, -0.5244, 1.5251, -0.3258, -0.3835, -0.0800, 0.3727, 0.2703],
  7567. [ 0.4983, -0.4797, 1.2142, -1.1775, -0.5485, -0.8428, 0.4820, 0.2132],
  7568. [ 0.3679, -0.5232, 1.6810, -0.1140, -0.4568, -0.6179, 0.4762, 0.1681]],
  7569. device='cuda:0', grad_fn=<AddmmBackward>)
  7570. landmarks are: tensor([[[ 0.5629, -0.3871, 0.7243, -0.9581, -0.5827, -0.9849, 0.1288,
  7571. 0.4103],
  7572. [ 0.6240, -0.4321, 1.8423, 0.1852, -0.5885, -0.1612, 0.6962,
  7573. 0.0111],
  7574. [ 0.6224, -0.4177, 1.9346, -0.3921, -0.3314, -0.3264, 1.1422,
  7575. 0.1608],
  7576. [ 0.5746, -0.3623, 1.7961, -0.1150, -0.3691, -0.0380, 0.2208,
  7577. 0.1439],
  7578. [ 0.6117, -0.3877, 1.8249, -0.5230, -0.3517, -1.2851, 0.4200,
  7579. -0.0529],
  7580. [ 0.5592, -0.3956, 1.5543, -0.2456, -0.5885, -0.1689, 0.1392,
  7581. 0.3968],
  7582. [ 0.5430, -0.4503, 1.2822, -1.1235, -0.6520, -0.7540, 0.4335,
  7583. 0.2545],
  7584. [ 0.6113, -0.4057, 1.7499, 0.3007, -0.5885, -0.5384, 0.4513,
  7585. -0.0957]]], device='cuda:0')
  7586. loss_train_step before backward: tensor(0.0294, device='cuda:0', grad_fn=<MseLossBackward>)
  7587. loss_train_step after backward: tensor(0.0294, device='cuda:0', grad_fn=<MseLossBackward>)
  7588. loss_train: 6.113160545006394
  7589. step: 80
  7590. running loss: 0.07641450681257993
  7591. Train Steps: 80/90 Loss: 0.0764 torch.Size([8, 600, 800])
  7592. torch.Size([8, 8])
  7593. tensor([[0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459],
  7594. [0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938],
  7595. [0.6134, 0.4090, 0.6926, 0.2819, 0.3538, 0.3233, 0.5563, 0.5667],
  7596. [0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904],
  7597. [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583],
  7598. [0.6189, 0.4029, 0.8375, 0.5767, 0.4745, 0.4829, 0.5551, 0.5598],
  7599. [0.6204, 0.4055, 0.8438, 0.5733, 0.4574, 0.4801, 0.5487, 0.5617],
  7600. [0.6126, 0.4067, 0.8638, 0.5383, 0.4188, 0.4850, 0.5016, 0.5392]],
  7601. device='cuda:0', dtype=torch.float64)
  7602. predictions are: tensor([[ 0.6616, -0.3725, 1.5009, -1.1986, -0.1821, -1.1744, 0.8349, 0.1502],
  7603. [ 0.5140, -0.4440, 1.6731, 0.1572, -0.5335, -0.1048, 0.4126, 0.1451],
  7604. [ 0.2424, -0.6267, 1.0445, -1.1448, -0.6994, -0.8297, 0.1809, 0.2472],
  7605. [ 0.5195, -0.4319, 1.5449, -1.2729, -0.2066, -1.1167, 0.6421, 0.1443],
  7606. [ 0.6524, -0.3687, 1.5164, -0.6111, -0.6951, -0.3538, 0.4975, 0.2012],
  7607. [ 0.4303, -0.4790, 1.5850, 0.0598, -0.2176, -0.2040, 0.3833, 0.2430],
  7608. [ 0.2529, -0.5771, 1.5279, 0.0467, -0.2950, -0.1728, 0.3096, 0.2634],
  7609. [ 0.3993, -0.4989, 1.6358, 0.0904, -0.4416, -0.0155, 0.3070, 0.2172]],
  7610. device='cuda:0', grad_fn=<AddmmBackward>)
  7611. landmarks are: tensor([[[ 0.6520, -0.3912, 1.7095, -0.9079, -0.0284, -1.3621, 0.8096,
  7612. 0.2356],
  7613. [ 0.6395, -0.4213, 1.7037, 0.3623, -0.4326, -0.1073, 0.6560,
  7614. -0.0049],
  7615. [ 0.5473, -0.3966, 0.9131, -0.9838, -0.6520, -0.7925, 0.2834,
  7616. 0.3315],
  7617. [ 0.6126, -0.3871, 1.6229, -1.0773, -0.2132, -1.3698, 0.5829,
  7618. -0.0209],
  7619. [ 0.5761, -0.3966, 1.6171, -0.4845, -0.6346, -0.4691, 0.4739,
  7620. 0.2930],
  7621. [ 0.5726, -0.4249, 1.5824, 0.3777, -0.0942, -0.0556, 0.2782,
  7622. 0.2997],
  7623. [ 0.5795, -0.4129, 1.6113, 0.3623, -0.1733, -0.0684, 0.2487,
  7624. 0.3084],
  7625. [ 0.5436, -0.4076, 1.7037, 0.2006, -0.3517, -0.0457, 0.0311,
  7626. 0.2048]]], device='cuda:0')
  7627. loss_train_step before backward: tensor(0.0210, device='cuda:0', grad_fn=<MseLossBackward>)
  7628. loss_train_step after backward: tensor(0.0210, device='cuda:0', grad_fn=<MseLossBackward>)
  7629. loss_train: 6.13416051492095
  7630. step: 81
  7631. running loss: 0.07573037672741914
  7632. Train Steps: 81/90 Loss: 0.0757 torch.Size([8, 600, 800])
  7633. torch.Size([8, 8])
  7634. tensor([[ nan, nan, 0.7412, 0.2200, 0.4450, 0.1517, 0.5312, 0.4983],
  7635. [0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367],
  7636. [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672],
  7637. [0.6122, 0.4006, 0.8850, 0.4217, 0.4088, 0.5517, 0.6063, 0.5517],
  7638. [0.6250, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6088, 0.5183],
  7639. [0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240],
  7640. [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212],
  7641. [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309]],
  7642. device='cuda:0', dtype=torch.float64)
  7643. predictions are: tensor([[-0.8470, -1.3406, 0.9710, -1.2401, -0.3874, -1.3271, 0.1391, 0.1894],
  7644. [ 0.9333, -0.1303, 1.5740, -0.8103, -0.3536, -0.8830, 0.5989, 0.1798],
  7645. [ 0.9719, -0.1713, 1.7116, 0.3834, -0.6196, -0.0734, 0.4194, 0.1456],
  7646. [ 0.8045, -0.2672, 1.7701, -0.0866, -0.4186, 0.3582, 0.5741, 0.2192],
  7647. [ 0.7323, -0.2955, 1.8138, 0.0565, -0.1777, 0.2783, 0.5645, 0.2337],
  7648. [-0.4810, -1.0622, 1.0033, -1.1718, -0.4243, -1.2152, 0.1338, 0.2037],
  7649. [ 0.8303, -0.2347, 1.4033, -0.9560, -0.4656, -0.8805, 0.6346, 0.1497],
  7650. [ 0.6676, -0.2687, 1.7781, 0.0762, -0.3945, -0.3554, 0.4669, 0.1738]],
  7651. device='cuda:0', grad_fn=<AddmmBackward>)
  7652. landmarks are: tensor([[[-2.2859, -2.2859, 1.1379, -1.2697, -0.2305, -1.5854, 0.1679,
  7653. 0.0159],
  7654. [ 0.6174, -0.4201, 1.7309, -0.8784, -0.4735, -0.9524, 0.6242,
  7655. 0.1931],
  7656. [ 0.6128, -0.4375, 1.7788, 0.2699, -0.6346, -0.2536, 0.5463,
  7657. -0.1278],
  7658. [ 0.5417, -0.4355, 1.8018, -0.3383, -0.3979, 0.2622, 0.5143,
  7659. 0.2622],
  7660. [ 0.6012, -0.3846, 1.7326, -0.0457, -0.1381, 0.2853, 0.5259,
  7661. 0.1082],
  7662. [ 0.5676, -0.4112, 1.1898, -1.2467, -0.2940, -1.4622, 0.2103,
  7663. 0.1343],
  7664. [ 0.6125, -0.4345, 1.4308, -1.1384, -0.4213, -1.0031, 0.7190,
  7665. 0.1214],
  7666. [ 0.6520, -0.3623, 1.8885, 0.0313, -0.5538, -0.5384, 0.6926,
  7667. 0.1661]]], device='cuda:0')
  7668. loss_train_step before backward: tensor(0.0915, device='cuda:0', grad_fn=<MseLossBackward>)
  7669. loss_train_step after backward: tensor(0.0915, device='cuda:0', grad_fn=<MseLossBackward>)
  7670. loss_train: 6.2256993018090725
  7671. step: 82
  7672. running loss: 0.07592316221718381
  7673.  
  7674. Train Steps: 82/90 Loss: 0.0759 torch.Size([8, 600, 800])
  7675. torch.Size([8, 8])
  7676. tensor([[0.6053, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168],
  7677. [0.6204, 0.4007, 0.7838, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167],
  7678. [0.6156, 0.4125, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084],
  7679. [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5413, 0.5717],
  7680. [0.6179, 0.4082, 0.6688, 0.2667, 0.3588, 0.3317, 0.5750, 0.5783],
  7681. [0.6261, 0.3987, 0.8688, 0.4917, 0.4300, 0.5333, 0.7010, 0.5309],
  7682. [0.6164, 0.4102, 0.8850, 0.4867, 0.4213, 0.5983, 0.5609, 0.5038],
  7683. [0.6262, 0.4163, 0.8850, 0.5183, 0.3763, 0.4150, 0.6025, 0.5500]],
  7684. device='cuda:0', dtype=torch.float64)
  7685. predictions are: tensor([[-0.4372, -1.0773, 0.9684, -1.1526, -0.4889, -1.4327, 0.1665, 0.1314],
  7686. [ 0.8245, -0.2356, 1.2412, -1.2070, -0.3522, -1.4770, 0.4970, 0.1102],
  7687. [ 0.6625, -0.3387, 1.8488, 0.0791, -0.2400, 0.1239, 0.5469, 0.1703],
  7688. [ 0.6149, -0.3869, 1.8176, 0.0247, -0.3495, -0.1160, 0.5422, 0.1320],
  7689. [ 0.5138, -0.4426, 1.1079, -0.9809, -0.7097, -0.8575, 0.2407, 0.2462],
  7690. [ 0.6288, -0.3711, 1.8621, 0.0934, -0.3120, 0.1545, 0.6857, 0.1629],
  7691. [ 0.4593, -0.4756, 1.8160, 0.1128, -0.2644, 0.2836, 0.5064, 0.1714],
  7692. [ 0.5796, -0.3572, 1.7196, 0.0223, -0.5261, -0.5082, 0.3354, 0.1745]],
  7693. device='cuda:0', grad_fn=<AddmmBackward>)
  7694. landmarks are: tensor([[[ 0.5102, -0.4223, 0.8999, -1.4011, -0.4383, -1.3082, 0.2267,
  7695. 0.1013],
  7696. [ 0.5796, -0.4351, 1.3342, -1.3159, -0.2189, -1.4853, 0.4046,
  7697. 0.1005],
  7698. [ 0.5573, -0.3808, 1.8018, -0.0534, -0.1843, 0.4008, 0.3769,
  7699. 0.0622],
  7700. [ 0.5432, -0.4462, 1.7557, -0.0380, -0.3113, -0.0765, 0.2141,
  7701. 0.3546],
  7702. [ 0.5680, -0.4005, 0.8030, -1.0542, -0.6289, -0.7540, 0.3700,
  7703. 0.3854],
  7704. [ 0.6059, -0.4442, 1.7268, -0.0149, -0.2998, 0.1775, 0.9521,
  7705. 0.1661],
  7706. [ 0.5614, -0.3912, 1.8018, -0.0380, -0.3402, 0.4778, 0.3049,
  7707. 0.0412],
  7708. [ 0.6066, -0.3632, 1.8018, 0.1082, -0.5480, -0.3691, 0.4970,
  7709. 0.2545]]], device='cuda:0')
  7710. loss_train_step before backward: tensor(0.0369, device='cuda:0', grad_fn=<MseLossBackward>)
  7711. loss_train_step after backward: tensor(0.0369, device='cuda:0', grad_fn=<MseLossBackward>)
  7712. loss_train: 6.262647982686758
  7713. step: 83
  7714. running loss: 0.07545359015285251
  7715. Train Steps: 83/90 Loss: 0.0755 torch.Size([8, 600, 800])
  7716. torch.Size([8, 8])
  7717. tensor([[0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433],
  7718. [0.6276, 0.4120, 0.8738, 0.3133, 0.4225, 0.2217, 0.6203, 0.4892],
  7719. [0.6124, 0.4075, 0.7696, 0.4153, 0.3475, 0.3767, 0.5157, 0.5427],
  7720. [0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317],
  7721. [0.6266, 0.4067, 0.8588, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267],
  7722. [0.6339, 0.4149, 0.8800, 0.5000, 0.3900, 0.5283, 0.7541, 0.5424],
  7723. [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760],
  7724. [0.6206, 0.4001, 0.8900, 0.3933, 0.3588, 0.3567, 0.5837, 0.5083]],
  7725. device='cuda:0', dtype=torch.float64)
  7726. predictions are: tensor([[ 0.5588, -0.4253, 1.3032, -0.8714, -0.5930, -0.6349, 0.4121, 0.1612],
  7727. [ 0.5840, -0.3910, 1.6718, -0.5996, -0.2422, -0.9303, 0.5476, 0.1022],
  7728. [ 0.5503, -0.4172, 1.4038, -0.4003, -0.5918, -0.4394, 0.2558, 0.1828],
  7729. [ 0.0503, -0.7139, 1.2194, -0.9917, -0.2453, -1.0379, 0.3115, 0.1644],
  7730. [ 0.6324, -0.3348, 1.6029, -0.8426, -0.1930, -0.8667, 0.6539, 0.1151],
  7731. [ 0.6758, -0.3652, 1.8777, 0.4324, -0.4339, 0.3469, 0.6145, 0.1447],
  7732. [ 0.5173, -0.4611, 1.6718, 0.3990, -0.4012, 0.1117, 0.4887, 0.2058],
  7733. [ 0.4877, -0.4613, 1.7446, -0.1365, -0.5940, -0.5164, 0.3340, 0.0479]],
  7734. device='cuda:0', grad_fn=<AddmmBackward>)
  7735. landmarks are: tensor([[[ 0.5665, -0.3963, 1.3457, -0.9926, -0.6058, -0.6308, 0.5721,
  7736. 0.2237],
  7737. [ 0.6128, -0.3828, 1.7499, -0.8386, -0.3344, -1.2620, 0.5792,
  7738. -0.0263],
  7739. [ 0.5427, -0.4035, 1.2688, -0.3675, -0.6808, -0.5461, 0.0959,
  7740. 0.2206],
  7741. [ 0.5849, -0.3836, 1.2649, -1.2236, -0.3460, -1.2313, 0.4508,
  7742. 0.1698],
  7743. [ 0.6084, -0.4076, 1.6806, -0.9618, -0.2998, -0.9695, 0.6356,
  7744. 0.1467],
  7745. [ 0.6421, -0.3695, 1.7788, 0.0236, -0.4845, 0.1544, 1.1971,
  7746. 0.2196],
  7747. [ 0.6421, -0.4008, 1.6055, 0.2160, -0.5076, -0.0534, 1.1020,
  7748. 0.3745],
  7749. [ 0.5807, -0.4378, 1.8249, -0.4691, -0.6289, -0.6385, 0.4104,
  7750. 0.0620]]], device='cuda:0')
  7751. loss_train_step before backward: tensor(0.0327, device='cuda:0', grad_fn=<MseLossBackward>)
  7752. loss_train_step after backward: tensor(0.0327, device='cuda:0', grad_fn=<MseLossBackward>)
  7753. loss_train: 6.2953284196555614
  7754. step: 84
  7755. running loss: 0.0749443859482805
  7756. Train Steps: 84/90 Loss: 0.0749 torch.Size([8, 600, 800])
  7757. torch.Size([8, 8])
  7758. tensor([[0.6171, 0.4127, 0.8900, 0.4800, 0.4325, 0.5783, 0.5769, 0.5090],
  7759. [0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6262, 0.5167],
  7760. [0.6277, 0.4057, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823],
  7761. [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317],
  7762. [0.6280, 0.4101, 0.9050, 0.4533, 0.3775, 0.3217, 0.6338, 0.4915],
  7763. [0.6264, 0.3972, 0.8853, 0.4771, 0.3853, 0.4511, 0.6293, 0.5334],
  7764. [0.6075, 0.4000, 0.8513, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280],
  7765. [0.6167, 0.4048, 0.6831, 0.3639, 0.3763, 0.3017, 0.5700, 0.5883]],
  7766. device='cuda:0', dtype=torch.float64)
  7767. predictions are: tensor([[ 0.5941, -0.4105, 1.7887, 0.0018, -0.3170, 0.2443, 0.5468, 0.1424],
  7768. [ 0.8466, -0.2356, 1.7616, -0.3460, -0.4721, -0.6073, 0.5775, 0.1030],
  7769. [ 0.6348, -0.4047, 1.4215, -1.2086, -0.3035, -1.2415, 0.7195, 0.0656],
  7770. [ 0.6285, -0.3556, 1.6495, 0.1077, -0.5253, -0.5284, 0.3942, 0.1090],
  7771. [ 0.5837, -0.3673, 1.7452, -0.1876, -0.4923, -0.7255, 0.4623, 0.0394],
  7772. [ 0.6060, -0.3984, 1.7194, -0.0774, -0.5157, -0.1249, 0.5227, 0.1249],
  7773. [ 0.3936, -0.5132, 1.6742, 0.0635, -0.1754, 0.0217, 0.4127, 0.1566],
  7774. [-0.0444, -0.7899, 1.1317, -0.8592, -0.4773, -0.8693, 0.1572, 0.2092]],
  7775. device='cuda:0', grad_fn=<AddmmBackward>)
  7776. landmarks are: tensor([[[ 0.5645, -0.3797, 1.8249, -0.0688, -0.2882, 0.3854, 0.3789,
  7777. 0.0652],
  7778. [ 0.6094, -0.3947, 1.8885, -0.2998, -0.5769, -0.6770, 0.6067,
  7779. 0.1005],
  7780. [ 0.6132, -0.4122, 1.5478, -1.0619, -0.2709, -1.4314, 0.5500,
  7781. -0.0583],
  7782. [ 0.5901, -0.4157, 1.7557, 0.1929, -0.5423, -0.5923, 0.3584,
  7783. 0.1698],
  7784. [ 0.6148, -0.3918, 1.8942, -0.1920, -0.5423, -0.8002, 0.6414,
  7785. -0.0156],
  7786. [ 0.6075, -0.4514, 1.8032, -0.0822, -0.5061, -0.2023, 0.6208,
  7787. 0.1779],
  7788. [ 0.5202, -0.4382, 1.6460, 0.1082, -0.2029, 0.1754, 0.1067,
  7789. 0.1530],
  7790. [ 0.5626, -0.4162, 0.8692, -0.6051, -0.5480, -0.8925, 0.3469,
  7791. 0.4316]]], device='cuda:0')
  7792. loss_train_step before backward: tensor(0.0203, device='cuda:0', grad_fn=<MseLossBackward>)
  7793. loss_train_step after backward: tensor(0.0203, device='cuda:0', grad_fn=<MseLossBackward>)
  7794. loss_train: 6.315675836056471
  7795. step: 85
  7796. running loss: 0.0743020686594879
  7797. Train Steps: 85/90 Loss: 0.0743 torch.Size([8, 600, 800])
  7798. torch.Size([8, 8])
  7799. tensor([[0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350],
  7800. [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250],
  7801. [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656],
  7802. [0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297],
  7803. [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947],
  7804. [0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240],
  7805. [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461],
  7806. [0.6127, 0.4118, 0.8650, 0.5083, 0.4088, 0.5367, 0.5300, 0.5456]],
  7807. device='cuda:0', dtype=torch.float64)
  7808. predictions are: tensor([[ 1.0889, -0.0847, 1.4050, -1.0740, -0.3015, -1.3084, 0.4984, 0.0900],
  7809. [ 0.7538, -0.3029, 1.8606, 0.0660, -0.4251, 0.0576, 0.5933, 0.1187],
  7810. [ 0.3395, -0.5567, 1.6117, -0.3520, -0.5353, -0.2820, 0.4529, 0.1650],
  7811. [ 0.7772, -0.2671, 1.8426, 0.1250, -0.5436, -0.4465, 0.5795, 0.0673],
  7812. [ 0.4460, -0.5134, 1.7618, -0.0583, -0.5205, -0.1142, 0.5861, 0.0676],
  7813. [-0.0893, -0.8130, 1.1007, -1.2895, -0.3983, -1.5112, 0.2485, 0.0711],
  7814. [ 0.5626, -0.4279, 1.7289, 0.1548, -0.2305, -0.0760, 0.4865, 0.1444],
  7815. [ 0.4831, -0.4688, 1.6880, 0.0459, -0.3398, -0.0117, 0.4747, 0.1757]],
  7816. device='cuda:0', grad_fn=<AddmmBackward>)
  7817. landmarks are: tensor([[[ 5.7679e-01, -4.0308e-01, 1.3838e+00, -1.1527e+00, -2.1876e-01,
  7818. -1.4216e+00, 4.3790e-01, 1.8502e-01],
  7819. [ 5.9913e-01, -3.8029e-01, 1.8018e+00, -5.3426e-02, -3.4596e-01,
  7820. 1.8522e-01, 5.3741e-01, 1.3903e-01],
  7821. [ 5.2835e-01, -4.4288e-01, 1.5940e+00, -2.8437e-01, -5.8268e-01,
  7822. -1.4580e-01, 2.8226e-01, 3.2671e-01],
  7823. [ 6.5036e-01, -3.6471e-01, 1.7730e+00, 2.9299e-01, -6.0577e-01,
  7824. -2.3818e-01, 7.1085e-01, 1.6077e-01],
  7825. [ 5.8031e-01, -4.3672e-01, 1.8423e+00, -3.0331e-02, -5.9423e-01,
  7826. 2.3557e-02, 6.5034e-01, -9.2270e-04],
  7827. [ 5.6761e-01, -4.1124e-01, 1.1898e+00, -1.2467e+00, -2.9400e-01,
  7828. -1.4622e+00, 2.1029e-01, 1.3434e-01],
  7829. [ 5.4428e-01, -3.8314e-01, 1.7095e+00, 1.6212e-01, -2.0162e-01,
  7830. 1.3903e-01, 1.4368e-01, 2.3637e-01],
  7831. [ 5.4428e-01, -3.8399e-01, 1.7095e+00, 6.2048e-02, -3.9792e-01,
  7832. 1.9292e-01, 1.6218e-01, 2.3412e-01]]], device='cuda:0')
  7833. loss_train_step before backward: tensor(0.0275, device='cuda:0', grad_fn=<MseLossBackward>)
  7834. loss_train_step after backward: tensor(0.0275, device='cuda:0', grad_fn=<MseLossBackward>)
  7835. loss_train: 6.343135943636298
  7836. step: 86
  7837. running loss: 0.07375739469344533
  7838.  
  7839. Train Steps: 86/90 Loss: 0.0738 torch.Size([8, 600, 800])
  7840. torch.Size([8, 8])
  7841. tensor([[0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083],
  7842. [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927],
  7843. [0.6236, 0.3967, 0.8675, 0.5400, 0.3862, 0.4517, 0.5825, 0.5200],
  7844. [0.6190, 0.4135, 0.8000, 0.4883, 0.3566, 0.3647, 0.5613, 0.5900],
  7845. [0.6273, 0.4143, 0.8750, 0.5700, 0.3987, 0.4717, 0.6013, 0.5467],
  7846. [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837],
  7847. [0.6278, 0.4253, 0.8875, 0.5017, 0.4113, 0.2750, 0.5413, 0.6196],
  7848. [0.6202, 0.4079, 0.8025, 0.2500, 0.3763, 0.3217, 0.6125, 0.5533]],
  7849. device='cuda:0', dtype=torch.float64)
  7850. predictions are: tensor([[ 0.3789, -0.5709, 1.7497, -0.0941, -0.4350, -0.0917, 0.5710, 0.1533],
  7851. [ 0.6201, -0.4236, 1.7590, -0.4153, -0.6013, -0.2876, 0.6609, 0.0435],
  7852. [ 0.5135, -0.4883, 1.8328, 0.1927, -0.4158, -0.2144, 0.5694, 0.0815],
  7853. [ 0.7109, -0.2940, 1.6941, -0.1075, -0.5754, -0.5636, 0.3950, 0.1171],
  7854. [ 0.5652, -0.4250, 1.7889, 0.1220, -0.4005, -0.2076, 0.5230, 0.0870],
  7855. [ 0.5411, -0.4651, 1.1210, -1.0702, -0.5344, -1.0046, 0.2457, 0.1836],
  7856. [ 0.5682, -0.4166, 1.6892, -0.0446, -0.2701, -0.6571, 0.4846, 0.1540],
  7857. [ 0.8239, -0.2596, 1.5885, -0.9988, -0.3349, -0.9175, 0.5501, 0.0685]],
  7858. device='cuda:0', grad_fn=<AddmmBackward>)
  7859. landmarks are: tensor([[[ 0.5713, -0.4105, 1.7557, 0.0467, -0.6520, -0.2690, 0.3988,
  7860. 0.5239],
  7861. [ 0.6010, -0.4345, 1.8480, -0.3537, -0.6462, -0.2613, 0.6524,
  7862. -0.0099],
  7863. [ 0.5944, -0.4538, 1.7210, 0.2083, -0.5018, -0.1997, 0.4046,
  7864. 0.1159],
  7865. [ 0.5732, -0.3761, 1.4092, -0.0303, -0.6388, -0.6012, 0.3065,
  7866. 0.4393],
  7867. [ 0.6115, -0.3724, 1.7557, 0.3469, -0.4441, -0.1073, 0.4912,
  7868. 0.2391],
  7869. [ 0.5629, -0.3871, 0.7243, -0.9581, -0.5827, -0.9849, 0.1288,
  7870. 0.4103],
  7871. [ 0.6139, -0.3216, 1.8134, 0.0313, -0.3864, -1.0157, 0.2144,
  7872. 0.5762],
  7873. [ 0.5788, -0.4020, 1.4208, -1.1312, -0.5480, -0.8002, 0.5432,
  7874. 0.2699]]], device='cuda:0')
  7875. loss_train_step before backward: tensor(0.0253, device='cuda:0', grad_fn=<MseLossBackward>)
  7876. loss_train_step after backward: tensor(0.0253, device='cuda:0', grad_fn=<MseLossBackward>)
  7877. loss_train: 6.368460373952985
  7878. step: 87
  7879. running loss: 0.07320069395348258
  7880. Train Steps: 87/90 Loss: 0.0732 torch.Size([8, 600, 800])
  7881. torch.Size([8, 8])
  7882. tensor([[0.6110, 0.3984, 0.8750, 0.4933, 0.4625, 0.4950, 0.5578, 0.5676],
  7883. [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683],
  7884. [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426],
  7885. [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726],
  7886. [0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5737, 0.5350],
  7887. [0.6200, 0.4098, 0.8237, 0.2917, 0.4012, 0.2967, 0.6000, 0.5683],
  7888. [0.6043, 0.4022, 0.6887, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136],
  7889. [0.6113, 0.4006, 0.8700, 0.5350, 0.3638, 0.3767, 0.5097, 0.4882]],
  7890. device='cuda:0', dtype=torch.float64)
  7891. predictions are: tensor([[ 0.3796, -0.5730, 1.8118, 0.0626, -0.1858, 0.1095, 0.5429, 0.1698],
  7892. [ 0.2873, -0.6146, 1.8132, 0.1036, -0.1824, 0.1075, 0.5311, 0.1494],
  7893. [ 0.6287, -0.3633, 1.7577, -0.1201, -0.5401, -0.5168, 0.4512, 0.1066],
  7894. [ 0.7294, -0.3134, 1.3254, -0.9326, -0.5984, -0.9560, 0.4578, 0.1106],
  7895. [ 0.6293, -0.3388, 1.7535, 0.0480, -0.4836, -0.6253, 0.4843, 0.1036],
  7896. [ 0.7504, -0.3159, 1.5468, -0.8603, -0.3524, -0.7077, 0.5296, 0.1618],
  7897. [ 0.6767, -0.3592, 1.1190, -1.0683, -0.5741, -1.1130, 0.3685, 0.1548],
  7898. [ 0.5433, -0.4318, 1.8166, 0.2886, -0.6087, -0.4237, 0.5091, 0.0742]],
  7899. device='cuda:0', grad_fn=<AddmmBackward>)
  7900. landmarks are: tensor([[[ 5.3637e-01, -4.4573e-01, 1.7557e+00, -7.2363e-03, -1.4965e-01,
  7901. 4.6189e-04, 2.9049e-01, 3.3573e-01],
  7902. [ 5.3585e-01, -4.3703e-01, 1.7095e+00, -3.0331e-02, -8.0370e-02,
  7903. -3.8029e-02, 1.0439e-01, 3.3918e-01],
  7904. [ 5.7875e-01, -4.1347e-01, 1.8214e+00, -2.4075e-01, -6.0389e-01,
  7905. -7.8543e-01, 4.1155e-01, 2.2033e-01],
  7906. [ 4.9971e-01, -4.4465e-01, 1.1610e+00, -9.7721e-01, -6.0577e-01,
  7907. -1.0311e+00, 1.4038e-01, -1.0312e-01],
  7908. [ 5.9007e-01, -4.0000e-01, 1.8423e+00, -6.8822e-02, -5.3072e-01,
  7909. -9.2333e-01, 3.6420e-01, 1.8522e-01],
  7910. [ 5.7783e-01, -3.9299e-01, 1.5189e+00, -9.3872e-01, -4.3256e-01,
  7911. -9.1563e-01, 4.8545e-01, 3.3918e-01],
  7912. [ 5.0531e-01, -4.2810e-01, 8.9538e-01, -1.3698e+00, -5.4226e-01,
  7913. -1.1389e+00, 2.4525e-01, 8.6245e-02],
  7914. [ 5.3788e-01, -4.3580e-01, 1.7326e+00, 1.8522e-01, -6.0577e-01,
  7915. -5.4611e-01, 6.8408e-02, -3.0981e-02]]], device='cuda:0')
  7916. loss_train_step before backward: tensor(0.0248, device='cuda:0', grad_fn=<MseLossBackward>)
  7917. loss_train_step after backward: tensor(0.0248, device='cuda:0', grad_fn=<MseLossBackward>)
  7918. loss_train: 6.393300803378224
  7919. step: 88
  7920. running loss: 0.07265114549293437
  7921. Train Steps: 88/90 Loss: 0.0727 torch.Size([8, 600, 800])
  7922. torch.Size([8, 8])
  7923. tensor([[0.6091, 0.3997, 0.8314, 0.4334, 0.3788, 0.4550, 0.5213, 0.5656],
  7924. [0.6254, 0.3993, 0.8988, 0.4767, 0.3987, 0.5517, 0.6955, 0.5285],
  7925. [ nan, nan, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621],
  7926. [ nan, nan, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533],
  7927. [0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585],
  7928. [0.6127, 0.4066, 0.8550, 0.5567, 0.4662, 0.5141, 0.5070, 0.5412],
  7929. [0.6293, 0.4024, 0.8750, 0.5000, 0.4012, 0.5733, 0.7121, 0.5633],
  7930. [0.6289, 0.4024, 0.9088, 0.4567, 0.3937, 0.5633, 0.7058, 0.5609]],
  7931. device='cuda:0', dtype=torch.float64)
  7932. predictions are: tensor([[ 0.6858, -0.3454, 1.5921, -0.3030, -0.6412, -0.4857, 0.3554, 0.1860],
  7933. [ 0.6335, -0.4234, 1.8358, 0.0389, -0.5274, 0.0525, 0.6177, 0.1727],
  7934. [ 0.4526, -0.4640, 1.5374, -1.1822, -0.0993, -1.4783, 0.6228, 0.0749],
  7935. [ 0.0231, -0.7600, 1.2551, -0.8862, -0.6108, -1.3019, 0.1301, 0.1329],
  7936. [ 0.7427, -0.3045, 1.7052, -0.2803, -0.4116, -0.1438, 0.4753, 0.1974],
  7937. [ 0.8011, -0.2826, 1.6989, 0.2847, -0.3568, -0.3909, 0.3260, 0.1932],
  7938. [ 0.8804, -0.2614, 1.8125, 0.1778, -0.4982, 0.1325, 0.6062, 0.2211],
  7939. [ 0.7056, -0.3785, 1.8138, -0.0180, -0.5265, 0.0883, 0.6572, 0.2042]],
  7940. device='cuda:0', grad_fn=<AddmmBackward>)
  7941. landmarks are: tensor([[[ 0.5275, -0.4396, 1.5543, -0.2841, -0.5365, -0.1843, 0.1221,
  7942. 0.3265],
  7943. [ 0.6026, -0.4417, 1.8654, -0.0842, -0.4441, 0.2622, 0.9265,
  7944. 0.1554],
  7945. [-2.2859, -2.2859, 1.7210, -0.9772, 0.1852, -1.3698, 0.7986,
  7946. 0.3104],
  7947. [-2.2859, -2.2859, 1.5074, -1.0388, -0.5423, -0.9849, 0.2199,
  7948. 0.2699],
  7949. [ 0.5115, -0.4332, 1.7557, -0.3152, -0.2536, 0.3931, 0.4139,
  7950. 0.2936],
  7951. [ 0.5443, -0.4079, 1.6633, 0.2853, -0.1325, 0.0888, 0.0558,
  7952. 0.2138],
  7953. [ 0.6207, -0.4273, 1.7557, 0.0236, -0.4326, 0.3623, 1.0033,
  7954. 0.3157],
  7955. [ 0.6191, -0.4273, 1.9115, -0.1766, -0.4672, 0.3161, 0.9741,
  7956. 0.3050]]], device='cuda:0')
  7957. loss_train_step before backward: tensor(0.3238, device='cuda:0', grad_fn=<MseLossBackward>)
  7958. loss_train_step after backward: tensor(0.3238, device='cuda:0', grad_fn=<MseLossBackward>)
  7959. loss_train: 6.717064084485173
  7960. step: 89
  7961. running loss: 0.07547263016275475
  7962. Train Steps: 89/90 Loss: 0.0755 torch.Size([8, 600, 800])
  7963. torch.Size([8, 8])
  7964. tensor([[0.6252, 0.4158, 0.8988, 0.4083, 0.3788, 0.4783, 0.6225, 0.5633],
  7965. [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717],
  7966. [0.6332, 0.4118, 0.9238, 0.4267, 0.4012, 0.4733, 0.7525, 0.5436],
  7967. [0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938],
  7968. [0.6132, 0.4037, 0.6963, 0.2217, 0.4100, 0.1950, 0.5395, 0.5175],
  7969. [0.6263, 0.4039, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804],
  7970. [ nan, nan, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729],
  7971. [0.6231, 0.3973, 0.8650, 0.3950, 0.3625, 0.3183, 0.5837, 0.5167]],
  7972. device='cuda:0', dtype=torch.float64)
  7973. predictions are: tensor([[ 0.6784, -0.3806, 1.9874, 0.0074, -0.5508, 0.0863, 0.5777, 0.2291],
  7974. [ 0.4791, -0.4740, 1.2944, -0.7627, -0.4662, -0.9881, 0.2634, 0.2488],
  7975. [ 0.7636, -0.3399, 1.8175, -0.1967, -0.4223, 0.0778, 0.7374, 0.2327],
  7976. [ 0.6685, -0.3923, 1.8977, 0.5432, -0.4429, -0.0133, 0.5661, 0.1591],
  7977. [ 0.6327, -0.3888, 1.1758, -0.9603, -0.4457, -1.3570, 0.2291, 0.1677],
  7978. [ 0.5583, -0.4858, 1.8936, 0.0071, -0.5720, 0.0042, 0.5830, 0.1419],
  7979. [-0.0687, -0.8334, 1.0642, -1.0043, -0.3935, -1.2683, 0.1451, 0.2017],
  7980. [ 0.8477, -0.2481, 1.7314, -0.2783, -0.5498, -0.6370, 0.5573, 0.1633]],
  7981. device='cuda:0', grad_fn=<AddmmBackward>)
  7982. landmarks are: tensor([[[ 0.6017, -0.3654, 1.8654, -0.3998, -0.5365, -0.0765, 0.5894,
  7983. 0.3161],
  7984. [ 0.5746, -0.3882, 1.1436, -1.2005, -0.4903, -1.0157, 0.4393,
  7985. 0.3546],
  7986. [ 0.6388, -0.3840, 1.9808, -0.3152, -0.4326, -0.0996, 1.1897,
  7987. 0.2249],
  7988. [ 0.6395, -0.4213, 1.7037, 0.3623, -0.4326, -0.1073, 0.6560,
  7989. -0.0049],
  7990. [ 0.5465, -0.4214, 0.9300, -1.2620, -0.3921, -1.3852, 0.2062,
  7991. 0.1043],
  7992. [ 0.6072, -0.4206, 1.8711, -0.2536, -0.6115, -0.1304, 0.6812,
  7993. -0.0670],
  7994. [-2.2859, -2.2859, 0.8824, -1.2727, -0.3691, -1.4237, 0.0943,
  7995. 0.3604],
  7996. [ 0.5924, -0.4507, 1.7095, -0.4614, -0.6115, -0.8156, 0.4104,
  7997. 0.1005]]], device='cuda:0')
  7998. loss_train_step before backward: tensor(0.1345, device='cuda:0', grad_fn=<MseLossBackward>)
  7999. loss_train_step after backward: tensor(0.1345, device='cuda:0', grad_fn=<MseLossBackward>)
  8000. loss_train: 6.85155183263123
  8001. step: 90
  8002. running loss: 0.07612835369590255
  8003.  
  8004. Valid Steps: 10/10 Loss: nan 61
  8005. --------------------------------------------------
  8006. Epoch: 2 Train Loss: 0.0761 Valid Loss: nan
  8007. --------------------------------------------------
  8008. size of train loader is: 90
  8009. torch.Size([8, 600, 800])
  8010. torch.Size([8, 8])
  8011. tensor([[0.6260, 0.4153, 0.9000, 0.4533, 0.4025, 0.2633, 0.6223, 0.4967],
  8012. [0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945],
  8013. [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383],
  8014. [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5837, 0.5500],
  8015. [0.6300, 0.4133, 0.8538, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413],
  8016. [0.6205, 0.4016, 0.8350, 0.2717, 0.3987, 0.2550, 0.5787, 0.5133],
  8017. [0.6166, 0.4008, 0.8563, 0.5667, 0.4388, 0.4933, 0.5575, 0.5567],
  8018. [0.6185, 0.4067, 0.8838, 0.4450, 0.4037, 0.4733, 0.5213, 0.5142]],
  8019. device='cuda:0', dtype=torch.float64)
  8020. predictions are: tensor([[ 0.4644, -0.4431, 1.6816, -0.0788, -0.5316, -0.8118, 0.3678, 0.1957],
  8021. [ 0.3627, -0.5644, 1.7695, -0.1866, -0.3259, -0.4964, 0.6251, 0.2581],
  8022. [ 0.7497, -0.3301, 1.4605, -0.7863, -0.7153, -0.5758, 0.6303, 0.2166],
  8023. [ 0.5202, -0.4933, 1.6469, -0.0964, -0.5065, 0.0323, 0.4776, 0.2630],
  8024. [ 0.3867, -0.5193, 1.5639, -0.8843, -0.0533, -0.8858, 0.7233, 0.2240],
  8025. [ 0.3793, -0.5255, 1.4437, -0.7890, -0.4778, -0.8875, 0.3171, 0.1730],
  8026. [ 0.6539, -0.3837, 1.6159, 0.2937, -0.4373, -0.0220, 0.4100, 0.2954],
  8027. [ 0.3520, -0.5612, 1.6328, -0.0203, -0.6282, -0.1089, 0.2260, 0.2060]],
  8028. device='cuda:0', grad_fn=<AddmmBackward>)
  8029. landmarks are: tensor([[[ 0.6055, -0.3676, 1.8711, -0.1920, -0.4268, -1.0696, 0.5887,
  8030. 0.0081],
  8031. [ 0.6471, -0.3719, 1.9866, -0.3921, -0.3806, -0.5538, 1.0070,
  8032. 0.4600],
  8033. [ 0.5779, -0.4275, 1.5894, -0.8362, -0.6577, -0.5153, 0.5605,
  8034. 0.2006],
  8035. [ 0.5604, -0.4620, 1.7961, -0.1997, -0.3229, 0.1082, 0.4104,
  8036. 0.2545],
  8037. [ 0.6240, -0.3768, 1.6575, -1.2851, 0.2949, -1.2467, 1.1276,
  8038. 0.2142],
  8039. [ 0.5800, -0.4312, 1.5709, -1.0311, -0.4441, -1.1081, 0.3873,
  8040. 0.0851],
  8041. [ 0.5620, -0.4346, 1.6691, 0.3315, -0.2594, -0.0072, 0.2891,
  8042. 0.2853],
  8043. [ 0.5708, -0.4075, 1.7961, -0.2305, -0.4210, -0.0996, 0.1219,
  8044. 0.0893]]], device='cuda:0')
  8045. loss_train_step before backward: tensor(0.0298, device='cuda:0', grad_fn=<MseLossBackward>)
  8046. loss_train_step after backward: tensor(0.0298, device='cuda:0', grad_fn=<MseLossBackward>)
  8047. loss_train: 0.02976425550878048
  8048. step: 1
  8049. running loss: 0.02976425550878048
  8050. Train Steps: 1/90 Loss: 0.0298 torch.Size([8, 600, 800])
  8051. torch.Size([8, 8])
  8052. tensor([[0.6205, 0.4016, 0.8350, 0.2717, 0.3987, 0.2550, 0.5787, 0.5133],
  8053. [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317],
  8054. [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413],
  8055. [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083],
  8056. [0.6350, 0.4043, 0.8738, 0.5650, 0.3850, 0.4750, 0.6401, 0.4950],
  8057. [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283],
  8058. [0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266],
  8059. [0.6224, 0.4179, 0.8700, 0.5683, 0.4037, 0.4683, 0.5650, 0.5600]],
  8060. device='cuda:0', dtype=torch.float64)
  8061. predictions are: tensor([[ 0.2171, -0.6386, 1.4917, -0.9178, -0.3628, -0.8859, 0.3591, 0.1758],
  8062. [ 0.4784, -0.4798, 1.6804, 0.0476, -0.5666, -0.5235, 0.3980, 0.2267],
  8063. [ 0.2964, -0.6088, 1.7696, -0.4092, -0.2328, -0.7607, 0.5548, 0.2144],
  8064. [ 0.5832, -0.4517, 1.6893, -0.2107, -0.5384, -0.1259, 0.5803, 0.2860],
  8065. [ 0.6926, -0.3706, 1.8076, 0.1113, -0.5554, -0.0664, 0.6463, 0.2072],
  8066. [ 0.3910, -0.5487, 1.6927, -0.3536, -0.6326, -0.4973, 0.4058, 0.2500],
  8067. [ 0.3494, -0.5530, 1.1658, -0.9490, -0.4045, -0.9441, 0.3029, 0.3491],
  8068. [ 0.7162, -0.3370, 1.7149, 0.1530, -0.5052, -0.0180, 0.5345, 0.3138]],
  8069. device='cuda:0', grad_fn=<AddmmBackward>)
  8070. landmarks are: tensor([[[ 5.7997e-01, -4.3118e-01, 1.5709e+00, -1.0311e+00, -4.4411e-01,
  8071. -1.1081e+00, 3.8730e-01, 8.5142e-02],
  8072. [ 5.9013e-01, -4.1570e-01, 1.7557e+00, 1.9292e-01, -5.4226e-01,
  8073. -5.9230e-01, 3.5843e-01, 1.6982e-01],
  8074. [ 6.2236e-01, -4.3453e-01, 1.9404e+00, -2.9207e-01, -3.1709e-01,
  8075. -8.7714e-01, 1.0655e+00, 2.1421e-01],
  8076. [ 5.7131e-01, -4.1045e-01, 1.7557e+00, 4.6651e-02, -6.5196e-01,
  8077. -2.6898e-01, 3.9885e-01, 5.2394e-01],
  8078. [ 6.4706e-01, -4.1832e-01, 1.7499e+00, 3.2379e-01, -5.0762e-01,
  8079. -9.1917e-02, 6.7064e-01, 4.6189e-04],
  8080. [ 5.8799e-01, -3.8868e-01, 1.8423e+00, -3.3056e-01, -6.2309e-01,
  8081. -5.2302e-01, 4.0462e-01, 1.5443e-01],
  8082. [ 5.7460e-01, -4.0208e-01, 1.0801e+00, -1.1312e+00, -3.2286e-01,
  8083. -1.1081e+00, 4.8034e-01, 6.0842e-01],
  8084. [ 5.8909e-01, -3.5574e-01, 1.7326e+00, 3.3918e-01, -4.2102e-01,
  8085. -1.2271e-01, 3.2379e-01, 3.0069e-01]]], device='cuda:0')
  8086. loss_train_step before backward: tensor(0.0238, device='cuda:0', grad_fn=<MseLossBackward>)
  8087. loss_train_step after backward: tensor(0.0238, device='cuda:0', grad_fn=<MseLossBackward>)
  8088. loss_train: 0.053598225116729736
  8089. step: 2
  8090. running loss: 0.026799112558364868
  8091. Train Steps: 2/90 Loss: 0.0268 torch.Size([8, 600, 800])
  8092. torch.Size([8, 8])
  8093. tensor([[0.6218, 0.4098, 0.7238, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350],
  8094. [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220],
  8095. [0.6222, 0.3937, 0.8350, 0.5617, 0.4138, 0.4600, 0.5800, 0.5233],
  8096. [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329],
  8097. [0.6275, 0.4081, 0.8063, 0.2017, 0.4825, 0.1583, 0.6156, 0.4869],
  8098. [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650],
  8099. [0.6252, 0.4158, 0.8988, 0.4083, 0.3788, 0.4783, 0.6225, 0.5633],
  8100. [0.6289, 0.4019, 0.8113, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332]],
  8101. device='cuda:0', dtype=torch.float64)
  8102. predictions are: tensor([[ 0.4246, -0.5008, 1.2521, -0.9797, -0.3414, -1.0541, 0.5136, 0.3221],
  8103. [-0.0271, -0.7760, 1.0477, -0.9932, -0.3399, -1.3101, 0.2273, 0.2716],
  8104. [ 0.7331, -0.3530, 1.8778, 0.4137, -0.4636, 0.0050, 0.6195, 0.2694],
  8105. [ 0.5305, -0.4416, 1.7704, -0.2950, -0.6993, -0.4477, 0.4500, 0.2283],
  8106. [ 0.0741, -0.7198, 1.4817, -0.8829, -0.1518, -1.0529, 0.5162, 0.2382],
  8107. [-0.0500, -0.7776, 1.3050, -0.7022, -0.4658, -0.7708, 0.1221, 0.3052],
  8108. [ 0.6535, -0.3872, 2.0202, -0.0684, -0.6126, 0.1059, 0.6727, 0.2721],
  8109. [ 0.8910, -0.2266, 1.7665, 0.1437, -0.5468, 0.1158, 0.6713, 0.2996]],
  8110. device='cuda:0', grad_fn=<AddmmBackward>)
  8111. landmarks are: tensor([[[ 0.5861, -0.3929, 1.0570, -1.4314, -0.3286, -1.1235, 0.6182,
  8112. 0.1852],
  8113. [ 0.5566, -0.4160, 0.9935, -1.3313, -0.2824, -1.5161, 0.2144,
  8114. 0.1253],
  8115. [ 0.5880, -0.4676, 1.5709, 0.3084, -0.3748, -0.1612, 0.3931,
  8116. 0.1313],
  8117. [ 0.5680, -0.4393, 1.5920, -0.6672, -0.6453, -0.5457, 0.5149,
  8118. 0.1753],
  8119. [ 0.6124, -0.4010, 1.4381, -1.3544, -0.0573, -1.5546, 0.5573,
  8120. -0.0369],
  8121. [ 0.5320, -0.4189, 1.3053, -1.0773, -0.5711, -0.9849, 0.2267,
  8122. 0.3237],
  8123. [ 0.6017, -0.3654, 1.8654, -0.3998, -0.5365, -0.0765, 0.5894,
  8124. 0.3161],
  8125. [ 0.6191, -0.4297, 1.4612, 0.2391, -0.4961, 0.0313, 1.1166,
  8126. 0.1768]]], device='cuda:0')
  8127. loss_train_step before backward: tensor(0.0572, device='cuda:0', grad_fn=<MseLossBackward>)
  8128. loss_train_step after backward: tensor(0.0572, device='cuda:0', grad_fn=<MseLossBackward>)
  8129. loss_train: 0.11081135272979736
  8130. step: 3
  8131. running loss: 0.03693711757659912
  8132. Train Steps: 3/90 Loss: 0.0369 torch.Size([8, 600, 800])
  8133. torch.Size([8, 8])
  8134. tensor([[0.6219, 0.4097, 0.8738, 0.3400, 0.3563, 0.4117, 0.5975, 0.5683],
  8135. [0.6177, 0.4086, 0.8738, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700],
  8136. [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5988, 0.5283],
  8137. [0.6286, 0.4060, 0.9188, 0.4333, 0.3675, 0.4167, 0.7034, 0.5528],
  8138. [0.6109, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117],
  8139. [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209],
  8140. [ nan, nan, 0.7268, 0.2333, 0.4125, 0.1933, 0.5112, 0.5383],
  8141. [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795]],
  8142. device='cuda:0', dtype=torch.float64)
  8143. predictions are: tensor([[ 0.3962, -0.5233, 1.6373, -0.6012, -0.6109, -0.6016, 0.4131, 0.2866],
  8144. [ 0.8978, -0.2230, 1.7190, -0.6673, -0.4464, 0.1228, 0.8288, 0.3088],
  8145. [ 0.8697, -0.2394, 1.7297, 0.0657, -0.4729, -0.1425, 0.5340, 0.2836],
  8146. [ 0.4009, -0.5431, 1.8679, -0.2984, -0.4186, -0.5946, 0.6249, 0.2787],
  8147. [ 0.3967, -0.5239, 1.7884, -0.1002, -0.5768, -0.4514, 0.3042, 0.2203],
  8148. [ 0.2635, -0.5390, 1.5640, -0.2989, -0.2955, -0.9008, 0.3495, 0.3293],
  8149. [-0.8458, -1.3516, 0.9245, -1.1352, -0.3281, -1.4677, 0.0543, 0.2653],
  8150. [ 0.8833, -0.2316, 1.8017, -0.0581, -0.4491, -0.0319, 0.8155, 0.3043]],
  8151. device='cuda:0', grad_fn=<AddmmBackward>)
  8152. landmarks are: tensor([[[ 0.5867, -0.3937, 1.7499, -0.7155, -0.6404, -0.3844, 0.4739,
  8153. 0.3392],
  8154. [ 0.5671, -0.3988, 1.7499, -0.4614, -0.5423, 0.3007, 0.5894,
  8155. 0.3469],
  8156. [ 0.5603, -0.4319, 1.7788, 0.1775, -0.5538, -0.0611, 0.4797,
  8157. 0.1544],
  8158. [ 0.6174, -0.4105, 1.9577, -0.2844, -0.5885, -0.3614, 0.9631,
  8159. 0.2676],
  8160. [ 0.5356, -0.4343, 1.7441, -0.1150, -0.6462, -0.3614, 0.3238,
  8161. 0.0774],
  8162. [ 0.6174, -0.3118, 1.6402, -0.2074, -0.1958, -1.0927, 0.2267,
  8163. 0.5822],
  8164. [-2.2859, -2.2859, 1.0712, -1.2085, -0.3806, -1.3929, 0.0755,
  8165. 0.2006],
  8166. [ 0.6355, -0.4080, 1.6113, 0.1852, -0.4730, 0.1467, 0.9996,
  8167. 0.3905]]], device='cuda:0')
  8168. loss_train_step before backward: tensor(0.0703, device='cuda:0', grad_fn=<MseLossBackward>)
  8169. loss_train_step after backward: tensor(0.0703, device='cuda:0', grad_fn=<MseLossBackward>)
  8170. loss_train: 0.18106389790773392
  8171. step: 4
  8172. running loss: 0.04526597447693348
  8173.  
  8174. Train Steps: 4/90 Loss: 0.0453 torch.Size([8, 600, 800])
  8175. torch.Size([8, 8])
  8176. tensor([[0.6229, 0.4066, 0.7612, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350],
  8177. [ nan, nan, 0.6900, 0.1917, 0.3937, 0.2367, 0.5240, 0.5246],
  8178. [0.6201, 0.4017, 0.8871, 0.4621, 0.3517, 0.4675, 0.5999, 0.5106],
  8179. [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6088, 0.5583],
  8180. [0.6286, 0.4086, 0.8408, 0.2801, 0.4163, 0.2800, 0.6725, 0.5393],
  8181. [0.6275, 0.4048, 0.8488, 0.2883, 0.4463, 0.2033, 0.6321, 0.5155],
  8182. [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333],
  8183. [0.6250, 0.4236, 0.8638, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799]],
  8184. device='cuda:0', dtype=torch.float64)
  8185. predictions are: tensor([[ 0.4974, -0.4397, 1.3189, -0.9185, -0.3947, -0.9882, 0.4938, 0.2704],
  8186. [-1.2384, -1.5859, 1.0754, -0.9062, -0.3067, -1.0850, 0.0830, 0.2715],
  8187. [ 0.8753, -0.2470, 1.9504, 0.1446, -0.6124, 0.1460, 0.6680, 0.2374],
  8188. [ 0.4194, -0.4939, 1.1516, -0.8221, -0.5055, -0.9677, 0.3401, 0.3332],
  8189. [ 0.2939, -0.5620, 1.6364, -0.6533, -0.2373, -0.7871, 0.6217, 0.2830],
  8190. [ 0.6833, -0.3461, 1.5232, -0.9011, -0.2480, -1.0147, 0.7707, 0.2400],
  8191. [ 0.9637, -0.1672, 2.0570, 0.1913, -0.6407, 0.2832, 0.6602, 0.2757],
  8192. [ 0.3353, -0.5188, 1.7350, -0.2497, -0.4663, -0.4829, 0.4751, 0.3412]],
  8193. device='cuda:0', grad_fn=<AddmmBackward>)
  8194. landmarks are: tensor([[[ 0.5911, -0.4080, 1.2303, -0.9156, -0.3229, -1.2851, 0.4508,
  8195. 0.1852],
  8196. [-2.2859, -2.2859, 0.9012, -1.4006, -0.4672, -1.1928, 0.1342,
  8197. 0.1373],
  8198. [ 0.5783, -0.4306, 1.8114, -0.1515, -0.6617, -0.1268, 0.4851,
  8199. 0.0727],
  8200. [ 0.5992, -0.3968, 0.9307, -1.3497, -0.4730, -1.0465, 0.5259,
  8201. 0.2930],
  8202. [ 0.6174, -0.3984, 1.5975, -0.9921, -0.3633, -0.9926, 0.8205,
  8203. 0.2050],
  8204. [ 0.6126, -0.4161, 1.6344, -0.9541, -0.2247, -1.3467, 0.6339,
  8205. 0.0953],
  8206. [ 0.6003, -0.3747, 1.8885, -0.1997, -0.5018, -0.0149, 0.5490,
  8207. 0.1775],
  8208. [ 0.6008, -0.3293, 1.7037, -0.5461, -0.4152, -0.8309, 0.3234,
  8209. 0.3928]]], device='cuda:0')
  8210. loss_train_step before backward: tensor(0.0653, device='cuda:0', grad_fn=<MseLossBackward>)
  8211. loss_train_step after backward: tensor(0.0653, device='cuda:0', grad_fn=<MseLossBackward>)
  8212. loss_train: 0.24632898718118668
  8213. step: 5
  8214. running loss: 0.04926579743623734
  8215. Train Steps: 5/90 Loss: 0.0493 torch.Size([8, 600, 800])
  8216. torch.Size([8, 8])
  8217. tensor([[0.6230, 0.4113, 0.7213, 0.1983, 0.4325, 0.2367, 0.6262, 0.5400],
  8218. [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167],
  8219. [0.6193, 0.4050, 0.7313, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656],
  8220. [0.6166, 0.4008, 0.8563, 0.5667, 0.4388, 0.4933, 0.5575, 0.5567],
  8221. [0.6132, 0.4118, 0.8200, 0.3633, 0.3563, 0.5400, 0.5787, 0.5136],
  8222. [0.6293, 0.4097, 0.8800, 0.2517, 0.5262, 0.2600, 0.7430, 0.5378],
  8223. [0.6165, 0.4106, 0.7575, 0.1733, 0.3838, 0.2650, 0.5680, 0.5116],
  8224. [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967]],
  8225. device='cuda:0', dtype=torch.float64)
  8226. predictions are: tensor([[-0.0326, -0.7844, 1.3051, -0.8212, -0.3678, -1.0649, 0.3889, 0.3106],
  8227. [ 0.4314, -0.4934, 1.6887, -0.4932, -0.6464, -0.7139, 0.3917, 0.2290],
  8228. [ 0.2542, -0.5932, 1.1377, -0.9746, -0.4692, -1.0381, 0.3510, 0.3077],
  8229. [ 0.7676, -0.3140, 1.8666, 0.5899, -0.4083, 0.2689, 0.5522, 0.3187],
  8230. [ 0.6571, -0.3569, 1.7394, -0.4057, -0.6427, 0.3132, 0.6880, 0.2942],
  8231. [ 0.4241, -0.5237, 1.5921, -1.0009, -0.0236, -1.0447, 0.9168, 0.2845],
  8232. [ 0.3779, -0.5320, 1.3481, -0.8482, -0.5710, -1.0561, 0.4776, 0.2223],
  8233. [-0.1864, -0.8626, 1.5975, -0.7790, -0.1724, -0.9359, 0.5434, 0.2635]],
  8234. device='cuda:0', grad_fn=<AddmmBackward>)
  8235. landmarks are: tensor([[[ 5.9169e-01, -3.8607e-01, 1.0455e+00, -1.3698e+00, -2.8822e-01,
  8236. -1.1928e+00, 6.0670e-01, 2.0831e-01],
  8237. [ 5.4825e-01, -4.2490e-01, 1.5305e+00, -7.3857e-01, -6.1155e-01,
  8238. -8.6944e-01, 3.3533e-01, 1.0054e-01],
  8239. [ 5.7460e-01, -4.1527e-01, 1.0917e+00, -1.1620e+00, -4.0370e-01,
  8240. -1.3082e+00, 3.2339e-01, 3.2671e-01],
  8241. [ 5.6195e-01, -4.3457e-01, 1.6691e+00, 3.3149e-01, -2.5935e-01,
  8242. -7.2363e-03, 2.8915e-01, 2.8530e-01],
  8243. [ 5.4660e-01, -3.8397e-01, 1.5016e+00, -6.0770e-01, -6.4042e-01,
  8244. 2.0831e-01, 3.8714e-01, 8.6245e-02],
  8245. [ 6.2072e-01, -3.9360e-01, 1.7788e+00, -1.1235e+00, 1.4480e-01,
  8246. -1.0850e+00, 1.1459e+00, 1.9818e-01],
  8247. [ 5.6184e-01, -3.8945e-01, 1.2129e+00, -1.4853e+00, -5.1339e-01,
  8248. -1.0619e+00, 3.3778e-01, 7.7228e-02],
  8249. [ 6.0479e-01, -3.7229e-01, 1.6517e+00, -1.0773e+00, 4.6189e-04,
  8250. -1.5161e+00, 5.8660e-01, 8.0947e-03]]], device='cuda:0')
  8251. loss_train_step before backward: tensor(0.0639, device='cuda:0', grad_fn=<MseLossBackward>)
  8252. loss_train_step after backward: tensor(0.0639, device='cuda:0', grad_fn=<MseLossBackward>)
  8253. loss_train: 0.3102681040763855
  8254. step: 6
  8255. running loss: 0.05171135067939758
  8256. Train Steps: 6/90 Loss: 0.0517 torch.Size([8, 600, 800])
  8257. torch.Size([8, 8])
  8258. tensor([[0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012],
  8259. [ nan, nan, 0.6469, 0.1943, 0.4025, 0.2000, 0.5125, 0.5533],
  8260. [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892],
  8261. [0.6147, 0.4112, 0.7988, 0.3200, 0.3775, 0.2767, 0.5150, 0.5550],
  8262. [0.6205, 0.4062, 0.8337, 0.2683, 0.3675, 0.4283, 0.6338, 0.5250],
  8263. [0.6189, 0.4033, 0.8650, 0.5267, 0.4487, 0.5150, 0.5925, 0.5050],
  8264. [0.6200, 0.4098, 0.8237, 0.2917, 0.4012, 0.2967, 0.6000, 0.5683],
  8265. [0.6127, 0.4118, 0.8650, 0.5083, 0.4088, 0.5367, 0.5300, 0.5456]],
  8266. device='cuda:0', dtype=torch.float64)
  8267. predictions are: tensor([[ 1.0117, -0.1618, 1.9431, 0.1343, -0.5366, -0.3326, 0.7663, 0.1990],
  8268. [-1.0768, -1.5103, 0.9567, -1.3619, -0.3016, -1.6007, 0.2505, 0.2646],
  8269. [ 0.7019, -0.3086, 1.7838, -0.1395, -0.7029, -0.5296, 0.4112, 0.2605],
  8270. [-0.0792, -0.8122, 1.4107, -0.9500, -0.5097, -1.0802, 0.2752, 0.2552],
  8271. [ 0.5814, -0.4038, 1.4747, -1.1459, -0.5425, -0.5329, 0.7750, 0.2706],
  8272. [ 0.6678, -0.3636, 1.8524, -0.0345, -0.2189, -0.0571, 0.5808, 0.2862],
  8273. [ 0.3820, -0.5253, 1.6017, -1.1825, -0.3065, -1.0830, 0.6300, 0.2799],
  8274. [ 0.6951, -0.3400, 1.8050, -0.0378, -0.3050, 0.0485, 0.5817, 0.3112]],
  8275. device='cuda:0', grad_fn=<AddmmBackward>)
  8276. landmarks are: tensor([[[ 0.6249, -0.4352, 1.8018, 0.2545, -0.6173, -0.1997, 0.6401,
  8277. 0.0291],
  8278. [-2.2859, -2.2859, 0.7023, -1.3883, -0.4268, -1.3621, 0.0813,
  8279. 0.2699],
  8280. [ 0.5454, -0.4053, 1.6633, -0.1766, -0.6058, -0.5923, 0.1577,
  8281. 0.4357],
  8282. [ 0.5532, -0.3864, 1.4035, -0.8079, -0.5423, -1.0080, 0.0928,
  8283. 0.2776],
  8284. [ 0.5799, -0.4099, 1.5651, -1.0465, -0.5885, -0.3075, 0.6413,
  8285. 0.1390],
  8286. [ 0.5726, -0.4230, 1.7095, 0.1467, -0.2132, 0.0928, 0.4508,
  8287. 0.0467],
  8288. [ 0.5778, -0.3930, 1.5189, -0.9387, -0.4326, -0.9156, 0.4855,
  8289. 0.3392],
  8290. [ 0.5443, -0.3840, 1.7095, 0.0620, -0.3979, 0.1929, 0.1622,
  8291. 0.2341]]], device='cuda:0')
  8292. loss_train_step before backward: tensor(0.0636, device='cuda:0', grad_fn=<MseLossBackward>)
  8293. loss_train_step after backward: tensor(0.0636, device='cuda:0', grad_fn=<MseLossBackward>)
  8294. loss_train: 0.37389400601387024
  8295. step: 7
  8296. running loss: 0.05341342943055289
  8297. Train Steps: 7/90 Loss: 0.0534 torch.Size([8, 600, 800])
  8298. torch.Size([8, 8])
  8299. tensor([[0.6136, 0.4085, 0.6688, 0.2317, 0.3862, 0.2367, 0.5517, 0.5783],
  8300. [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389],
  8301. [0.6212, 0.4159, 0.8675, 0.5783, 0.4088, 0.4317, 0.5613, 0.5917],
  8302. [0.6161, 0.4024, 0.8838, 0.4583, 0.3688, 0.3733, 0.5311, 0.5344],
  8303. [0.6200, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913],
  8304. [0.6179, 0.4118, 0.7278, 0.4237, 0.3588, 0.3400, 0.5675, 0.5917],
  8305. [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5687, 0.5567],
  8306. [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413]],
  8307. device='cuda:0', dtype=torch.float64)
  8308. predictions are: tensor([[ 0.0096, -0.7602, 0.9648, -1.2757, -0.4615, -1.1769, 0.3001, 0.3017],
  8309. [ 0.2265, -0.6059, 1.3263, -1.2305, -0.4143, -1.1986, 0.5272, 0.2549],
  8310. [ 0.4254, -0.4732, 1.6860, 0.2153, -0.4557, -0.1081, 0.4277, 0.3285],
  8311. [ 0.9220, -0.1908, 1.9226, -0.0915, -0.6569, -0.3724, 0.5336, 0.1467],
  8312. [ 0.7344, -0.3511, 1.9044, 0.0650, -0.4269, 0.1083, 0.6133, 0.1700],
  8313. [ 0.4440, -0.4838, 1.4899, -0.6572, -0.6243, -0.7794, 0.3746, 0.2894],
  8314. [-0.5248, -1.0886, 1.2168, -1.2470, -0.2981, -1.0941, 0.3924, 0.3101],
  8315. [ 0.3848, -0.5339, 1.8826, -1.0373, 0.0069, -1.0555, 1.0492, 0.1640]],
  8316. device='cuda:0', grad_fn=<AddmmBackward>)
  8317. landmarks are: tensor([[[ 0.5483, -0.3991, 0.8030, -1.2159, -0.5018, -1.1928, 0.2624,
  8318. 0.3852],
  8319. [ 0.5697, -0.4466, 1.1973, -1.1871, -0.4571, -0.9965, 0.5219,
  8320. 0.2032],
  8321. [ 0.5836, -0.3649, 1.7210, 0.3854, -0.3979, -0.2921, 0.3065,
  8322. 0.4470],
  8323. [ 0.5598, -0.4273, 1.7961, -0.1689, -0.5827, -0.5615, 0.1671,
  8324. 0.1824],
  8325. [ 0.5776, -0.4784, 1.7730, 0.1236, -0.4037, 0.0697, 0.5132,
  8326. -0.0168],
  8327. [ 0.5680, -0.3840, 1.0756, -0.3290, -0.6289, -0.7155, 0.3353,
  8328. 0.4470],
  8329. [ 0.5824, -0.3625, 1.0405, -0.9325, -0.4268, -1.1389, 0.3411,
  8330. 0.2853],
  8331. [ 0.6108, -0.4201, 1.8711, -0.7848, -0.0053, -1.2236, 1.0362,
  8332. 0.2142]]], device='cuda:0')
  8333. loss_train_step before backward: tensor(0.0577, device='cuda:0', grad_fn=<MseLossBackward>)
  8334. loss_train_step after backward: tensor(0.0577, device='cuda:0', grad_fn=<MseLossBackward>)
  8335. loss_train: 0.43161700665950775
  8336. step: 8
  8337. running loss: 0.05395212583243847
  8338.  
  8339. Train Steps: 8/90 Loss: 0.0540 torch.Size([8, 600, 800])
  8340. torch.Size([8, 8])
  8341. tensor([[0.6284, 0.4127, 0.8538, 0.5867, 0.4363, 0.5083, 0.6038, 0.5433],
  8342. [0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412],
  8343. [0.6079, 0.3964, 0.7420, 0.2958, 0.3563, 0.2917, 0.5351, 0.4980],
  8344. [0.6245, 0.4100, 0.7762, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417],
  8345. [0.6162, 0.4134, 0.6700, 0.2467, 0.3962, 0.2533, 0.5737, 0.5467],
  8346. [0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118],
  8347. [0.6271, 0.4040, 0.9138, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413],
  8348. [0.6109, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567]],
  8349. device='cuda:0', dtype=torch.float64)
  8350. predictions are: tensor([[ 0.4623, -0.4913, 1.8371, 0.3857, -0.4122, 0.2493, 0.5271, 0.1908],
  8351. [ 0.6949, -0.3453, 1.8927, -0.5485, -0.4945, -0.7875, 0.6493, 0.1366],
  8352. [ 0.2863, -0.5883, 1.2984, -0.9925, -0.5895, -0.7914, 0.3332, 0.1886],
  8353. [-0.0663, -0.7860, 1.2408, -1.1712, -0.1767, -1.2552, 0.4436, 0.2503],
  8354. [ 0.0550, -0.7215, 1.0977, -1.0060, -0.5058, -1.0154, 0.3654, 0.2937],
  8355. [ 0.2194, -0.6018, 1.1889, -0.9221, -0.3864, -0.8253, 0.3989, 0.3544],
  8356. [ 0.7135, -0.3359, 2.0395, -0.5939, -0.1271, -0.8327, 0.9730, 0.1247],
  8357. [ 0.1684, -0.6724, 1.2012, -1.1064, -0.4767, -0.9897, 0.3972, 0.2452]],
  8358. device='cuda:0', grad_fn=<AddmmBackward>)
  8359. landmarks are: tensor([[[ 0.6166, -0.3795, 1.6575, 0.4239, -0.2709, 0.0620, 0.5028,
  8360. 0.2237],
  8361. [ 0.5784, -0.4006, 1.7911, -0.5701, -0.5192, -1.0331, 0.4137,
  8362. 0.2139],
  8363. [ 0.5219, -0.4550, 1.1415, -0.9196, -0.6404, -0.9387, 0.1856,
  8364. 0.0141],
  8365. [ 0.5985, -0.3921, 1.2995, -1.0927, 0.0062, -1.5854, 0.4277,
  8366. 0.2160],
  8367. [ 0.5603, -0.3764, 0.8088, -1.1466, -0.4557, -1.1158, 0.3642,
  8368. 0.2391],
  8369. [ 0.6077, -0.3226, 0.9993, -0.6462, -0.2651, -1.3082, 0.2946,
  8370. 0.5401],
  8371. [ 0.6108, -0.4201, 1.9346, -0.5538, -0.1497, -1.0773, 1.0545,
  8372. 0.2142],
  8373. [ 0.5356, -0.4217, 1.0339, -1.4776, -0.5076, -1.1081, 0.4219,
  8374. 0.2853]]], device='cuda:0')
  8375. loss_train_step before backward: tensor(0.0441, device='cuda:0', grad_fn=<MseLossBackward>)
  8376. loss_train_step after backward: tensor(0.0441, device='cuda:0', grad_fn=<MseLossBackward>)
  8377. loss_train: 0.4756833612918854
  8378. step: 9
  8379. running loss: 0.05285370681020948
  8380. Train Steps: 9/90 Loss: 0.0529 torch.Size([8, 600, 800])
  8381. torch.Size([8, 8])
  8382. tensor([[0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433],
  8383. [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167],
  8384. [0.6273, 0.4143, 0.8750, 0.5700, 0.3987, 0.4717, 0.6013, 0.5467],
  8385. [0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162],
  8386. [0.6275, 0.4050, 0.9038, 0.3767, 0.3838, 0.3533, 0.7074, 0.5575],
  8387. [0.6193, 0.4108, 0.7438, 0.2700, 0.3650, 0.3683, 0.6238, 0.5717],
  8388. [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6088, 0.5583],
  8389. [0.6346, 0.4086, 0.7938, 0.5500, 0.3962, 0.4867, 0.7343, 0.5702]],
  8390. device='cuda:0', dtype=torch.float64)
  8391. predictions are: tensor([[ 0.2563, -0.6047, 1.2636, -1.1187, -0.5007, -0.8370, 0.3980, 0.2478],
  8392. [ 0.4033, -0.5214, 1.6198, -0.9454, -0.4773, -0.9102, 0.5505, 0.2541],
  8393. [ 0.1984, -0.6554, 1.7125, 0.0773, -0.3777, -0.4354, 0.3894, 0.1948],
  8394. [ 0.6152, -0.3854, 1.7423, -0.4495, -0.5086, -0.2344, 0.5880, 0.1688],
  8395. [ 0.7010, -0.3407, 1.9443, -0.7245, -0.2994, -1.1247, 0.8264, 0.1196],
  8396. [ 0.2452, -0.5995, 1.1966, -1.1032, -0.5102, -0.7389, 0.3625, 0.2906],
  8397. [ 0.0631, -0.7514, 1.0383, -1.2927, -0.3640, -1.4295, 0.3175, 0.2634],
  8398. [ 0.5745, -0.3833, 1.5549, -0.1126, -0.4069, -0.2537, 0.5014, 0.2783]],
  8399. device='cuda:0', grad_fn=<AddmmBackward>)
  8400. landmarks are: tensor([[[ 0.5665, -0.3963, 1.3457, -0.9926, -0.6058, -0.6308, 0.5721,
  8401. 0.2237],
  8402. [ 0.5768, -0.3899, 1.3861, -0.7771, -0.5885, -0.5461, 0.5028,
  8403. 0.5624],
  8404. [ 0.6115, -0.3724, 1.7557, 0.3469, -0.4441, -0.1073, 0.4912,
  8405. 0.2391],
  8406. [ 0.6059, -0.4442, 1.8921, -0.3422, -0.6230, -0.1461, 0.7489,
  8407. 0.0985],
  8408. [ 0.6125, -0.4153, 1.8885, -0.5461, -0.5134, -0.6539, 0.9814,
  8409. 0.2890],
  8410. [ 0.5747, -0.3886, 1.1494, -1.0388, -0.6000, -0.5846, 0.5952,
  8411. 0.3546],
  8412. [ 0.5992, -0.3968, 0.9307, -1.3497, -0.4730, -1.0465, 0.5259,
  8413. 0.2930],
  8414. [ 0.6454, -0.3984, 1.3804, 0.2545, -0.4557, -0.0380, 1.1057,
  8415. 0.3478]]], device='cuda:0')
  8416. loss_train_step before backward: tensor(0.0468, device='cuda:0', grad_fn=<MseLossBackward>)
  8417. loss_train_step after backward: tensor(0.0468, device='cuda:0', grad_fn=<MseLossBackward>)
  8418. loss_train: 0.5225014016032219
  8419. step: 10
  8420. running loss: 0.05225014016032219
  8421. Train Steps: 10/90 Loss: 0.0523 torch.Size([8, 600, 800])
  8422. torch.Size([8, 8])
  8423. tensor([[0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833],
  8424. [0.6200, 0.3999, 0.8653, 0.5207, 0.4100, 0.5125, 0.5975, 0.5103],
  8425. [0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767],
  8426. [0.6091, 0.3997, 0.8314, 0.4334, 0.3788, 0.4550, 0.5213, 0.5656],
  8427. [0.6293, 0.4024, 0.8750, 0.5000, 0.4012, 0.5733, 0.7121, 0.5633],
  8428. [0.6273, 0.4100, 0.7137, 0.2133, 0.4000, 0.2650, 0.6075, 0.5633],
  8429. [0.6277, 0.4118, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938],
  8430. [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773]],
  8431. device='cuda:0', dtype=torch.float64)
  8432. predictions are: tensor([[-0.4314, -1.0916, 0.9281, -1.1852, -0.4754, -1.1881, 0.1755, 0.3128],
  8433. [ 0.6351, -0.3751, 1.7415, 0.0037, -0.3866, -0.0694, 0.4654, 0.1710],
  8434. [ 0.9734, -0.1112, 1.7456, -0.4610, -0.5858, -0.7201, 0.5528, 0.2130],
  8435. [ 0.7597, -0.2628, 1.6129, -0.4537, -0.5275, -0.2665, 0.4179, 0.2051],
  8436. [ 0.7301, -0.2926, 1.8193, -0.0412, -0.3160, 0.2844, 0.7483, 0.2097],
  8437. [-0.0348, -0.7843, 1.0376, -1.5226, -0.3714, -1.4824, 0.4634, 0.2735],
  8438. [ 0.3638, -0.5736, 1.8100, -0.7922, -0.3260, -1.3180, 0.7065, 0.1236],
  8439. [ 0.1663, -0.6514, 1.0272, -1.2652, -0.3324, -1.2683, 0.3238, 0.3170]],
  8440. device='cuda:0', grad_fn=<AddmmBackward>)
  8441. landmarks are: tensor([[[ 0.5614, -0.4080, 0.7741, -0.8848, -0.5423, -0.9156, 0.3584,
  8442. 0.4085],
  8443. [ 0.5778, -0.4389, 1.7107, 0.1192, -0.3921, 0.0815, 0.4741,
  8444. 0.0711],
  8445. [ 0.6058, -0.3216, 1.8423, -0.2536, -0.5885, -0.6000, 0.3353,
  8446. 0.3777],
  8447. [ 0.5275, -0.4396, 1.5543, -0.2841, -0.5365, -0.1843, 0.1221,
  8448. 0.3265],
  8449. [ 0.6207, -0.4273, 1.7557, 0.0236, -0.4326, 0.3623, 1.0033,
  8450. 0.3157],
  8451. [ 0.6115, -0.3924, 1.0109, -1.3005, -0.4383, -1.0619, 0.5201,
  8452. 0.3161],
  8453. [ 0.6135, -0.3841, 1.8654, -0.5153, -0.4614, -1.0619, 0.6195,
  8454. -0.0049],
  8455. [ 0.5841, -0.3574, 1.0859, -0.9541, -0.2824, -1.2851, 0.3460,
  8456. 0.3808]]], device='cuda:0')
  8457. loss_train_step before backward: tensor(0.0607, device='cuda:0', grad_fn=<MseLossBackward>)
  8458. loss_train_step after backward: tensor(0.0607, device='cuda:0', grad_fn=<MseLossBackward>)
  8459. loss_train: 0.5832267887890339
  8460. step: 11
  8461. running loss: 0.05302061716263944
  8462. Train Steps: 11/90 Loss: 0.0530 torch.Size([8, 600, 800])
  8463. torch.Size([8, 8])
  8464. tensor([[0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383],
  8465. [0.6182, 0.3972, 0.8552, 0.5914, 0.3683, 0.4181, 0.5688, 0.5378],
  8466. [0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804],
  8467. [0.6222, 0.3937, 0.8350, 0.5617, 0.4138, 0.4600, 0.5800, 0.5233],
  8468. [0.6266, 0.4070, 0.8712, 0.5600, 0.3713, 0.4783, 0.5775, 0.6100],
  8469. [0.6271, 0.4040, 0.9138, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413],
  8470. [0.6229, 0.4066, 0.7612, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350],
  8471. [0.6353, 0.4128, 0.8488, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540]],
  8472. device='cuda:0', dtype=torch.float64)
  8473. predictions are: tensor([[ 0.6115, -0.3336, 1.6059, -0.4732, -0.3632, 0.0570, 0.5158, 0.2598],
  8474. [ 0.6079, -0.3892, 1.4494, -0.0570, -0.6873, -0.5817, 0.3053, 0.2103],
  8475. [ 0.7655, -0.2788, 1.4467, -0.8692, -0.6469, -0.6972, 0.4732, 0.1591],
  8476. [ 0.3337, -0.5651, 1.4831, -0.1191, -0.5095, -0.3914, 0.3344, 0.2294],
  8477. [ 0.4696, -0.4529, 1.4901, 0.0385, -0.6042, -0.0918, 0.3001, 0.2966],
  8478. [ 0.6387, -0.3689, 1.6943, -1.0324, -0.1716, -1.3331, 0.7935, 0.1650],
  8479. [ 0.3009, -0.5571, 1.0399, -1.4437, -0.4154, -1.3370, 0.4036, 0.2582],
  8480. [-0.2349, -0.9462, 1.4178, -1.5820, 0.0318, -1.4545, 0.7773, 0.2556]],
  8481. device='cuda:0', grad_fn=<AddmmBackward>)
  8482. landmarks are: tensor([[[ 0.5664, -0.4321, 1.8249, -0.2074, -0.2651, 0.4162, 0.5663,
  8483. 0.2006],
  8484. [ 0.5697, -0.4514, 1.6642, 0.4459, -0.5850, -0.3546, 0.3414,
  8485. 0.1982],
  8486. [ 0.6071, -0.4119, 1.7788, -0.5153, -0.6000, -0.5692, 0.6586,
  8487. -0.0670],
  8488. [ 0.5880, -0.4676, 1.5709, 0.3084, -0.3748, -0.1612, 0.3931,
  8489. 0.1313],
  8490. [ 0.6081, -0.4059, 1.7383, 0.3007, -0.5711, -0.0765, 0.3815,
  8491. 0.5316],
  8492. [ 0.6108, -0.4201, 1.9346, -0.5538, -0.1497, -1.0773, 1.0545,
  8493. 0.2142],
  8494. [ 0.5911, -0.4080, 1.2303, -0.9156, -0.3229, -1.2851, 0.4508,
  8495. 0.1852],
  8496. [ 0.6487, -0.3792, 1.6344, -1.0850, 0.2659, -1.5397, 0.8059,
  8497. 0.2730]]], device='cuda:0')
  8498. loss_train_step before backward: tensor(0.0612, device='cuda:0', grad_fn=<MseLossBackward>)
  8499. loss_train_step after backward: tensor(0.0612, device='cuda:0', grad_fn=<MseLossBackward>)
  8500. loss_train: 0.6444090195000172
  8501. step: 12
  8502. running loss: 0.05370075162500143
  8503.  
  8504. Train Steps: 12/90 Loss: 0.0537 torch.Size([8, 600, 800])
  8505. torch.Size([8, 8])
  8506. tensor([[0.6145, 0.4008, 0.8750, 0.5383, 0.3975, 0.4650, 0.5563, 0.5533],
  8507. [0.6124, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485],
  8508. [0.6030, 0.3969, 0.7988, 0.3917, 0.3450, 0.3667, 0.5266, 0.4700],
  8509. [0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250],
  8510. [0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343],
  8511. [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170],
  8512. [0.6371, 0.4092, 0.8337, 0.5850, 0.3950, 0.5117, 0.6559, 0.5262],
  8513. [0.6308, 0.3990, 0.8688, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133]],
  8514. device='cuda:0', dtype=torch.float64)
  8515. predictions are: tensor([[ 0.3441, -0.5397, 1.5219, 0.0440, -0.4011, -0.1755, 0.2975, 0.2755],
  8516. [ 0.1604, -0.6528, 0.9091, -1.3186, -0.5518, -1.0986, 0.2071, 0.3260],
  8517. [ 0.6327, -0.3453, 1.3894, -0.7139, -0.6229, -0.6713, 0.3165, 0.2111],
  8518. [ 0.5513, -0.4404, 1.7052, -0.7359, -0.4866, -1.0530, 0.6075, 0.1834],
  8519. [ 0.4135, -0.5046, 1.4512, -0.8828, -0.2866, -0.9949, 0.6271, 0.1815],
  8520. [ 0.3910, -0.5328, 1.3725, -1.4320, -0.0645, -1.5060, 0.8007, 0.2209],
  8521. [ 0.7513, -0.2595, 1.5456, -0.0458, -0.5058, -0.1437, 0.5182, 0.2074],
  8522. [ 0.5279, -0.4191, 1.5979, -0.1190, -0.4333, -0.1138, 0.4706, 0.2032]],
  8523. device='cuda:0', grad_fn=<AddmmBackward>)
  8524. landmarks are: tensor([[[ 0.5526, -0.4347, 1.7557, 0.2006, -0.4499, -0.1381, 0.2834,
  8525. 0.2699],
  8526. [ 0.5425, -0.3998, 0.9263, -0.8683, -0.6000, -1.0157, 0.0990,
  8527. 0.2476],
  8528. [ 0.4992, -0.4525, 1.4035, -0.4768, -0.6924, -0.5923, 0.1465,
  8529. -0.1151],
  8530. [ 0.6111, -0.3828, 1.8885, -0.3844, -0.5654, -0.8079, 0.5663,
  8531. 0.1390],
  8532. [ 0.6125, -0.4369, 1.9173, -0.5384, -0.2594, -0.8386, 0.9741,
  8533. 0.1821],
  8534. [ 0.6471, -0.3840, 1.5767, -1.0311, -0.0457, -1.5007, 0.6889,
  8535. 0.1020],
  8536. [ 0.6569, -0.3960, 1.5651, 0.4162, -0.4614, 0.0774, 0.7438,
  8537. 0.1447],
  8538. [ 0.6275, -0.4430, 1.7268, 0.1082, -0.4614, 0.0159, 0.6644,
  8539. 0.0851]]], device='cuda:0')
  8540. loss_train_step before backward: tensor(0.0372, device='cuda:0', grad_fn=<MseLossBackward>)
  8541. loss_train_step after backward: tensor(0.0372, device='cuda:0', grad_fn=<MseLossBackward>)
  8542. loss_train: 0.6816200353205204
  8543. step: 13
  8544. running loss: 0.0524323104092708
  8545. Train Steps: 13/90 Loss: 0.0524 torch.Size([8, 600, 800])
  8546. torch.Size([8, 8])
  8547. tensor([[0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742],
  8548. [0.6179, 0.4118, 0.7278, 0.4237, 0.3588, 0.3400, 0.5675, 0.5917],
  8549. [0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869],
  8550. [0.6187, 0.4104, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683],
  8551. [ nan, nan, 0.7512, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617],
  8552. [0.6069, 0.3975, 0.8625, 0.5083, 0.4388, 0.5483, 0.5650, 0.4967],
  8553. [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268],
  8554. [0.6229, 0.4066, 0.7612, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350]],
  8555. device='cuda:0', dtype=torch.float64)
  8556. predictions are: tensor([[ 0.6906, -0.3250, 1.7244, 0.2441, -0.6534, -0.5400, 0.4928, 0.0983],
  8557. [ 0.7984, -0.2507, 1.3635, -0.5323, -0.6371, -0.7391, 0.3126, 0.2900],
  8558. [ 0.6796, -0.2984, 1.4948, -1.3600, -0.1548, -1.2440, 0.7837, 0.1997],
  8559. [ 0.6440, -0.3274, 1.0033, -1.1558, -0.4957, -1.1206, 0.3775, 0.2980],
  8560. [-0.6363, -1.1912, 0.9427, -1.3059, -0.3338, -1.3175, 0.2988, 0.2964],
  8561. [ 0.5409, -0.4155, 1.6573, 0.1337, -0.2963, 0.1204, 0.4237, 0.1673],
  8562. [ 0.4925, -0.4208, 1.6593, -0.0255, -0.3210, 0.0899, 0.5519, 0.1780],
  8563. [ 0.5728, -0.3631, 1.1779, -1.0559, -0.3516, -1.1277, 0.5243, 0.2186]],
  8564. device='cuda:0', grad_fn=<AddmmBackward>)
  8565. landmarks are: tensor([[[ 0.6113, -0.4057, 1.7499, 0.3007, -0.5885, -0.5384, 0.4513,
  8566. -0.0957],
  8567. [ 0.5680, -0.3840, 1.0756, -0.3290, -0.6289, -0.7155, 0.3353,
  8568. 0.4470],
  8569. [ 0.6127, -0.3944, 1.5189, -1.2467, -0.1323, -1.4622, 0.5646,
  8570. -0.0369],
  8571. [ 0.5718, -0.3905, 1.0053, -1.3305, -0.4614, -1.1235, 0.4450,
  8572. 0.3392],
  8573. [-2.2859, -2.2859, 1.1841, -1.3082, -0.3055, -1.3621, 0.3007,
  8574. 0.3084],
  8575. [ 0.5174, -0.4497, 1.6979, 0.0620, -0.2594, 0.2468, 0.3238,
  8576. 0.0082],
  8577. [ 0.5828, -0.4682, 1.7031, -0.0497, -0.2458, 0.0818, 0.6381,
  8578. 0.1474],
  8579. [ 0.5911, -0.4080, 1.2303, -0.9156, -0.3229, -1.2851, 0.4508,
  8580. 0.1852]]], device='cuda:0')
  8581. loss_train_step before backward: tensor(0.0724, device='cuda:0', grad_fn=<MseLossBackward>)
  8582. loss_train_step after backward: tensor(0.0724, device='cuda:0', grad_fn=<MseLossBackward>)
  8583. loss_train: 0.7540136612951756
  8584. step: 14
  8585. running loss: 0.053858118663941114
  8586. Train Steps: 14/90 Loss: 0.0539 torch.Size([8, 600, 800])
  8587. torch.Size([8, 8])
  8588. tensor([[0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467],
  8589. [0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631],
  8590. [0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767],
  8591. [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290],
  8592. [0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274],
  8593. [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261],
  8594. [0.6262, 0.4163, 0.8850, 0.5183, 0.3763, 0.4150, 0.6025, 0.5500],
  8595. [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402]],
  8596. device='cuda:0', dtype=torch.float64)
  8597. predictions are: tensor([[ 3.8594e-01, -4.9731e-01, 1.3426e+00, -7.9972e-01, -5.6520e-01,
  8598. -5.4576e-01, 5.2629e-01, 1.8728e-01],
  8599. [ 4.3697e-01, -4.5928e-01, 1.3931e+00, -5.5027e-01, -4.7881e-01,
  8600. -1.3916e-01, 3.5979e-01, 2.4473e-01],
  8601. [ 8.3286e-01, -2.0214e-01, 1.5449e+00, -3.9255e-01, -5.2480e-01,
  8602. -7.8448e-01, 4.5361e-01, 2.3245e-01],
  8603. [ 5.0168e-01, -4.3097e-01, 1.3078e+00, -9.1020e-01, -5.8755e-01,
  8604. -6.2334e-01, 5.8320e-01, 1.7699e-01],
  8605. [ 7.1887e-01, -3.1333e-01, 1.5627e+00, -1.1709e+00, -4.1819e-04,
  8606. -1.5454e+00, 8.5491e-01, 2.0092e-01],
  8607. [ 4.4513e-01, -4.4595e-01, 1.5378e+00, -2.1944e-01, -4.1386e-01,
  8608. -3.5445e-01, 3.4656e-01, 1.7412e-01],
  8609. [ 6.1162e-01, -3.6164e-01, 1.5413e+00, -1.8645e-01, -4.5866e-01,
  8610. -7.5103e-01, 3.8984e-01, 2.1871e-01],
  8611. [ 2.9098e-01, -5.9695e-01, 1.4410e+00, 1.1175e-01, -3.4529e-01,
  8612. -4.0318e-01, 3.0012e-01, 1.9724e-01]], device='cuda:0',
  8613. grad_fn=<AddmmBackward>)
  8614. landmarks are: tensor([[[ 0.5484, -0.4591, 1.5940, -0.7771, -0.6693, -0.3460, 0.4219,
  8615. 0.2391],
  8616. [ 0.5027, -0.4289, 1.5478, -0.4229, -0.6346, -0.0303, 0.3223,
  8617. 0.3148],
  8618. [ 0.6058, -0.3216, 1.8423, -0.2536, -0.5885, -0.6000, 0.3353,
  8619. 0.3777],
  8620. [ 0.5697, -0.4538, 1.5308, -0.8703, -0.6572, -0.3639, 0.5739,
  8621. 0.1576],
  8622. [ 0.6537, -0.3719, 1.6979, -0.8617, -0.0169, -1.3621, 0.6926,
  8623. 0.1501],
  8624. [ 0.5746, -0.3623, 1.7961, -0.1150, -0.3691, -0.0380, 0.2208,
  8625. 0.1439],
  8626. [ 0.6066, -0.3632, 1.8018, 0.1082, -0.5480, -0.3691, 0.4970,
  8627. 0.2545],
  8628. [ 0.5780, -0.4565, 1.6221, 0.2532, -0.3728, -0.1718, 0.4357,
  8629. 0.2091]]], device='cuda:0')
  8630. loss_train_step before backward: tensor(0.0246, device='cuda:0', grad_fn=<MseLossBackward>)
  8631. loss_train_step after backward: tensor(0.0246, device='cuda:0', grad_fn=<MseLossBackward>)
  8632. loss_train: 0.7785959504544735
  8633. step: 15
  8634. running loss: 0.0519063966969649
  8635. Train Steps: 15/90 Loss: 0.0519 torch.Size([8, 600, 800])
  8636. torch.Size([8, 8])
  8637. tensor([[0.6022, 0.3994, 0.8025, 0.3350, 0.3350, 0.4400, 0.5565, 0.5025],
  8638. [0.6271, 0.4040, 0.9138, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413],
  8639. [0.6097, 0.4000, 0.7325, 0.2667, 0.3450, 0.3517, 0.5284, 0.5045],
  8640. [0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742],
  8641. [0.6122, 0.4006, 0.8850, 0.4217, 0.4088, 0.5517, 0.6063, 0.5517],
  8642. [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082],
  8643. [0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148],
  8644. [0.6261, 0.4029, 0.8720, 0.3364, 0.3665, 0.3753, 0.6531, 0.5183]],
  8645. device='cuda:0', dtype=torch.float64)
  8646. predictions are: tensor([[ 0.3495, -0.5137, 1.2867, -0.7977, -0.5919, -0.4932, 0.3754, 0.2001],
  8647. [ 0.7947, -0.2719, 1.7608, -0.7762, -0.0301, -1.4611, 0.8805, 0.1307],
  8648. [ 0.0306, -0.7537, 1.0149, -1.0745, -0.5246, -0.9820, 0.2472, 0.2447],
  8649. [ 0.6324, -0.3694, 1.6234, 0.0647, -0.5520, -0.8712, 0.4582, 0.1424],
  8650. [ 0.7833, -0.2346, 1.5707, -0.3463, -0.3770, 0.0528, 0.5279, 0.2255],
  8651. [ 0.7305, -0.2862, 1.5712, 0.0126, -0.2571, 0.0699, 0.4072, 0.2189],
  8652. [ 0.4863, -0.4808, 1.5423, 0.0540, -0.5026, -0.2989, 0.2758, 0.1930],
  8653. [ 0.5454, -0.4141, 1.3707, -0.7669, -0.4660, -0.6807, 0.5137, 0.2013]],
  8654. device='cuda:0', grad_fn=<AddmmBackward>)
  8655. landmarks are: tensor([[[ 0.4956, -0.4413, 1.4208, -0.7386, -0.7386, -0.2536, 0.2843,
  8656. 0.0351],
  8657. [ 0.6108, -0.4201, 1.9346, -0.5538, -0.1497, -1.0773, 1.0545,
  8658. 0.2142],
  8659. [ 0.5303, -0.4384, 1.0975, -1.0542, -0.6924, -0.6616, 0.1548,
  8660. 0.0442],
  8661. [ 0.6113, -0.4057, 1.7499, 0.3007, -0.5885, -0.5384, 0.4513,
  8662. -0.0957],
  8663. [ 0.5417, -0.4355, 1.8018, -0.3383, -0.3979, 0.2622, 0.5143,
  8664. 0.2622],
  8665. [ 0.5450, -0.4706, 1.7643, 0.0722, -0.3708, 0.3200, 0.4854,
  8666. 0.0612],
  8667. [ 0.5773, -0.4316, 1.7399, 0.1287, -0.5153, -0.0817, 0.4313,
  8668. 0.0919],
  8669. [ 0.6059, -0.4249, 1.7420, -0.7321, -0.5931, -0.5523, 0.7307,
  8670. 0.1080]]], device='cuda:0')
  8671. loss_train_step before backward: tensor(0.0321, device='cuda:0', grad_fn=<MseLossBackward>)
  8672. loss_train_step after backward: tensor(0.0321, device='cuda:0', grad_fn=<MseLossBackward>)
  8673. loss_train: 0.8107367865741253
  8674. step: 16
  8675. running loss: 0.05067104916088283
  8676.  
  8677. Train Steps: 16/90 Loss: 0.0507 torch.Size([8, 600, 800])
  8678. torch.Size([8, 8])
  8679. tensor([[0.6078, 0.4033, 0.8019, 0.3055, 0.3450, 0.4200, 0.6025, 0.5550],
  8680. [0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933],
  8681. [0.6264, 0.4049, 0.8988, 0.4633, 0.3813, 0.4983, 0.6326, 0.4843],
  8682. [0.6175, 0.3957, 0.8700, 0.4817, 0.4662, 0.5133, 0.5800, 0.5517],
  8683. [0.6097, 0.4000, 0.7325, 0.2667, 0.3450, 0.3517, 0.5284, 0.5045],
  8684. [0.6087, 0.3951, 0.8387, 0.5833, 0.4188, 0.4933, 0.5146, 0.4830],
  8685. [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220],
  8686. [0.6204, 0.4055, 0.8438, 0.5733, 0.4574, 0.4801, 0.5487, 0.5617]],
  8687. device='cuda:0', dtype=torch.float64)
  8688. predictions are: tensor([[ 0.6864, -0.3161, 1.3957, -0.8741, -0.6393, -0.4115, 0.5721, 0.1944],
  8689. [ 0.6869, -0.2809, 1.5808, 0.1829, -0.1508, -0.4284, 0.3990, 0.2674],
  8690. [ 1.0103, -0.1350, 1.8808, -0.1997, -0.6168, -0.1198, 0.6180, 0.0456],
  8691. [ 0.4554, -0.4768, 1.7386, -0.1356, -0.2300, -0.1834, 0.4971, 0.1328],
  8692. [ 0.2051, -0.6472, 1.1485, -1.0484, -0.5905, -0.9150, 0.3315, 0.1749],
  8693. [ 0.6647, -0.3565, 1.6234, 0.2364, -0.4079, -0.3115, 0.3650, 0.1571],
  8694. [ 0.0915, -0.7536, 0.9806, -1.4417, -0.3319, -1.7779, 0.4103, 0.1761],
  8695. [ 0.6232, -0.3394, 1.5547, 0.1898, -0.2631, -0.2693, 0.3681, 0.2310]],
  8696. device='cuda:0', grad_fn=<AddmmBackward>)
  8697. landmarks are: tensor([[[ 0.5215, -0.4232, 1.4182, -0.8747, -0.6924, -0.3460, 0.4970,
  8698. 0.2776],
  8699. [ 0.5959, -0.3579, 1.6055, 0.3623, -0.0573, -0.2074, 0.3122,
  8700. 0.4547],
  8701. [ 0.6075, -0.4159, 1.8654, -0.1458, -0.5249, 0.0159, 0.6359,
  8702. -0.0490],
  8703. [ 0.5662, -0.4581, 1.7326, -0.0611, -0.1323, 0.0851, 0.3931,
  8704. 0.2622],
  8705. [ 0.5303, -0.4384, 1.0975, -1.0542, -0.6924, -0.6616, 0.1548,
  8706. 0.0442],
  8707. [ 0.5258, -0.4610, 1.5882, 0.4085, -0.3517, -0.0072, 0.0910,
  8708. -0.0550],
  8709. [ 0.5566, -0.4160, 0.9935, -1.3313, -0.2824, -1.5161, 0.2144,
  8710. 0.1253],
  8711. [ 0.5795, -0.4129, 1.6113, 0.3623, -0.1733, -0.0684, 0.2487,
  8712. 0.3084]]], device='cuda:0')
  8713. loss_train_step before backward: tensor(0.0270, device='cuda:0', grad_fn=<MseLossBackward>)
  8714. loss_train_step after backward: tensor(0.0270, device='cuda:0', grad_fn=<MseLossBackward>)
  8715. loss_train: 0.8377538155764341
  8716. step: 17
  8717. running loss: 0.04927963621037848
  8718. Train Steps: 17/90 Loss: 0.0493 torch.Size([8, 600, 800])
  8719. torch.Size([8, 8])
  8720. tensor([[0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083],
  8721. [0.6115, 0.3998, 0.7063, 0.2383, 0.4037, 0.1950, 0.5320, 0.4993],
  8722. [0.6251, 0.4163, 0.8662, 0.4467, 0.3625, 0.3567, 0.6038, 0.5533],
  8723. [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123],
  8724. [0.6222, 0.3937, 0.8350, 0.5617, 0.4138, 0.4600, 0.5800, 0.5233],
  8725. [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274],
  8726. [0.6137, 0.4038, 0.8563, 0.4050, 0.3813, 0.2550, 0.5106, 0.4954],
  8727. [0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153]],
  8728. device='cuda:0', dtype=torch.float64)
  8729. predictions are: tensor([[ 0.4288, -0.4854, 1.6467, 0.1719, -0.4551, 0.0582, 0.4286, 0.1805],
  8730. [ 0.5306, -0.4610, 0.9986, -1.3014, -0.4066, -1.3574, 0.3123, 0.1790],
  8731. [ 0.7175, -0.2925, 1.6305, -0.1981, -0.5236, -0.6448, 0.4414, 0.1654],
  8732. [ 0.5649, -0.4179, 1.0909, -1.1380, -0.4752, -1.0641, 0.3576, 0.1617],
  8733. [ 0.4000, -0.5173, 1.6446, 0.3024, -0.3259, -0.0621, 0.4557, 0.1149],
  8734. [ 0.5420, -0.4195, 1.7541, 0.0212, -0.4687, -0.2140, 0.5426, 0.1165],
  8735. [ 0.5727, -0.4093, 1.5485, -0.5428, -0.4685, -0.9696, 0.3440, 0.1536],
  8736. [ 0.6472, -0.3293, 1.6709, 0.1153, -0.1787, 0.1713, 0.4548, 0.1555]],
  8737. device='cuda:0', grad_fn=<AddmmBackward>)
  8738. landmarks are: tensor([[[ 0.5713, -0.4105, 1.7557, 0.0467, -0.6520, -0.2690, 0.3988,
  8739. 0.5239],
  8740. [ 0.5384, -0.4393, 0.9762, -1.1851, -0.4210, -1.3852, 0.1712,
  8741. 0.0201],
  8742. [ 0.6016, -0.3633, 1.7152, -0.2228, -0.6115, -0.6385, 0.5028,
  8743. 0.2699],
  8744. [ 0.5342, -0.4280, 1.0051, -1.0619, -0.5711, -1.0388, 0.2226,
  8745. 0.0802],
  8746. [ 0.5880, -0.4676, 1.5709, 0.3084, -0.3748, -0.1612, 0.3931,
  8747. 0.1313],
  8748. [ 0.6108, -0.4008, 1.9088, -0.0253, -0.5769, -0.3075, 0.7905,
  8749. 0.1499],
  8750. [ 0.5485, -0.4209, 1.6691, -0.4152, -0.5249, -1.1081, 0.0725,
  8751. 0.0021],
  8752. [ 0.5548, -0.4682, 1.7309, 0.0966, -0.1394, 0.1757, 0.5045,
  8753. 0.0942]]], device='cuda:0')
  8754. loss_train_step before backward: tensor(0.0131, device='cuda:0', grad_fn=<MseLossBackward>)
  8755. loss_train_step after backward: tensor(0.0131, device='cuda:0', grad_fn=<MseLossBackward>)
  8756. loss_train: 0.8508864175528288
  8757. step: 18
  8758. running loss: 0.04727146764182382
  8759. Train Steps: 18/90 Loss: 0.0473 torch.Size([8, 600, 800])
  8760. torch.Size([8, 8])
  8761. tensor([[0.6143, 0.4040, 0.8237, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973],
  8762. [0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367],
  8763. [0.6097, 0.4024, 0.8488, 0.3717, 0.3875, 0.5517, 0.5836, 0.5591],
  8764. [0.6339, 0.4118, 0.7988, 0.5800, 0.3912, 0.4583, 0.7343, 0.5760],
  8765. [0.6198, 0.4164, 0.8700, 0.5067, 0.4625, 0.5650, 0.5464, 0.5197],
  8766. [0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938],
  8767. [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896],
  8768. [0.6275, 0.4157, 0.8337, 0.5800, 0.3763, 0.4200, 0.5547, 0.6125]],
  8769. device='cuda:0', dtype=torch.float64)
  8770. predictions are: tensor([[ 0.4882, -0.4754, 1.3462, -1.0380, -0.4194, -1.2863, 0.3462, 0.1151],
  8771. [ 0.6720, -0.3324, 1.6458, -0.4131, -0.3798, -1.0912, 0.4214, 0.1275],
  8772. [ 0.5824, -0.3736, 1.5879, -0.6035, -0.3317, 0.1944, 0.6438, 0.1732],
  8773. [ 0.5314, -0.4397, 1.4854, -0.0471, -0.4339, -0.2989, 0.4611, 0.1603],
  8774. [ 0.3135, -0.5562, 1.6645, 0.0351, -0.2306, 0.0211, 0.3456, 0.1590],
  8775. [ 0.6142, -0.3721, 1.6466, 0.2753, -0.4500, -0.0996, 0.4397, 0.0354],
  8776. [ 0.5489, -0.4391, 1.4079, -0.5499, -0.6380, -0.7958, 0.2248, 0.2013],
  8777. [ 0.6976, -0.3084, 1.5207, 0.3261, -0.4993, -0.1174, 0.3380, 0.2032]],
  8778. device='cuda:0', grad_fn=<AddmmBackward>)
  8779. landmarks are: tensor([[[ 0.5515, -0.4201, 1.5189, -0.7463, -0.4037, -1.3082, 0.0869,
  8780. 0.0111],
  8781. [ 0.5900, -0.3932, 1.8307, -0.3921, -0.4268, -1.1851, 0.3758,
  8782. 0.1929],
  8783. [ 0.5301, -0.4273, 1.6344, -0.5692, -0.4961, 0.2622, 0.4098,
  8784. 0.2966],
  8785. [ 0.6421, -0.3840, 1.4035, 0.3931, -0.4788, -0.1689, 1.1057,
  8786. 0.3745],
  8787. [ 0.5770, -0.3624, 1.7326, 0.0543, -0.1497, 0.3238, 0.2378,
  8788. 0.1146],
  8789. [ 0.6395, -0.4213, 1.7037, 0.3623, -0.4326, -0.1073, 0.6560,
  8790. -0.0049],
  8791. [ 0.5485, -0.3997, 1.4445, -0.4895, -0.6000, -0.8309, 0.1878,
  8792. 0.4374],
  8793. [ 0.6124, -0.3658, 1.5651, 0.3931, -0.5480, -0.3460, 0.2761,
  8794. 0.5431]]], device='cuda:0')
  8795. loss_train_step before backward: tensor(0.0252, device='cuda:0', grad_fn=<MseLossBackward>)
  8796. loss_train_step after backward: tensor(0.0252, device='cuda:0', grad_fn=<MseLossBackward>)
  8797. loss_train: 0.8761011473834515
  8798. step: 19
  8799. running loss: 0.046110586704392185
  8800. Train Steps: 19/90 Loss: 0.0461 torch.Size([8, 600, 800])
  8801. torch.Size([8, 8])
  8802. tensor([[0.6196, 0.4088, 0.8888, 0.4583, 0.4500, 0.5683, 0.6138, 0.5883],
  8803. [0.6216, 0.4167, 0.8588, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667],
  8804. [0.6133, 0.4066, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358],
  8805. [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960],
  8806. [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717],
  8807. [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220],
  8808. [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583],
  8809. [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834]],
  8810. device='cuda:0', dtype=torch.float64)
  8811. predictions are: tensor([[ 0.7217, -0.2845, 1.9075, 0.2389, -0.2370, 0.5157, 0.5562, 0.1797],
  8812. [ 0.6534, -0.3390, 1.8560, 0.8615, -0.4630, 0.5337, 0.4289, 0.1413],
  8813. [ 0.4276, -0.5171, 1.1000, -0.8050, -0.5874, -1.0325, 0.1186, 0.1669],
  8814. [ 0.5049, -0.4677, 1.8860, 0.3048, -0.2772, 0.1354, 0.3989, 0.0883],
  8815. [ 0.4303, -0.5046, 1.0362, -0.7526, -0.5927, -0.9086, 0.1313, 0.2303],
  8816. [ 0.2386, -0.6472, 1.1012, -0.9978, -0.3905, -1.3685, 0.2712, 0.1517],
  8817. [ 0.6727, -0.3058, 1.6422, -0.0121, -0.6491, -0.0651, 0.3661, 0.1395],
  8818. [ 0.7852, -0.3123, 1.5772, -0.9977, -0.1710, -1.3506, 0.8036, 0.0506]],
  8819. device='cuda:0', grad_fn=<AddmmBackward>)
  8820. landmarks are: tensor([[[ 0.5759, -0.3978, 1.8192, -0.1689, -0.2074, 0.3392, 0.5490,
  8821. 0.4316],
  8822. [ 0.5853, -0.3614, 1.6806, 0.2930, -0.4499, 0.1005, 0.3815,
  8823. 0.3315],
  8824. [ 0.5470, -0.4081, 0.8492, -1.0773, -0.5307, -1.1620, 0.0912,
  8825. 0.1890],
  8826. [ 0.5205, -0.4577, 1.7326, -0.1304, -0.1862, 0.0447, 0.1301,
  8827. 0.0051],
  8828. [ 0.5633, -0.3883, 0.8648, -1.0288, -0.5192, -1.1081, 0.2776,
  8829. 0.3546],
  8830. [ 0.5566, -0.4160, 0.9935, -1.3313, -0.2824, -1.5161, 0.2144,
  8831. 0.1253],
  8832. [ 0.5761, -0.3966, 1.6171, -0.4845, -0.6346, -0.4691, 0.4739,
  8833. 0.2930],
  8834. [ 0.6126, -0.4057, 1.4439, -1.3159, -0.1150, -1.5777, 0.5537,
  8835. -0.0530]]], device='cuda:0')
  8836. loss_train_step before backward: tensor(0.0413, device='cuda:0', grad_fn=<MseLossBackward>)
  8837.  
  8838. loss_train_step after backward: tensor(0.0413, device='cuda:0', grad_fn=<MseLossBackward>)
  8839. loss_train: 0.9173775799572468
  8840. step: 20
  8841. running loss: 0.045868878997862336
  8842. Train Steps: 20/90 Loss: 0.0459 torch.Size([8, 600, 800])
  8843. torch.Size([8, 8])
  8844. tensor([[0.6264, 0.3972, 0.8853, 0.4771, 0.3853, 0.4511, 0.6293, 0.5334],
  8845. [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377],
  8846. [0.6147, 0.4081, 0.8538, 0.3400, 0.3663, 0.3150, 0.5142, 0.4875],
  8847. [0.6250, 0.3993, 0.9138, 0.4333, 0.3763, 0.5217, 0.6995, 0.5320],
  8848. [ nan, nan, 0.7512, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617],
  8849. [0.6203, 0.4096, 0.8862, 0.4267, 0.3538, 0.4117, 0.6025, 0.5650],
  8850. [0.6185, 0.4067, 0.8838, 0.4450, 0.4037, 0.4733, 0.5213, 0.5142],
  8851. [ nan, nan, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999]],
  8852. device='cuda:0', dtype=torch.float64)
  8853. predictions are: tensor([[ 0.7893, -0.2475, 1.7366, 0.2809, -0.4846, 0.0459, 0.5071, 0.1226],
  8854. [ 0.5180, -0.4447, 1.6718, 0.4874, -0.3442, 0.0526, 0.3643, 0.1179],
  8855. [ 1.0198, -0.1212, 1.6432, -0.3591, -0.6406, -0.6661, 0.2950, 0.1079],
  8856. [ 0.9893, -0.1316, 1.7666, 0.0423, -0.4163, 0.3851, 0.7108, 0.1294],
  8857. [-0.4476, -1.0955, 1.0936, -1.0599, -0.2677, -1.2739, 0.2665, 0.2393],
  8858. [ 1.0255, -0.0901, 1.7566, -0.0326, -0.5185, -0.2372, 0.4457, 0.1847],
  8859. [ 0.7385, -0.2714, 1.6558, 0.0937, -0.4627, 0.0625, 0.1644, 0.1464],
  8860. [-0.2213, -0.9596, 1.1754, -1.1070, -0.3095, -1.3147, 0.2532, 0.1814]],
  8861. device='cuda:0', grad_fn=<AddmmBackward>)
  8862. landmarks are: tensor([[[ 0.6075, -0.4514, 1.8032, -0.0822, -0.5061, -0.2023, 0.6208,
  8863. 0.1779],
  8864. [ 0.5775, -0.4607, 1.6741, 0.1962, -0.4036, -0.1212, 0.4588,
  8865. 0.1979],
  8866. [ 0.5532, -0.4008, 1.6575, -0.7155, -0.5942, -0.8309, 0.0890,
  8867. -0.0340],
  8868. [ 0.6010, -0.4417, 1.9346, -0.2844, -0.5480, 0.1236, 0.9448,
  8869. 0.1715],
  8870. [-2.2859, -2.2859, 1.1841, -1.3082, -0.3055, -1.3621, 0.3007,
  8871. 0.3084],
  8872. [ 0.5790, -0.3940, 1.8076, -0.3152, -0.6520, -0.3844, 0.4970,
  8873. 0.3238],
  8874. [ 0.5708, -0.4075, 1.7961, -0.2305, -0.4210, -0.0996, 0.1219,
  8875. 0.0893],
  8876. [-2.2859, -2.2859, 1.3400, -1.0388, -0.3055, -1.4930, 0.1157,
  8877. 0.0231]]], device='cuda:0')
  8878. loss_train_step before backward: tensor(0.2049, device='cuda:0', grad_fn=<MseLossBackward>)
  8879. loss_train_step after backward: tensor(0.2049, device='cuda:0', grad_fn=<MseLossBackward>)
  8880. loss_train: 1.122323576360941
  8881. step: 21
  8882. running loss: 0.05344397982671147
  8883. Train Steps: 21/90 Loss: 0.0534 torch.Size([8, 600, 800])
  8884. torch.Size([8, 8])
  8885. tensor([[0.6245, 0.4115, 0.8700, 0.4883, 0.4625, 0.5517, 0.6100, 0.5217],
  8886. [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967],
  8887. [0.6250, 0.4106, 0.8700, 0.3717, 0.3588, 0.4967, 0.6038, 0.5167],
  8888. [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466],
  8889. [0.6176, 0.3911, 0.8738, 0.4217, 0.3488, 0.4033, 0.6025, 0.4817],
  8890. [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5637, 0.5633],
  8891. [0.6113, 0.4006, 0.8700, 0.5350, 0.3638, 0.3767, 0.5097, 0.4882],
  8892. [ nan, nan, 0.7412, 0.2200, 0.4450, 0.1517, 0.5312, 0.4983]],
  8893. device='cuda:0', dtype=torch.float64)
  8894. predictions are: tensor([[ 0.8088, -0.2508, 1.7559, 0.1688, -0.2472, 0.3217, 0.4909, 0.2069],
  8895. [ 0.9829, -0.1435, 1.6416, -0.8073, -0.1479, -1.0794, 0.5959, 0.1515],
  8896. [ 0.8976, -0.1853, 1.6383, -0.2905, -0.5018, 0.1907, 0.5584, 0.1761],
  8897. [ 0.5843, -0.3556, 1.3610, -0.4377, -0.6175, -0.2176, 0.2037, 0.2051],
  8898. [ 0.6115, -0.3894, 1.6423, -0.0639, -0.6122, -0.2054, 0.3690, 0.0912],
  8899. [ 0.8306, -0.2088, 1.6910, 0.3271, -0.6185, -0.2251, 0.2638, 0.2357],
  8900. [ 0.4336, -0.5279, 1.7140, 0.4916, -0.5995, -0.3352, 0.2393, 0.1133],
  8901. [-0.9037, -1.4468, 1.1021, -1.1638, -0.2849, -1.4912, 0.2202, 0.1359]],
  8902. device='cuda:0', grad_fn=<AddmmBackward>)
  8903. landmarks are: tensor([[[ 5.9873e-01, -3.8522e-01, 1.7326e+00, -3.0331e-02, -1.4965e-01,
  8904. 2.6220e-01, 5.3164e-01, 1.2363e-01],
  8905. [ 6.0479e-01, -3.7229e-01, 1.6517e+00, -1.0773e+00, 4.6189e-04,
  8906. -1.5161e+00, 5.8660e-01, 8.0947e-03],
  8907. [ 6.0098e-01, -3.8961e-01, 1.7326e+00, -5.6921e-01, -6.2887e-01,
  8908. 8.1601e-03, 5.0277e-01, 1.0054e-01],
  8909. [ 5.0762e-01, -4.4426e-01, 1.2337e+00, -5.0235e-01, -6.8083e-01,
  8910. -3.6135e-01, 8.6614e-02, 2.3862e-01],
  8911. [ 5.6692e-01, -4.7937e-01, 1.7499e+00, -3.3826e-01, -6.7506e-01,
  8912. -4.2294e-01, 4.9700e-01, -6.1124e-02],
  8913. [ 5.9024e-01, -3.4927e-01, 1.7961e+00, -7.2363e-03, -5.9423e-01,
  8914. -5.6151e-01, 3.1801e-01, 3.1609e-01],
  8915. [ 5.3788e-01, -4.3580e-01, 1.7326e+00, 1.8522e-01, -6.0577e-01,
  8916. -5.4611e-01, 6.8408e-02, -3.0981e-02],
  8917. [-2.2859e+00, -2.2859e+00, 1.1379e+00, -1.2697e+00, -2.3048e-01,
  8918. -1.5854e+00, 1.6790e-01, 1.5858e-02]]], device='cuda:0')
  8919. loss_train_step before backward: tensor(0.0680, device='cuda:0', grad_fn=<MseLossBackward>)
  8920. loss_train_step after backward: tensor(0.0680, device='cuda:0', grad_fn=<MseLossBackward>)
  8921. loss_train: 1.1902863197028637
  8922. step: 22
  8923. running loss: 0.05410392362285744
  8924. Train Steps: 22/90 Loss: 0.0541 torch.Size([8, 600, 800])
  8925. torch.Size([8, 8])
  8926. tensor([[0.6259, 0.4156, 0.8812, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960],
  8927. [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617],
  8928. [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334],
  8929. [0.6346, 0.4092, 0.7712, 0.5917, 0.4037, 0.4767, 0.7343, 0.5725],
  8930. [0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272],
  8931. [0.6156, 0.4125, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084],
  8932. [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550],
  8933. [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383]],
  8934. device='cuda:0', dtype=torch.float64)
  8935. predictions are: tensor([[ 0.5841, -0.4228, 1.7081, -0.6441, -0.2441, -1.1153, 0.5499, 0.0982],
  8936. [ 0.4278, -0.5337, 1.6374, -0.0478, -0.6051, -0.2112, 0.1915, 0.1604],
  8937. [ 0.6027, -0.4234, 1.2634, -0.8264, -0.6503, -0.8771, 0.4061, 0.1357],
  8938. [ 0.4495, -0.5041, 1.4335, 0.1495, -0.5340, -0.1502, 0.3682, 0.2350],
  8939. [ 0.5330, -0.4750, 1.5403, -0.4251, -0.6036, -0.9126, 0.1916, 0.1295],
  8940. [ 0.4893, -0.4719, 1.7569, 0.1496, -0.3485, 0.2388, 0.4058, 0.1906],
  8941. [ 0.4003, -0.5475, 1.7447, -0.0725, -0.2012, -0.1510, 0.3865, 0.1682],
  8942. [ 0.4466, -0.5168, 1.7576, 0.0200, -0.3755, 0.2759, 0.4909, 0.1671]],
  8943. device='cuda:0', grad_fn=<AddmmBackward>)
  8944. landmarks are: tensor([[[ 0.6052, -0.3663, 1.7845, -0.8156, -0.0804, -1.4237, 0.5866,
  8945. 0.0051],
  8946. [ 0.5488, -0.4221, 1.8018, -0.2459, -0.4845, -0.3075, 0.2309,
  8947. 0.3087],
  8948. [ 0.5730, -0.4562, 1.2195, -1.2440, -0.5497, -0.7711, 0.5704,
  8949. 0.1779],
  8950. [ 0.6454, -0.3960, 1.2764, 0.4470, -0.4210, -0.0842, 1.1057,
  8951. 0.3585],
  8952. [ 0.5711, -0.4015, 1.6979, -0.6770, -0.5365, -1.0619, 0.1712,
  8953. 0.1494],
  8954. [ 0.5573, -0.3808, 1.8018, -0.0534, -0.1843, 0.4008, 0.3769,
  8955. 0.0622],
  8956. [ 0.5351, -0.4321, 1.7326, -0.1381, 0.0463, 0.1049, 0.4162,
  8957. 0.2776],
  8958. [ 0.5664, -0.4321, 1.8249, -0.2074, -0.2651, 0.4162, 0.5663,
  8959. 0.2006]]], device='cuda:0')
  8960. loss_train_step before backward: tensor(0.0284, device='cuda:0', grad_fn=<MseLossBackward>)
  8961. loss_train_step after backward: tensor(0.0284, device='cuda:0', grad_fn=<MseLossBackward>)
  8962. loss_train: 1.2187356799840927
  8963. step: 23
  8964. running loss: 0.05298850782539533
  8965. Train Steps: 23/90 Loss: 0.0530 torch.Size([8, 600, 800])
  8966. torch.Size([8, 8])
  8967. tensor([[0.6266, 0.4067, 0.8588, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267],
  8968. [0.6203, 0.4078, 0.8800, 0.5083, 0.3900, 0.5000, 0.6100, 0.5583],
  8969. [0.6265, 0.4088, 0.8025, 0.1850, 0.4163, 0.2500, 0.6290, 0.4947],
  8970. [0.6268, 0.4102, 0.8938, 0.3667, 0.4025, 0.2833, 0.6275, 0.5183],
  8971. [0.6166, 0.4008, 0.8563, 0.5667, 0.4388, 0.4933, 0.5575, 0.5567],
  8972. [0.6136, 0.4117, 0.8700, 0.5167, 0.4188, 0.5083, 0.5147, 0.5495],
  8973. [0.6201, 0.3970, 0.8413, 0.4950, 0.4413, 0.5183, 0.6088, 0.5400],
  8974. [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5988, 0.5667]],
  8975. device='cuda:0', dtype=torch.float64)
  8976. predictions are: tensor([[ 0.6472, -0.3689, 1.6346, -0.8492, -0.3853, -0.9779, 0.5521, 0.1281],
  8977. [ 0.3331, -0.6111, 1.7036, 0.2330, -0.5683, 0.1859, 0.3486, 0.1685],
  8978. [ 0.6470, -0.3728, 1.4712, -1.0258, -0.3932, -1.1447, 0.4923, 0.0939],
  8979. [ 0.5844, -0.4290, 1.7303, -0.4750, -0.4808, -0.8322, 0.4734, 0.1240],
  8980. [ 0.2903, -0.6353, 1.5557, 0.3560, -0.3045, 0.0172, 0.2414, 0.2126],
  8981. [ 0.2699, -0.6419, 1.6350, 0.2703, -0.4484, 0.1321, 0.2392, 0.1861],
  8982. [ 0.1457, -0.7198, 1.6238, -0.0021, -0.3622, 0.0578, 0.3708, 0.1807],
  8983. [ 0.7875, -0.2717, 1.6199, -0.5207, -0.6258, -0.5832, 0.3459, 0.1732]],
  8984. device='cuda:0', grad_fn=<AddmmBackward>)
  8985. landmarks are: tensor([[[ 6.0837e-01, -4.0762e-01, 1.6806e+00, -9.6182e-01, -2.9977e-01,
  8986. -9.6952e-01, 6.3557e-01, 1.4673e-01],
  8987. [ 5.7939e-01, -4.0231e-01, 1.7788e+00, 6.2048e-02, -4.8453e-01,
  8988. 2.3557e-02, 5.3164e-01, 2.9299e-01],
  8989. [ 6.0785e-01, -3.9761e-01, 1.4208e+00, -1.4314e+00, -3.6328e-01,
  8990. -1.1312e+00, 6.1950e-01, -9.2270e-04],
  8991. [ 6.0918e-01, -3.9130e-01, 1.8423e+00, -5.9230e-01, -4.2679e-01,
  8992. -9.7721e-01, 6.1247e-01, 1.0824e-01],
  8993. [ 5.6195e-01, -4.3457e-01, 1.6691e+00, 3.3149e-01, -2.5935e-01,
  8994. -7.2363e-03, 2.8915e-01, 2.8530e-01],
  8995. [ 5.4821e-01, -3.8414e-01, 1.7326e+00, 1.0054e-01, -3.5173e-01,
  8996. 6.2048e-02, 9.1240e-02, 2.5215e-01],
  8997. [ 5.7812e-01, -4.5219e-01, 1.5998e+00, 4.6189e-04, -2.4781e-01,
  8998. 1.0824e-01, 5.2587e-01, 2.0831e-01],
  8999. [ 5.7829e-01, -3.9330e-01, 1.6748e+00, -6.1540e-01, -5.7691e-01,
  9000. -6.4619e-01, 4.7968e-01, 3.3149e-01]]], device='cuda:0')
  9001. loss_train_step before backward: tensor(0.0203, device='cuda:0', grad_fn=<MseLossBackward>)
  9002.  
  9003. loss_train_step after backward: tensor(0.0203, device='cuda:0', grad_fn=<MseLossBackward>)
  9004. loss_train: 1.2390678264200687
  9005. step: 24
  9006. running loss: 0.051627826100836195
  9007. Train Steps: 24/90 Loss: 0.0516 torch.Size([8, 600, 800])
  9008. torch.Size([8, 8])
  9009. tensor([[0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733],
  9010. [0.6236, 0.4084, 0.7738, 0.2133, 0.3663, 0.3233, 0.5813, 0.5567],
  9011. [0.6128, 0.4022, 0.8738, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064],
  9012. [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329],
  9013. [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892],
  9014. [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413],
  9015. [0.6145, 0.4007, 0.8775, 0.4533, 0.4562, 0.5533, 0.6088, 0.5533],
  9016. [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395]],
  9017. device='cuda:0', dtype=torch.float64)
  9018. predictions are: tensor([[ 0.1949, -0.6706, 1.7265, 0.0499, -0.5140, -0.1403, 0.4085, 0.2467],
  9019. [ 0.7997, -0.3032, 1.3810, -1.0131, -0.5489, -0.8264, 0.3527, 0.1973],
  9020. [ 0.3795, -0.5790, 1.7047, -0.0913, -0.1195, -0.1106, 0.3014, 0.1788],
  9021. [ 0.8255, -0.2797, 1.6323, -0.6465, -0.6656, -0.5635, 0.3617, 0.1430],
  9022. [ 0.2209, -0.6711, 1.5705, -0.1209, -0.6546, -0.3792, 0.2130, 0.2171],
  9023. [ 0.3898, -0.5788, 1.8787, -0.3252, -0.2116, -0.8951, 0.6498, 0.0578],
  9024. [ 0.4747, -0.5116, 1.7857, -0.2169, -0.2496, 0.1502, 0.5376, 0.1775],
  9025. [ 0.3268, -0.6048, 1.6250, 0.1335, -0.5727, -0.5164, 0.2695, 0.1240]],
  9026. device='cuda:0', grad_fn=<AddmmBackward>)
  9027. landmarks are: tensor([[[ 0.6121, -0.3138, 1.8423, 0.0082, -0.6462, -0.3075, 0.3469,
  9028. 0.3623],
  9029. [ 0.5946, -0.3995, 1.2880, -1.3005, -0.5942, -0.7925, 0.3988,
  9030. 0.2853],
  9031. [ 0.5446, -0.4280, 1.7499, 0.0543, 0.0156, 0.1301, 0.1918,
  9032. 0.0532],
  9033. [ 0.5680, -0.4393, 1.5920, -0.6672, -0.6453, -0.5457, 0.5149,
  9034. 0.1753],
  9035. [ 0.5454, -0.4053, 1.6633, -0.1766, -0.6058, -0.5923, 0.1577,
  9036. 0.4357],
  9037. [ 0.6224, -0.4345, 1.9404, -0.2921, -0.3171, -0.8771, 1.0655,
  9038. 0.2142],
  9039. [ 0.5526, -0.4351, 1.7672, -0.1920, -0.1785, 0.2699, 0.5259,
  9040. 0.2699],
  9041. [ 0.5697, -0.4514, 1.7420, 0.2672, -0.6055, -0.6312, 0.3449,
  9042. 0.2058]]], device='cuda:0')
  9043. loss_train_step before backward: tensor(0.0248, device='cuda:0', grad_fn=<MseLossBackward>)
  9044. loss_train_step after backward: tensor(0.0248, device='cuda:0', grad_fn=<MseLossBackward>)
  9045. loss_train: 1.2638748027384281
  9046. step: 25
  9047. running loss: 0.05055499210953712
  9048. Train Steps: 25/90 Loss: 0.0506 torch.Size([8, 600, 800])
  9049. torch.Size([8, 8])
  9050. tensor([[0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5787, 0.5550],
  9051. [ nan, nan, 0.6469, 0.1943, 0.4025, 0.2000, 0.5125, 0.5533],
  9052. [0.6310, 0.4017, 0.8563, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006],
  9053. [0.6200, 0.4059, 0.8700, 0.4900, 0.4163, 0.5000, 0.6162, 0.5467],
  9054. [0.6198, 0.4115, 0.7762, 0.2717, 0.3713, 0.3200, 0.5837, 0.5683],
  9055. [0.6161, 0.4024, 0.8838, 0.4583, 0.3688, 0.3733, 0.5311, 0.5344],
  9056. [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586],
  9057. [0.6261, 0.4066, 0.8325, 0.2150, 0.4763, 0.2667, 0.7002, 0.5633]],
  9058. device='cuda:0', dtype=torch.float64)
  9059. predictions are: tensor([[ 0.7765, -0.3130, 1.6100, -0.5837, -0.6955, -0.3657, 0.2162, 0.2099],
  9060. [-1.2546, -1.6816, 1.0257, -1.0399, -0.3609, -1.2230, 0.1650, 0.1903],
  9061. [ 0.6993, -0.4141, 1.8532, 0.5163, -0.5222, 0.1399, 0.4955, 0.1509],
  9062. [ 0.7386, -0.3365, 1.8392, 0.0971, -0.3573, 0.2887, 0.4597, 0.2088],
  9063. [ 0.6512, -0.3750, 1.5533, -0.6460, -0.5720, -0.5968, 0.2478, 0.2290],
  9064. [ 0.5814, -0.4463, 1.8323, 0.1737, -0.6186, -0.1798, 0.2413, 0.1486],
  9065. [ 0.5449, -0.4473, 1.8061, -0.7797, -0.1826, -0.8444, 0.7036, 0.1541],
  9066. [ 0.8164, -0.3197, 1.7249, -0.7490, -0.1492, -1.0003, 0.7974, 0.0999]],
  9067. device='cuda:0', grad_fn=<AddmmBackward>)
  9068. landmarks are: tensor([[[ 0.5647, -0.4129, 1.4901, -1.0619, -0.6462, -0.5846, 0.3873,
  9069. 0.2776],
  9070. [-2.2859, -2.2859, 0.7023, -1.3883, -0.4268, -1.3621, 0.0813,
  9071. 0.2699],
  9072. [ 0.6286, -0.4303, 1.6691, 0.3931, -0.5827, -0.0919, 0.6359,
  9073. 0.0261],
  9074. [ 0.5776, -0.4109, 1.7326, -0.0226, -0.3633, 0.0236, 0.5605,
  9075. 0.2391],
  9076. [ 0.5768, -0.3852, 1.2995, -1.0311, -0.5711, -0.8079, 0.4104,
  9077. 0.3392],
  9078. [ 0.5598, -0.4273, 1.7961, -0.1689, -0.5827, -0.5615, 0.1671,
  9079. 0.1824],
  9080. [ 0.6075, -0.4129, 1.6055, -1.0080, -0.2420, -1.0080, 0.9704,
  9081. 0.2944],
  9082. [ 0.6059, -0.4080, 1.5594, -1.2928, -0.0861, -1.0542, 0.9485,
  9083. 0.3157]]], device='cuda:0')
  9084. loss_train_step before backward: tensor(0.0563, device='cuda:0', grad_fn=<MseLossBackward>)
  9085. loss_train_step after backward: tensor(0.0563, device='cuda:0', grad_fn=<MseLossBackward>)
  9086. loss_train: 1.3201934099197388
  9087. step: 26
  9088. running loss: 0.05077666961229765
  9089. Train Steps: 26/90 Loss: 0.0508 torch.Size([8, 600, 800])
  9090. torch.Size([8, 8])
  9091. tensor([[0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600],
  9092. [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5787, 0.5117],
  9093. [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650],
  9094. [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092],
  9095. [0.6256, 0.4199, 0.8638, 0.5800, 0.3987, 0.4383, 0.5600, 0.5950],
  9096. [0.6136, 0.4117, 0.8700, 0.5167, 0.4188, 0.5083, 0.5147, 0.5495],
  9097. [0.6273, 0.4100, 0.7137, 0.2133, 0.4000, 0.2650, 0.6075, 0.5633],
  9098. [0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494]],
  9099. device='cuda:0', dtype=torch.float64)
  9100. predictions are: tensor([[ 0.4843, -0.4923, 1.8988, -0.3992, -0.2177, 0.0468, 0.5391, 0.2222],
  9101. [ 0.4154, -0.5881, 1.2895, -1.1669, -0.4763, -1.3328, 0.2543, 0.1306],
  9102. [ 0.4518, -0.5199, 1.7362, 0.0782, -0.4067, -0.2372, 0.3323, 0.2551],
  9103. [ 0.5852, -0.4918, 1.9120, -0.1694, -0.4795, 0.0024, 0.5392, 0.1263],
  9104. [ 0.6552, -0.4101, 1.8002, 0.1895, -0.3935, -0.2230, 0.4755, 0.2813],
  9105. [ 0.5180, -0.4848, 1.8587, 0.0052, -0.4058, 0.0462, 0.3925, 0.1988],
  9106. [ 0.1950, -0.6670, 1.2533, -1.3154, -0.4737, -1.2690, 0.4056, 0.1900],
  9107. [ 0.2802, -0.6517, 1.8728, -0.1009, -0.4746, -0.4945, 0.5519, 0.1651]],
  9108. device='cuda:0', grad_fn=<AddmmBackward>)
  9109. landmarks are: tensor([[[ 0.5652, -0.4325, 1.7388, -0.1429, -0.0688, 0.3469, 0.4970,
  9110. 0.3007],
  9111. [ 0.5598, -0.4201, 1.1898, -1.3005, -0.3806, -1.3313, 0.3873,
  9112. 0.0774],
  9113. [ 0.5770, -0.4036, 1.7095, 0.3084, -0.3691, -0.2690, 0.2314,
  9114. 0.3238],
  9115. [ 0.5783, -0.4363, 1.7724, 0.0432, -0.5153, 0.0871, 0.4840,
  9116. 0.0663],
  9117. [ 0.6038, -0.3464, 1.7037, 0.3931, -0.4441, -0.2613, 0.3007,
  9118. 0.4624],
  9119. [ 0.5482, -0.3841, 1.7326, 0.1005, -0.3517, 0.0620, 0.0912,
  9120. 0.2522],
  9121. [ 0.6115, -0.3924, 1.0109, -1.3005, -0.4383, -1.0619, 0.5201,
  9122. 0.3161],
  9123. [ 0.6224, -0.4032, 1.5940, 0.2930, -0.5769, -0.2690, 0.8900,
  9124. 0.2516]]], device='cuda:0')
  9125. loss_train_step before backward: tensor(0.0273, device='cuda:0', grad_fn=<MseLossBackward>)
  9126. loss_train_step after backward: tensor(0.0273, device='cuda:0', grad_fn=<MseLossBackward>)
  9127. loss_train: 1.3474591448903084
  9128. step: 27
  9129. running loss: 0.049905894255196606
  9130. Train Steps: 27/90 Loss: 0.0499 torch.Size([8, 600, 800])
  9131. torch.Size([8, 8])
  9132. tensor([[0.6145, 0.4007, 0.8775, 0.4533, 0.4562, 0.5533, 0.6088, 0.5533],
  9133. [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267],
  9134. [0.6161, 0.4024, 0.8838, 0.4583, 0.3688, 0.3733, 0.5311, 0.5344],
  9135. [0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274],
  9136. [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852],
  9137. [0.6162, 0.4134, 0.6700, 0.2467, 0.3962, 0.2533, 0.5737, 0.5467],
  9138. [0.6099, 0.4030, 0.8638, 0.5117, 0.4983, 0.4965, 0.5086, 0.5388],
  9139. [0.6224, 0.4179, 0.8700, 0.5683, 0.4037, 0.4683, 0.5650, 0.5600]],
  9140. device='cuda:0', dtype=torch.float64)
  9141. predictions are: tensor([[ 0.4047, -0.5641, 1.8557, -0.3915, -0.3121, 0.0780, 0.5668, 0.2369],
  9142. [ 0.5390, -0.4681, 1.7114, -0.0121, -0.5081, -0.2583, 0.4403, 0.2166],
  9143. [ 0.3382, -0.6112, 1.8243, -0.2912, -0.6956, -0.5718, 0.3748, 0.1359],
  9144. [ 0.1285, -0.7148, 1.7671, -0.9421, -0.1662, -1.1593, 0.6499, 0.1509],
  9145. [ 0.5752, -0.4616, 1.7676, -0.2336, -0.2807, -0.1014, 0.4329, 0.2863],
  9146. [ 0.3819, -0.5542, 1.1212, -1.1021, -0.6073, -1.2107, 0.3160, 0.2424],
  9147. [ 0.5454, -0.4809, 1.7989, -0.1910, -0.1672, -0.2807, 0.4104, 0.2157],
  9148. [ 0.6235, -0.4284, 1.7667, 0.0837, -0.4849, -0.0909, 0.5290, 0.3062]],
  9149. device='cuda:0', grad_fn=<AddmmBackward>)
  9150. landmarks are: tensor([[[ 0.5526, -0.4351, 1.7672, -0.1920, -0.1785, 0.2699, 0.5259,
  9151. 0.2699],
  9152. [ 0.5891, -0.4550, 1.5132, 0.3546, -0.3691, -0.1535, 0.3815,
  9153. 0.1467],
  9154. [ 0.5598, -0.4273, 1.7961, -0.1689, -0.5827, -0.5615, 0.1671,
  9155. 0.1824],
  9156. [ 0.6537, -0.3719, 1.6979, -0.8617, -0.0169, -1.3621, 0.6926,
  9157. 0.1501],
  9158. [ 0.5500, -0.4060, 1.7326, 0.0236, -0.1554, -0.0226, 0.1438,
  9159. 0.4171],
  9160. [ 0.5603, -0.3764, 0.8088, -1.1466, -0.4557, -1.1158, 0.3642,
  9161. 0.2391],
  9162. [ 0.5309, -0.4246, 1.7037, 0.0774, 0.0158, 0.0075, 0.0635,
  9163. 0.2026],
  9164. [ 0.5891, -0.3557, 1.7326, 0.3392, -0.4210, -0.1227, 0.3238,
  9165. 0.3007]]], device='cuda:0')
  9166. loss_train_step before backward: tensor(0.0284, device='cuda:0', grad_fn=<MseLossBackward>)
  9167. loss_train_step after backward: tensor(0.0284, device='cuda:0', grad_fn=<MseLossBackward>)
  9168. loss_train: 1.3758654911071062
  9169. step: 28
  9170. running loss: 0.04913805325382522
  9171.  
  9172. Train Steps: 28/90 Loss: 0.0491 torch.Size([8, 600, 800])
  9173. torch.Size([8, 8])
  9174. tensor([[0.6124, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485],
  9175. [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767],
  9176. [0.6200, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183],
  9177. [ nan, nan, 0.7240, 0.2722, 0.3900, 0.2567, 0.5168, 0.5933],
  9178. [0.6171, 0.4127, 0.8900, 0.4800, 0.4325, 0.5783, 0.5769, 0.5090],
  9179. [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611],
  9180. [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283],
  9181. [0.6293, 0.4097, 0.8800, 0.2517, 0.5262, 0.2600, 0.7430, 0.5378]],
  9182. device='cuda:0', dtype=torch.float64)
  9183. predictions are: tensor([[ 0.4980, -0.4510, 1.1881, -0.9858, -0.5650, -0.8815, 0.2687, 0.2701],
  9184. [ 0.6081, -0.3681, 1.7448, -0.0104, -0.5931, -0.5104, 0.2804, 0.2786],
  9185. [ 0.8465, -0.2756, 1.5485, -0.9244, -0.5099, -0.8537, 0.6211, 0.1732],
  9186. [-1.4354, -1.7735, 1.1950, -0.9369, -0.3766, -1.0535, 0.1421, 0.2662],
  9187. [ 0.8859, -0.2152, 1.9825, 0.0037, -0.3014, 0.4025, 0.5873, 0.2754],
  9188. [ 0.9285, -0.2262, 1.8356, -0.0805, -0.3351, 0.0737, 0.4443, 0.3068],
  9189. [ 0.5997, -0.4376, 1.7979, 0.2760, -0.4896, -0.1813, 0.4074, 0.2335],
  9190. [ 0.5675, -0.4697, 1.8163, -1.0207, 0.0638, -1.0853, 1.0176, 0.1437]],
  9191. device='cuda:0', grad_fn=<AddmmBackward>)
  9192. landmarks are: tensor([[[ 0.5425, -0.3998, 0.9263, -0.8683, -0.6000, -1.0157, 0.0990,
  9193. 0.2476],
  9194. [ 0.5781, -0.3848, 1.7441, -0.0996, -0.5769, -0.8002, 0.2021,
  9195. 0.3778],
  9196. [ 0.5781, -0.4129, 1.4142, -0.9057, -0.5115, -0.9937, 0.4621,
  9197. 0.1080],
  9198. [-2.2859, -2.2859, 1.0580, -1.0288, -0.4845, -1.1004, 0.1011,
  9199. 0.4543],
  9200. [ 0.5645, -0.3797, 1.8249, -0.0688, -0.2882, 0.3854, 0.3789,
  9201. 0.0652],
  9202. [ 0.5491, -0.4290, 1.7788, -0.1073, -0.2651, -0.1073, 0.2555,
  9203. 0.3057],
  9204. [ 0.5900, -0.4377, 1.6113, 0.3623, -0.5018, -0.3229, 0.3700,
  9205. 0.1544],
  9206. [ 0.6207, -0.3936, 1.7788, -1.1235, 0.1448, -1.0850, 1.1459,
  9207. 0.1982]]], device='cuda:0')
  9208. loss_train_step before backward: tensor(0.0327, device='cuda:0', grad_fn=<MseLossBackward>)
  9209. loss_train_step after backward: tensor(0.0327, device='cuda:0', grad_fn=<MseLossBackward>)
  9210. loss_train: 1.4085994269698858
  9211. step: 29
  9212. running loss: 0.048572394033444335
  9213. Train Steps: 29/90 Loss: 0.0486 torch.Size([8, 600, 800])
  9214. torch.Size([8, 8])
  9215. tensor([[0.6176, 0.3911, 0.8738, 0.4217, 0.3488, 0.4033, 0.6025, 0.4817],
  9216. [0.6185, 0.4098, 0.8838, 0.4900, 0.4537, 0.5800, 0.6288, 0.5400],
  9217. [0.6147, 0.4026, 0.6600, 0.2467, 0.4088, 0.2150, 0.5489, 0.5773],
  9218. [0.6102, 0.3999, 0.8750, 0.5133, 0.3825, 0.4750, 0.5637, 0.5083],
  9219. [0.6189, 0.4049, 0.8888, 0.4417, 0.4213, 0.5200, 0.5988, 0.5633],
  9220. [0.6179, 0.4082, 0.6688, 0.2667, 0.3588, 0.3317, 0.5750, 0.5783],
  9221. [0.6137, 0.4038, 0.8563, 0.4050, 0.3813, 0.2550, 0.5106, 0.4954],
  9222. [0.6185, 0.4067, 0.8838, 0.4450, 0.4037, 0.4733, 0.5213, 0.5142]],
  9223. device='cuda:0', dtype=torch.float64)
  9224. predictions are: tensor([[ 0.8314, -0.2732, 1.7540, -0.4890, -0.5828, -0.5468, 0.6514, 0.1730],
  9225. [ 0.6223, -0.3878, 1.9407, -0.0566, -0.0362, 0.2270, 0.7201, 0.3392],
  9226. [-0.0648, -0.8505, 0.9972, -1.0496, -0.3593, -1.3296, 0.2686, 0.3011],
  9227. [ 0.6182, -0.4060, 1.8672, 0.1088, -0.4125, -0.1731, 0.4656, 0.2312],
  9228. [ 0.7090, -0.3358, 1.9056, -0.3081, -0.2719, 0.1284, 0.6481, 0.3014],
  9229. [ 0.5278, -0.4470, 1.1147, -0.9866, -0.5895, -0.8632, 0.3399, 0.3552],
  9230. [-0.2827, -0.9884, 1.6530, -0.4936, -0.3237, -1.1280, 0.2926, 0.1953],
  9231. [ 0.4771, -0.4732, 1.8016, -0.2574, -0.3934, -0.2234, 0.3285, 0.2492]],
  9232. device='cuda:0', grad_fn=<AddmmBackward>)
  9233. landmarks are: tensor([[[ 0.5669, -0.4794, 1.7499, -0.3383, -0.6751, -0.4229, 0.4970,
  9234. -0.0611],
  9235. [ 0.5709, -0.3933, 1.7961, -0.0226, -0.1901, 0.3931, 0.6182,
  9236. 0.2083],
  9237. [ 0.5532, -0.4264, 0.7626, -1.1466, -0.3979, -1.2928, 0.2494,
  9238. 0.3808],
  9239. [ 0.5326, -0.4390, 1.7557, 0.0851, -0.5192, -0.0919, 0.3180,
  9240. 0.0620],
  9241. [ 0.5726, -0.4159, 1.8192, -0.2459, -0.3402, 0.1159, 0.4797,
  9242. 0.3161],
  9243. [ 0.5680, -0.4005, 0.8030, -1.0542, -0.6289, -0.7540, 0.3700,
  9244. 0.3854],
  9245. [ 0.5485, -0.4209, 1.6691, -0.4152, -0.5249, -1.1081, 0.0725,
  9246. 0.0021],
  9247. [ 0.5708, -0.4075, 1.7961, -0.2305, -0.4210, -0.0996, 0.1219,
  9248. 0.0893]]], device='cuda:0')
  9249. loss_train_step before backward: tensor(0.0391, device='cuda:0', grad_fn=<MseLossBackward>)
  9250. loss_train_step after backward: tensor(0.0391, device='cuda:0', grad_fn=<MseLossBackward>)
  9251. loss_train: 1.447685481980443
  9252. step: 30
  9253. running loss: 0.04825618273268143
  9254. Train Steps: 30/90 Loss: 0.0483 torch.Size([8, 600, 800])
  9255. torch.Size([8, 8])
  9256. tensor([[0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343],
  9257. [0.6179, 0.4008, 0.8600, 0.4015, 0.3932, 0.2515, 0.5711, 0.5438],
  9258. [0.6134, 0.4090, 0.6926, 0.2819, 0.3538, 0.3233, 0.5563, 0.5667],
  9259. [0.6200, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148],
  9260. [0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5787, 0.5600],
  9261. [0.6307, 0.4029, 0.8650, 0.5200, 0.3763, 0.4017, 0.7311, 0.5366],
  9262. [0.6346, 0.4086, 0.7938, 0.5500, 0.3962, 0.4867, 0.7343, 0.5702],
  9263. [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301]],
  9264. device='cuda:0', dtype=torch.float64)
  9265. predictions are: tensor([[ 0.4710, -0.5241, 1.6559, -0.7972, -0.2771, -0.8978, 0.7490, 0.1722],
  9266. [ 0.2822, -0.5709, 1.4792, -0.7684, -0.3218, -1.0572, 0.2195, 0.2607],
  9267. [ 0.3433, -0.5362, 1.0719, -1.1185, -0.5896, -0.8732, 0.2522, 0.3263],
  9268. [ 0.3159, -0.5916, 1.7861, -0.2359, -0.5078, -0.4310, 0.3561, 0.2378],
  9269. [ 0.4398, -0.4827, 1.7726, -0.2350, -0.0829, 0.0155, 0.4125, 0.3373],
  9270. [ 0.3263, -0.5749, 1.7963, -0.0346, -0.3689, -0.5563, 0.4905, 0.1943],
  9271. [ 0.6169, -0.3935, 1.6085, -0.0899, -0.3845, -0.1183, 0.6086, 0.3813],
  9272. [ 0.6808, -0.3486, 1.7518, -0.1970, -0.3165, -0.0122, 0.5884, 0.2809]],
  9273. device='cuda:0', grad_fn=<AddmmBackward>)
  9274. landmarks are: tensor([[[ 0.6125, -0.4369, 1.9173, -0.5384, -0.2594, -0.8386, 0.9741,
  9275. 0.1821],
  9276. [ 0.5680, -0.4345, 1.6864, -0.4315, -0.4698, -1.1241, 0.3518,
  9277. 0.2261],
  9278. [ 0.5473, -0.3966, 0.9131, -0.9838, -0.6520, -0.7925, 0.2834,
  9279. 0.3315],
  9280. [ 0.5781, -0.4205, 1.8159, -0.0691, -0.6116, -0.3069, 0.4236,
  9281. 0.0919],
  9282. [ 0.5505, -0.4207, 1.7095, -0.0534, -0.0509, 0.1050, 0.3873,
  9283. 0.3007],
  9284. [ 0.6273, -0.4249, 1.7095, 0.1159, -0.5480, -0.4306, 1.0910,
  9285. 0.1928],
  9286. [ 0.6454, -0.3984, 1.3804, 0.2545, -0.4557, -0.0380, 1.1057,
  9287. 0.3478],
  9288. [ 0.6010, -0.4562, 1.7198, -0.0090, -0.3464, 0.0108, 0.6294,
  9289. 0.1627]]], device='cuda:0')
  9290. loss_train_step before backward: tensor(0.0302, device='cuda:0', grad_fn=<MseLossBackward>)
  9291. loss_train_step after backward: tensor(0.0302, device='cuda:0', grad_fn=<MseLossBackward>)
  9292. loss_train: 1.477879423648119
  9293. step: 31
  9294. running loss: 0.04767352979510061
  9295. Train Steps: 31/90 Loss: 0.0477 torch.Size([8, 600, 800])
  9296. torch.Size([8, 8])
  9297. tensor([[0.6179, 0.4082, 0.6688, 0.2667, 0.3588, 0.3317, 0.5750, 0.5783],
  9298. [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274],
  9299. [0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250],
  9300. [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960],
  9301. [0.6179, 0.4040, 0.7412, 0.1850, 0.3825, 0.2783, 0.5837, 0.5600],
  9302. [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082],
  9303. [0.6078, 0.4033, 0.8019, 0.3055, 0.3450, 0.4200, 0.6025, 0.5550],
  9304. [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413]],
  9305. device='cuda:0', dtype=torch.float64)
  9306. predictions are: tensor([[ 0.5117, -0.4196, 0.9786, -0.9416, -0.6249, -0.8022, 0.2050, 0.3726],
  9307. [ 0.5619, -0.3987, 1.7834, 0.0175, -0.4419, -0.3582, 0.4923, 0.2641],
  9308. [ 0.3236, -0.5390, 1.8023, -0.3149, -0.4430, -0.8161, 0.4877, 0.2087],
  9309. [ 0.4976, -0.4769, 1.7585, 0.0139, -0.1965, 0.0074, 0.3728, 0.2565],
  9310. [ 0.4925, -0.4334, 1.1824, -1.0599, -0.4519, -1.0876, 0.3643, 0.2796],
  9311. [ 0.7676, -0.2882, 1.7267, 0.1517, -0.2220, 0.2903, 0.4297, 0.2950],
  9312. [ 0.3197, -0.5553, 1.3923, -0.9086, -0.5696, -0.3927, 0.5116, 0.3125],
  9313. [-0.0600, -0.8208, 1.8714, -0.7073, 0.0863, -0.9754, 0.9008, 0.1728]],
  9314. device='cuda:0', grad_fn=<AddmmBackward>)
  9315. landmarks are: tensor([[[ 0.5680, -0.4005, 0.8030, -1.0542, -0.6289, -0.7540, 0.3700,
  9316. 0.3854],
  9317. [ 0.6108, -0.4008, 1.9088, -0.0253, -0.5769, -0.3075, 0.7905,
  9318. 0.1499],
  9319. [ 0.6111, -0.3828, 1.8885, -0.3844, -0.5654, -0.8079, 0.5663,
  9320. 0.1390],
  9321. [ 0.5205, -0.4577, 1.7326, -0.1304, -0.1862, 0.0447, 0.1301,
  9322. 0.0051],
  9323. [ 0.5680, -0.4201, 1.1379, -1.4314, -0.5192, -1.0003, 0.4104,
  9324. 0.3007],
  9325. [ 0.5450, -0.4706, 1.7643, 0.0722, -0.3708, 0.3200, 0.4854,
  9326. 0.0612],
  9327. [ 0.5215, -0.4232, 1.4182, -0.8747, -0.6924, -0.3460, 0.4970,
  9328. 0.2776],
  9329. [ 0.6108, -0.4201, 1.8711, -0.7848, -0.0053, -1.2236, 1.0362,
  9330. 0.2142]]], device='cuda:0')
  9331. loss_train_step before backward: tensor(0.0254, device='cuda:0', grad_fn=<MseLossBackward>)
  9332. loss_train_step after backward: tensor(0.0254, device='cuda:0', grad_fn=<MseLossBackward>)
  9333. loss_train: 1.5033203139901161
  9334. step: 32
  9335. running loss: 0.04697875981219113
  9336.  
  9337. Train Steps: 32/90 Loss: 0.0470 torch.Size([8, 600, 800])
  9338. torch.Size([8, 8])
  9339. tensor([[0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212],
  9340. [0.6276, 0.4120, 0.8738, 0.3133, 0.4225, 0.2217, 0.6203, 0.4892],
  9341. [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317],
  9342. [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446],
  9343. [ nan, nan, 0.8363, 0.3317, 0.3563, 0.3367, 0.5329, 0.5142],
  9344. [0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153],
  9345. [0.6126, 0.4039, 0.8237, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138],
  9346. [0.6289, 0.4019, 0.8113, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332]],
  9347. device='cuda:0', dtype=torch.float64)
  9348. predictions are: tensor([[ 0.6046, -0.4041, 1.3856, -1.1072, -0.3421, -1.0320, 0.7189, 0.1829],
  9349. [ 0.3036, -0.5672, 1.7304, -0.7838, -0.2106, -1.1523, 0.6788, 0.1557],
  9350. [ 0.8693, -0.1994, 1.6628, 0.1078, -0.4714, -0.5727, 0.3917, 0.2545],
  9351. [ 0.4444, -0.4605, 1.5921, -0.3418, -0.3893, -0.0181, 0.3984, 0.3219],
  9352. [-0.8497, -1.3179, 1.3370, -0.8653, -0.4921, -0.9819, 0.1373, 0.2156],
  9353. [ 0.7128, -0.3160, 1.6909, 0.0323, -0.1249, 0.1409, 0.4211, 0.3112],
  9354. [ 0.8694, -0.2089, 1.4843, -0.6585, -0.5412, -0.5137, 0.5688, 0.3325],
  9355. [ 0.6722, -0.3150, 1.5615, -0.0606, -0.3911, -0.0069, 0.5004, 0.3152]],
  9356. device='cuda:0', grad_fn=<AddmmBackward>)
  9357. landmarks are: tensor([[[ 0.6125, -0.4345, 1.4308, -1.1384, -0.4213, -1.0031, 0.7190,
  9358. 0.1214],
  9359. [ 0.6128, -0.3828, 1.7499, -0.8386, -0.3344, -1.2620, 0.5792,
  9360. -0.0263],
  9361. [ 0.5901, -0.4157, 1.7557, 0.1929, -0.5423, -0.5923, 0.3584,
  9362. 0.1698],
  9363. [ 0.5444, -0.3846, 1.6171, -0.1689, -0.5885, -0.0380, 0.1791,
  9364. 0.2296],
  9365. [-2.2859, -2.2859, 1.5767, -0.7540, -0.6404, -0.7309, 0.1753,
  9366. 0.0893],
  9367. [ 0.5548, -0.4682, 1.7309, 0.0966, -0.1394, 0.1757, 0.5045,
  9368. 0.0942],
  9369. [ 0.5438, -0.4206, 1.5189, -0.4537, -0.6115, -0.6231, 0.4365,
  9370. 0.5491],
  9371. [ 0.6191, -0.4297, 1.4612, 0.2391, -0.4961, 0.0313, 1.1166,
  9372. 0.1768]]], device='cuda:0')
  9373. loss_train_step before backward: tensor(0.0719, device='cuda:0', grad_fn=<MseLossBackward>)
  9374. loss_train_step after backward: tensor(0.0719, device='cuda:0', grad_fn=<MseLossBackward>)
  9375. loss_train: 1.575185589492321
  9376. step: 33
  9377. running loss: 0.047732896651282455
  9378. Train Steps: 33/90 Loss: 0.0477 torch.Size([8, 600, 800])
  9379. torch.Size([8, 8])
  9380. tensor([[0.6228, 0.4004, 0.8750, 0.5250, 0.3825, 0.5233, 0.6362, 0.5000],
  9381. [0.6068, 0.3963, 0.8650, 0.4317, 0.4037, 0.5083, 0.5253, 0.4999],
  9382. [0.6299, 0.4008, 0.8450, 0.5350, 0.4213, 0.5000, 0.6350, 0.5100],
  9383. [ nan, nan, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633],
  9384. [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650],
  9385. [0.6222, 0.4169, 0.8638, 0.5650, 0.4313, 0.4783, 0.5637, 0.5633],
  9386. [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383],
  9387. [0.6270, 0.4267, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183]],
  9388. device='cuda:0', dtype=torch.float64)
  9389. predictions are: tensor([[ 0.9045, -0.2045, 1.7665, -0.0190, -0.4189, -0.0906, 0.6821, 0.1926],
  9390. [ 0.6531, -0.3369, 1.7089, -0.5142, -0.4372, -0.1780, 0.5593, 0.1922],
  9391. [ 0.9902, -0.1364, 1.6734, -0.0408, -0.3670, -0.1510, 0.6068, 0.2448],
  9392. [-1.6674, -1.8711, 1.1930, -1.0522, -0.3653, -1.3092, 0.2291, 0.2265],
  9393. [ 0.6062, -0.3637, 1.6419, -0.6272, -0.6213, -0.5816, 0.3713, 0.2625],
  9394. [ 1.0549, -0.0778, 1.6522, 0.2349, -0.3027, -0.0812, 0.4714, 0.3484],
  9395. [ 0.8803, -0.1778, 1.6896, -0.1564, -0.2772, -0.2500, 0.4931, 0.2272],
  9396. [ 0.2314, -0.5895, 0.9697, -1.2898, -0.2493, -1.2955, 0.3786, 0.3206]],
  9397. device='cuda:0', grad_fn=<AddmmBackward>)
  9398. landmarks are: tensor([[[ 5.9082e-01, -4.3664e-01, 1.7557e+00, 1.3903e-01, -5.1917e-01,
  9399. 1.3133e-01, 6.5289e-01, 2.3557e-02],
  9400. [ 5.1680e-01, -4.5558e-01, 1.7095e+00, -2.9207e-01, -4.2102e-01,
  9401. 6.2048e-02, 1.4038e-01, 2.3124e-02],
  9402. [ 6.2361e-01, -4.3441e-01, 1.6171e+00, 1.8522e-01, -3.4018e-01,
  9403. 2.3557e-02, 6.4711e-01, 6.9746e-02],
  9404. [-2.2859e+00, -2.2859e+00, 1.2030e+00, -1.0288e+00, -4.9607e-01,
  9405. -1.1081e+00, 8.1293e-02, 3.1609e-01],
  9406. [ 5.7569e-01, -3.9169e-01, 1.7095e+00, -4.7683e-01, -6.3464e-01,
  9407. -4.2294e-01, 3.9307e-01, 3.2379e-01],
  9408. [ 5.8799e-01, -3.6051e-01, 1.7037e+00, 3.2379e-01, -2.9400e-01,
  9409. -7.6520e-02, 3.1801e-01, 3.1609e-01],
  9410. [ 5.3603e-01, -4.6490e-01, 1.6517e+00, 4.6189e-04, -2.8245e-01,
  9411. -6.8822e-02, 2.3086e-01, 2.0046e-01],
  9412. [ 6.1010e-01, -3.1524e-01, 1.0166e+00, -7.5396e-01, -2.2633e-02,
  9413. -1.4468e+00, 2.8226e-01, 5.7018e-01]]], device='cuda:0')
  9414. loss_train_step before backward: tensor(0.0464, device='cuda:0', grad_fn=<MseLossBackward>)
  9415. loss_train_step after backward: tensor(0.0464, device='cuda:0', grad_fn=<MseLossBackward>)
  9416. loss_train: 1.6215668469667435
  9417. step: 34
  9418. running loss: 0.047693142557845396
  9419. Train Steps: 34/90 Loss: 0.0477 torch.Size([8, 600, 800])
  9420. torch.Size([8, 8])
  9421. tensor([[0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183],
  9422. [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611],
  9423. [ nan, nan, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999],
  9424. [0.6126, 0.4067, 0.8638, 0.5383, 0.4188, 0.4850, 0.5016, 0.5392],
  9425. [0.6179, 0.4082, 0.6688, 0.2667, 0.3588, 0.3317, 0.5750, 0.5783],
  9426. [0.6212, 0.4033, 0.8938, 0.4167, 0.3813, 0.4267, 0.5613, 0.5583],
  9427. [0.6336, 0.4154, 0.8900, 0.2767, 0.4988, 0.2867, 0.7422, 0.5540],
  9428. [ nan, nan, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552]],
  9429. device='cuda:0', dtype=torch.float64)
  9430. predictions are: tensor([[ 0.8166, -0.1899, 1.3306, -0.7404, -0.4580, -0.7054, 0.4630, 0.3116],
  9431. [ 1.0022, -0.1160, 1.6696, 0.1123, -0.4108, 0.1231, 0.3683, 0.2487],
  9432. [-1.8722, -1.9516, 1.0990, -0.9600, -0.3088, -1.2137, 0.1891, 0.1847],
  9433. [ 1.2242, 0.0453, 1.6560, 0.5276, -0.4548, 0.1647, 0.3594, 0.2172],
  9434. [ 0.9550, -0.1220, 0.9757, -0.8584, -0.7268, -0.6853, 0.3104, 0.3065],
  9435. [ 0.8092, -0.2271, 1.7159, -0.2274, -0.6485, -0.1693, 0.3869, 0.2246],
  9436. [ 0.9884, -0.1324, 1.6578, -0.7221, -0.1089, -0.8412, 0.9937, 0.1319],
  9437. [-0.7703, -1.1830, 1.6248, -0.8926, 0.0598, -0.9897, 0.8520, 0.1865]],
  9438. device='cuda:0', grad_fn=<AddmmBackward>)
  9439. landmarks are: tensor([[[ 0.5759, -0.3948, 1.2072, -0.9849, -0.4672, -0.8156, 0.4912,
  9440. 0.5701],
  9441. [ 0.5491, -0.4290, 1.7788, -0.1073, -0.2651, -0.1073, 0.2555,
  9442. 0.3057],
  9443. [-2.2859, -2.2859, 1.3400, -1.0388, -0.3055, -1.4930, 0.1157,
  9444. 0.0231],
  9445. [ 0.5436, -0.4076, 1.7037, 0.2006, -0.3517, -0.0457, 0.0311,
  9446. 0.2048],
  9447. [ 0.5680, -0.4005, 0.8030, -1.0542, -0.6289, -0.7540, 0.3700,
  9448. 0.3854],
  9449. [ 0.5832, -0.4231, 1.8423, -0.3614, -0.5249, -0.3152, 0.3065,
  9450. 0.2930],
  9451. [ 0.6405, -0.3671, 1.8249, -1.0080, 0.0178, -0.9618, 1.1422,
  9452. 0.2730],
  9453. [-2.2859, -2.2859, 1.7557, -1.1466, 0.0871, -1.0773, 1.1239,
  9454. 0.2783]]], device='cuda:0')
  9455. loss_train_step before backward: tensor(0.1038, device='cuda:0', grad_fn=<MseLossBackward>)
  9456. loss_train_step after backward: tensor(0.1038, device='cuda:0', grad_fn=<MseLossBackward>)
  9457. loss_train: 1.725327029824257
  9458. step: 35
  9459. running loss: 0.04929505799497877
  9460. Train Steps: 35/90 Loss: 0.0493 torch.Size([8, 600, 800])
  9461. torch.Size([8, 8])
  9462. tensor([[0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650],
  9463. [0.6173, 0.4114, 0.7325, 0.2500, 0.4213, 0.1917, 0.5338, 0.5700],
  9464. [0.6097, 0.3988, 0.8650, 0.5250, 0.4213, 0.5200, 0.5675, 0.5050],
  9465. [0.6192, 0.4128, 0.8513, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633],
  9466. [0.6068, 0.3963, 0.8650, 0.4317, 0.4037, 0.5083, 0.5253, 0.4999],
  9467. [0.6182, 0.3930, 0.8841, 0.3892, 0.3556, 0.4967, 0.6222, 0.5279],
  9468. [ nan, nan, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633],
  9469. [0.6127, 0.4115, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495]],
  9470. device='cuda:0', dtype=torch.float64)
  9471. predictions are: tensor([[ 1.0376, -0.0549, 1.6175, 0.2950, -0.3642, -0.3018, 0.4173, 0.2902],
  9472. [-1.3129, -1.5726, 1.0380, -1.1441, -0.2995, -1.3562, 0.2976, 0.2221],
  9473. [ 1.0837, -0.0560, 1.7402, 0.1596, -0.3138, 0.0089, 0.4249, 0.1761],
  9474. [ 1.2470, 0.0356, 1.7279, 0.3775, -0.3578, 0.1207, 0.5581, 0.2648],
  9475. [ 0.7240, -0.2872, 1.7171, -0.3920, -0.4311, -0.0944, 0.5612, 0.1505],
  9476. [ 0.7184, -0.3035, 1.6816, -0.6363, -0.5410, -0.1442, 0.8086, 0.2079],
  9477. [-1.6307, -1.8106, 1.2022, -0.9604, -0.3940, -1.2524, 0.2497, 0.2113],
  9478. [ 0.9433, -0.1526, 1.0979, -1.1195, -0.5620, -0.9940, 0.4218, 0.2257]],
  9479. device='cuda:0', grad_fn=<AddmmBackward>)
  9480. landmarks are: tensor([[[ 0.5770, -0.4036, 1.7095, 0.3084, -0.3691, -0.2690, 0.2314,
  9481. 0.3238],
  9482. [ 0.5652, -0.3858, 1.0975, -1.1312, -0.3402, -1.4006, 0.1794,
  9483. 0.3469],
  9484. [ 0.5303, -0.4440, 1.7095, 0.1390, -0.3402, 0.1159, 0.3353,
  9485. 0.0467],
  9486. [ 0.5742, -0.3792, 1.6460, 0.3084, -0.3460, 0.1467, 0.4162,
  9487. 0.3161],
  9488. [ 0.5168, -0.4556, 1.7095, -0.2921, -0.4210, 0.0620, 0.1404,
  9489. 0.0231],
  9490. [ 0.5697, -0.4706, 1.7976, -0.4884, -0.6433, 0.0081, 0.5878,
  9491. 0.1525],
  9492. [-2.2859, -2.2859, 1.2030, -1.0288, -0.4961, -1.1081, 0.0813,
  9493. 0.3161],
  9494. [ 0.5442, -0.3855, 1.0224, -0.9541, -0.6115, -0.9233, 0.1745,
  9495. 0.2522]]], device='cuda:0')
  9496. loss_train_step before backward: tensor(0.1254, device='cuda:0', grad_fn=<MseLossBackward>)
  9497. loss_train_step after backward: tensor(0.1254, device='cuda:0', grad_fn=<MseLossBackward>)
  9498. loss_train: 1.8506858050823212
  9499. step: 36
  9500. running loss: 0.05140793903006448
  9501.  
  9502. Train Steps: 36/90 Loss: 0.0514 torch.Size([8, 600, 800])
  9503. torch.Size([8, 8])
  9504. tensor([[0.6364, 0.4165, 0.9088, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297],
  9505. [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334],
  9506. [0.6126, 0.3954, 0.8538, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350],
  9507. [0.6097, 0.4024, 0.8488, 0.3717, 0.3875, 0.5517, 0.5836, 0.5591],
  9508. [ nan, nan, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609],
  9509. [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012],
  9510. [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656],
  9511. [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679]],
  9512. device='cuda:0', dtype=torch.float64)
  9513. predictions are: tensor([[ 0.3210, -0.4926, 1.6567, -0.4000, -0.3599, -1.0264, 0.4894, 0.1499],
  9514. [ 0.6571, -0.3596, 1.0842, -1.2060, -0.6313, -1.0385, 0.5151, 0.1963],
  9515. [ 0.5058, -0.4447, 1.5437, 0.0189, -0.4176, -0.1923, 0.2316, 0.1914],
  9516. [ 0.1817, -0.6537, 1.4349, -0.7227, -0.4305, 0.0920, 0.4942, 0.2824],
  9517. [-1.3286, -1.6251, 1.6110, -0.9346, 0.0318, -1.2953, 0.6881, 0.1814],
  9518. [ 0.9706, -0.1940, 1.5802, 0.2639, -0.6015, -0.3286, 0.4616, 0.1453],
  9519. [ 0.5066, -0.4450, 1.4328, -0.3905, -0.5788, -0.2500, 0.2859, 0.2596],
  9520. [ 0.6623, -0.3516, 1.6435, 0.1237, -0.3628, 0.2921, 0.5939, 0.2574]],
  9521. device='cuda:0', grad_fn=<AddmmBackward>)
  9522. landmarks are: tensor([[[ 0.6537, -0.3623, 1.9115, -0.2690, -0.4037, -0.8309, 0.6926,
  9523. 0.1608],
  9524. [ 0.5730, -0.4562, 1.2195, -1.2440, -0.5497, -0.7711, 0.5704,
  9525. 0.1779],
  9526. [ 0.5435, -0.4597, 1.6575, 0.0159, -0.3229, -0.1150, 0.1877,
  9527. 0.1854],
  9528. [ 0.5301, -0.4273, 1.6344, -0.5692, -0.4961, 0.2622, 0.4098,
  9529. 0.2966],
  9530. [-2.2859, -2.2859, 1.8249, -0.8002, 0.0409, -1.2543, 0.8059,
  9531. 0.3050],
  9532. [ 0.6249, -0.4352, 1.8018, 0.2545, -0.6173, -0.1997, 0.6401,
  9533. 0.0291],
  9534. [ 0.5283, -0.4429, 1.5940, -0.2844, -0.5827, -0.1458, 0.2823,
  9535. 0.3267],
  9536. [ 0.6257, -0.4249, 1.5998, 0.1236, -0.3806, 0.3084, 0.9887,
  9537. 0.3371]]], device='cuda:0')
  9538. loss_train_step before backward: tensor(0.0419, device='cuda:0', grad_fn=<MseLossBackward>)
  9539. loss_train_step after backward: tensor(0.0419, device='cuda:0', grad_fn=<MseLossBackward>)
  9540. loss_train: 1.8925394713878632
  9541. step: 37
  9542. running loss: 0.051149715442915224
  9543. Train Steps: 37/90 Loss: 0.0511 torch.Size([8, 600, 800])
  9544. torch.Size([8, 8])
  9545. tensor([[0.6205, 0.4016, 0.8350, 0.2717, 0.3987, 0.2550, 0.5787, 0.5133],
  9546. [0.6200, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391],
  9547. [0.6255, 0.4017, 0.8688, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901],
  9548. [0.6128, 0.4118, 0.8638, 0.5333, 0.4625, 0.5267, 0.5193, 0.5475],
  9549. [0.6339, 0.4118, 0.7988, 0.5800, 0.3912, 0.4583, 0.7343, 0.5760],
  9550. [0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585],
  9551. [0.6273, 0.4143, 0.8750, 0.5700, 0.3987, 0.4717, 0.6013, 0.5467],
  9552. [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5787, 0.5550]],
  9553. device='cuda:0', dtype=torch.float64)
  9554. predictions are: tensor([[ 0.1391, -0.6635, 1.4068, -1.0223, -0.4018, -1.1637, 0.4464, 0.0945],
  9555. [ 0.4628, -0.4339, 1.3757, -0.9598, -0.3049, -1.1301, 0.4751, 0.1769],
  9556. [ 0.4638, -0.5336, 1.4785, -0.6946, -0.6105, -0.6798, 0.6793, 0.1371],
  9557. [ 0.3253, -0.5462, 1.5459, 0.2808, -0.2164, 0.0366, 0.2902, 0.2458],
  9558. [ 0.2621, -0.6162, 1.4384, 0.0273, -0.4463, -0.3500, 0.6303, 0.2597],
  9559. [ 0.2308, -0.6349, 1.5872, -0.2894, -0.2790, 0.2705, 0.4701, 0.2369],
  9560. [ 0.0074, -0.7388, 1.5592, 0.3414, -0.4608, -0.2093, 0.3406, 0.1944],
  9561. [ 0.0809, -0.7228, 1.3165, -0.9145, -0.6861, -0.7330, 0.3129, 0.2097]],
  9562. device='cuda:0', grad_fn=<AddmmBackward>)
  9563. landmarks are: tensor([[[ 0.5800, -0.4312, 1.5709, -1.0311, -0.4441, -1.1081, 0.3873,
  9564. 0.0851],
  9565. [ 0.5776, -0.3987, 1.6005, -0.9121, -0.3844, -1.2358, 0.4247,
  9566. 0.2043],
  9567. [ 0.6031, -0.4307, 1.7268, -0.8002, -0.6058, -0.6462, 0.6442,
  9568. -0.0220],
  9569. [ 0.5447, -0.3838, 1.7037, 0.1775, -0.1497, 0.1467, 0.1128,
  9570. 0.2431],
  9571. [ 0.6421, -0.3840, 1.4035, 0.3931, -0.4788, -0.1689, 1.1057,
  9572. 0.3745],
  9573. [ 0.5115, -0.4332, 1.7557, -0.3152, -0.2536, 0.3931, 0.4139,
  9574. 0.2936],
  9575. [ 0.6115, -0.3724, 1.7557, 0.3469, -0.4441, -0.1073, 0.4912,
  9576. 0.2391],
  9577. [ 0.5647, -0.4129, 1.4901, -1.0619, -0.6462, -0.5846, 0.3873,
  9578. 0.2776]]], device='cuda:0')
  9579. loss_train_step before backward: tensor(0.0379, device='cuda:0', grad_fn=<MseLossBackward>)
  9580. loss_train_step after backward: tensor(0.0379, device='cuda:0', grad_fn=<MseLossBackward>)
  9581. loss_train: 1.9304494857788086
  9582. step: 38
  9583. running loss: 0.05080130225733707
  9584. Train Steps: 38/90 Loss: 0.0508 torch.Size([8, 600, 800])
  9585. torch.Size([8, 8])
  9586. tensor([[0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626],
  9587. [0.6200, 0.4070, 0.8938, 0.4183, 0.3538, 0.4567, 0.6175, 0.5400],
  9588. [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466],
  9589. [0.6212, 0.4033, 0.8938, 0.4167, 0.3813, 0.4267, 0.5613, 0.5583],
  9590. [0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6063, 0.6083],
  9591. [0.6131, 0.4037, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680],
  9592. [0.6257, 0.4034, 0.8287, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901],
  9593. [0.6216, 0.4167, 0.8588, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667]],
  9594. device='cuda:0', dtype=torch.float64)
  9595. predictions are: tensor([[ 0.3057, -0.5980, 1.6707, 0.3744, -0.3315, -0.2668, 0.4939, 0.1397],
  9596. [ 0.3730, -0.5857, 1.6908, -0.3420, -0.5077, -0.1843, 0.5378, 0.1543],
  9597. [ 0.3094, -0.5495, 1.2988, -0.7486, -0.5875, -0.4744, 0.3223, 0.2321],
  9598. [ 0.0189, -0.7939, 1.7198, -0.3256, -0.4821, -0.3460, 0.3194, 0.2208],
  9599. [-0.0583, -0.8413, 1.6697, -0.4644, -0.5157, -0.3376, 0.4319, 0.2579],
  9600. [-0.0959, -0.8424, 0.9313, -1.0887, -0.5208, -1.0755, 0.2619, 0.2573],
  9601. [ 0.4857, -0.5217, 1.4339, -1.0939, -0.3772, -1.2367, 0.7471, 0.0930],
  9602. [ 0.5980, -0.3976, 1.7367, 0.5304, -0.3018, 0.2016, 0.4894, 0.2275]],
  9603. device='cuda:0', grad_fn=<AddmmBackward>)
  9604. landmarks are: tensor([[[ 0.6105, -0.4293, 1.5824, 0.5239, -0.4730, -0.0380, 0.5025,
  9605. -0.1492],
  9606. [ 0.5778, -0.4061, 1.8423, -0.3537, -0.6520, -0.1766, 0.5663,
  9607. 0.2083],
  9608. [ 0.5076, -0.4443, 1.2337, -0.5023, -0.6808, -0.3614, 0.0866,
  9609. 0.2386],
  9610. [ 0.5832, -0.4231, 1.8423, -0.3614, -0.5249, -0.3152, 0.3065,
  9611. 0.2930],
  9612. [ 0.5774, -0.3868, 1.6286, -0.5692, -0.6462, -0.2767, 0.5143,
  9613. 0.5239],
  9614. [ 0.5459, -0.4215, 0.9043, -0.9838, -0.5827, -1.0388, 0.1236,
  9615. 0.3378],
  9616. [ 0.6042, -0.4225, 1.5420, -1.2082, -0.4730, -1.0311, 0.6380,
  9617. -0.0220],
  9618. [ 0.5853, -0.3614, 1.6806, 0.2930, -0.4499, 0.1005, 0.3815,
  9619. 0.3315]]], device='cuda:0')
  9620. loss_train_step before backward: tensor(0.0421, device='cuda:0', grad_fn=<MseLossBackward>)
  9621. loss_train_step after backward: tensor(0.0421, device='cuda:0', grad_fn=<MseLossBackward>)
  9622. loss_train: 1.9725742973387241
  9623. step: 39
  9624. running loss: 0.05057882813689036
  9625. Train Steps: 39/90 Loss: 0.0506 torch.Size([8, 600, 800])
  9626. torch.Size([8, 8])
  9627. tensor([[ nan, nan, 0.8850, 0.3000, 0.5363, 0.2250, 0.7343, 0.5771],
  9628. [0.6185, 0.4079, 0.8838, 0.4617, 0.4838, 0.5650, 0.6175, 0.5850],
  9629. [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575],
  9630. [ nan, nan, 0.8938, 0.2850, 0.4662, 0.3117, 0.7406, 0.5528],
  9631. [0.6147, 0.4112, 0.7988, 0.3200, 0.3775, 0.2767, 0.5150, 0.5550],
  9632. [0.6135, 0.4115, 0.8838, 0.4667, 0.4288, 0.6050, 0.5778, 0.5097],
  9633. [0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507],
  9634. [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250]],
  9635. device='cuda:0', dtype=torch.float64)
  9636. predictions are: tensor([[-1.5030, -1.7617, 1.5962, -0.7638, -0.0465, -1.1517, 0.7468, 0.2570],
  9637. [ 0.6592, -0.3707, 1.6703, 0.0753, -0.2731, 0.2695, 0.3758, 0.2453],
  9638. [ 0.8878, -0.2623, 1.4946, -0.6248, -0.5640, -0.3283, 0.7205, 0.1871],
  9639. [-0.2709, -0.9534, 1.8355, -0.6016, -0.2469, -1.0011, 0.8533, 0.1493],
  9640. [ 0.1519, -0.6594, 1.1395, -0.7464, -0.7456, -0.9452, -0.0113, 0.1914],
  9641. [ 0.5785, -0.4321, 1.6872, 0.1249, -0.4040, 0.3933, 0.3312, 0.1839],
  9642. [ 0.5743, -0.4212, 0.8302, -1.1150, -0.6334, -1.2068, 0.3029, 0.2081],
  9643. [ 0.6235, -0.3960, 1.6506, 0.2356, -0.4930, 0.2425, 0.3186, 0.1754]],
  9644. device='cuda:0', grad_fn=<AddmmBackward>)
  9645. landmarks are: tensor([[[-2.2859, -2.2859, 1.8018, -0.9002, 0.1910, -1.2467, 1.1057,
  9646. 0.3799],
  9647. [ 0.5707, -0.4017, 1.7961, -0.1535, -0.0515, 0.3238, 0.5663,
  9648. 0.4162],
  9649. [ 0.6158, -0.4249, 1.8654, -0.9002, -0.3229, -0.3537, 0.9667,
  9650. 0.2890],
  9651. [-2.2859, -2.2859, 1.8423, -0.9695, -0.1323, -0.8463, 1.1349,
  9652. 0.2676],
  9653. [ 0.5532, -0.3864, 1.4035, -0.8079, -0.5423, -1.0080, 0.0928,
  9654. 0.2776],
  9655. [ 0.5477, -0.3851, 1.7961, -0.1304, -0.3055, 0.5085, 0.3830,
  9656. 0.0682],
  9657. [ 0.5185, -0.4252, 0.9647, -1.2928, -0.4788, -1.2390, 0.2617,
  9658. 0.2576],
  9659. [ 0.5991, -0.3803, 1.8018, -0.0534, -0.3460, 0.1852, 0.5374,
  9660. 0.1390]]], device='cuda:0')
  9661. loss_train_step before backward: tensor(0.1352, device='cuda:0', grad_fn=<MseLossBackward>)
  9662. loss_train_step after backward: tensor(0.1352, device='cuda:0', grad_fn=<MseLossBackward>)
  9663. loss_train: 2.107771832495928
  9664. step: 40
  9665. running loss: 0.05269429581239819
  9666.  
  9667. Train Steps: 40/90 Loss: 0.0527 torch.Size([8, 600, 800])
  9668. torch.Size([8, 8])
  9669. tensor([[ nan, nan, 0.7268, 0.2333, 0.4125, 0.1933, 0.5112, 0.5383],
  9670. [0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433],
  9671. [0.6329, 0.4196, 0.9238, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748],
  9672. [0.6125, 0.4076, 0.8488, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505],
  9673. [0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412],
  9674. [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5687, 0.5567],
  9675. [0.6200, 0.4049, 0.8638, 0.5617, 0.4125, 0.5100, 0.6013, 0.5317],
  9676. [ nan, nan, 0.7335, 0.2569, 0.3788, 0.2667, 0.5066, 0.5578]],
  9677. device='cuda:0', dtype=torch.float64)
  9678. predictions are: tensor([[-1.4297, -1.6886, 0.9586, -1.0350, -0.4966, -1.1245, 0.2422, 0.2725],
  9679. [ 0.7484, -0.3261, 1.7147, 0.2348, -0.1303, 0.2234, 0.4167, 0.1816],
  9680. [ 0.9006, -0.1798, 1.8771, -0.0190, -0.2798, -0.6128, 0.7785, 0.1212],
  9681. [ 0.9955, -0.1387, 1.5056, -0.5418, -0.7577, -0.3477, 0.2791, 0.1715],
  9682. [ 1.1263, -0.0555, 1.6218, -0.5745, -0.5147, -0.7123, 0.5116, 0.1374],
  9683. [ 0.1350, -0.6184, 1.1729, -1.1085, -0.3779, -0.8162, 0.4339, 0.3186],
  9684. [ 0.9295, -0.2291, 1.6967, 0.4022, -0.3601, 0.3717, 0.5515, 0.1648],
  9685. [-2.0216, -2.0805, 1.1421, -0.9266, -0.5279, -0.9488, 0.2146, 0.2698]],
  9686. device='cuda:0', grad_fn=<AddmmBackward>)
  9687. landmarks are: tensor([[[-2.2859, -2.2859, 1.0712, -1.2085, -0.3806, -1.3929, 0.0755,
  9688. 0.2006],
  9689. [ 0.5295, -0.4373, 1.6553, 0.1011, 0.0380, 0.0671, 0.0813,
  9690. 0.2237],
  9691. [ 0.6372, -0.3479, 1.9808, -0.1689, -0.3171, -0.9310, 1.1057,
  9692. 0.3692],
  9693. [ 0.5433, -0.4032, 1.6344, -0.4922, -0.5769, -0.5846, 0.0357,
  9694. 0.2567],
  9695. [ 0.5784, -0.4006, 1.7911, -0.5701, -0.5192, -1.0331, 0.4137,
  9696. 0.2139],
  9697. [ 0.5824, -0.3625, 1.0405, -0.9325, -0.4268, -1.1389, 0.3411,
  9698. 0.2853],
  9699. [ 0.5776, -0.4159, 1.7037, 0.3084, -0.3806, 0.0697, 0.4912,
  9700. 0.1698],
  9701. [-2.2859, -2.2859, 1.1020, -1.0994, -0.5365, -1.0542, 0.0542,
  9702. 0.2905]]], device='cuda:0')
  9703. loss_train_step before backward: tensor(0.0592, device='cuda:0', grad_fn=<MseLossBackward>)
  9704. loss_train_step after backward: tensor(0.0592, device='cuda:0', grad_fn=<MseLossBackward>)
  9705. loss_train: 2.1670092418789864
  9706. step: 41
  9707. running loss: 0.05285388394826796
  9708. Train Steps: 41/90 Loss: 0.0529 torch.Size([8, 600, 800])
  9709. torch.Size([8, 8])
  9710. tensor([[0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528],
  9711. [0.6161, 0.4024, 0.8662, 0.4683, 0.4935, 0.5364, 0.6063, 0.5567],
  9712. [0.6113, 0.4006, 0.8700, 0.5350, 0.3638, 0.3767, 0.5097, 0.4882],
  9713. [0.6266, 0.4070, 0.8712, 0.5600, 0.3713, 0.4783, 0.5775, 0.6100],
  9714. [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517],
  9715. [0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343],
  9716. [0.6236, 0.3967, 0.8675, 0.5400, 0.3862, 0.4517, 0.5825, 0.5200],
  9717. [0.6163, 0.4114, 0.7650, 0.2017, 0.3763, 0.2867, 0.5631, 0.5071]],
  9718. device='cuda:0', dtype=torch.float64)
  9719. predictions are: tensor([[-0.3570, -1.0477, 1.9500, -0.7166, -0.2661, -0.7335, 0.7896, 0.2005],
  9720. [ 0.1600, -0.7055, 1.7179, -0.2601, -0.1441, 0.0563, 0.3919, 0.2460],
  9721. [ 0.2115, -0.6359, 1.6255, -0.0376, -0.6835, -0.6661, 0.2499, 0.1468],
  9722. [-0.0534, -0.8461, 1.5375, 0.2503, -0.5311, -0.0212, 0.3671, 0.3205],
  9723. [ 0.3429, -0.5866, 1.6651, -0.2636, -0.2702, 0.2513, 0.4242, 0.2724],
  9724. [ 0.3025, -0.6347, 1.5513, -0.8119, -0.4023, -0.8750, 0.6761, 0.1629],
  9725. [ 0.1845, -0.6954, 1.6102, 0.1688, -0.5439, -0.2238, 0.3499, 0.2069],
  9726. [ 0.6012, -0.4584, 1.0800, -1.4243, -0.6784, -1.1147, 0.3999, 0.1815]],
  9727. device='cuda:0', grad_fn=<AddmmBackward>)
  9728. landmarks are: tensor([[[ 0.6388, -0.3623, 1.9173, -0.7386, -0.3517, -0.5846, 1.1495,
  9729. 0.2676],
  9730. [ 0.5598, -0.4273, 1.7152, -0.1227, -0.0065, 0.1917, 0.5143,
  9731. 0.2853],
  9732. [ 0.5379, -0.4358, 1.7326, 0.1852, -0.6058, -0.5461, 0.0684,
  9733. -0.0310],
  9734. [ 0.6081, -0.4059, 1.7383, 0.3007, -0.5711, -0.0765, 0.3815,
  9735. 0.5316],
  9736. [ 0.5897, -0.3804, 1.7788, -0.0226, -0.2016, 0.3007, 0.6125,
  9737. 0.2622],
  9738. [ 0.6125, -0.4369, 1.9173, -0.5384, -0.2594, -0.8386, 0.9741,
  9739. 0.1821],
  9740. [ 0.5944, -0.4538, 1.7210, 0.2083, -0.5018, -0.1997, 0.4046,
  9741. 0.1159],
  9742. [ 0.5606, -0.3856, 1.2476, -1.3544, -0.5480, -0.9618, 0.3152,
  9743. 0.0562]]], device='cuda:0')
  9744. loss_train_step before backward: tensor(0.0622, device='cuda:0', grad_fn=<MseLossBackward>)
  9745. loss_train_step after backward: tensor(0.0622, device='cuda:0', grad_fn=<MseLossBackward>)
  9746. loss_train: 2.229224521666765
  9747. step: 42
  9748. running loss: 0.05307677432539917
  9749. Train Steps: 42/90 Loss: 0.0531 torch.Size([8, 600, 800])
  9750. torch.Size([8, 8])
  9751. tensor([[0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482],
  9752. [0.6143, 0.4040, 0.8237, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973],
  9753. [0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133],
  9754. [0.6145, 0.4007, 0.8775, 0.4533, 0.4562, 0.5533, 0.6088, 0.5533],
  9755. [0.6115, 0.4081, 0.6725, 0.2433, 0.4088, 0.1933, 0.5167, 0.5544],
  9756. [0.6212, 0.4171, 0.7875, 0.3633, 0.3813, 0.2933, 0.5675, 0.5700],
  9757. [0.6111, 0.4033, 0.8300, 0.3267, 0.3588, 0.3333, 0.5444, 0.5637],
  9758. [0.6064, 0.3953, 0.8738, 0.4417, 0.3663, 0.4683, 0.5511, 0.5416]],
  9759. device='cuda:0', dtype=torch.float64)
  9760. predictions are: tensor([[ 0.2737, -0.6350, 1.8644, -0.3020, -0.5419, -0.4866, 0.5516, 0.1753],
  9761. [-0.2903, -0.9783, 1.4274, -0.8604, -0.4192, -1.0997, 0.3088, 0.1842],
  9762. [ 0.2797, -0.6562, 1.8521, 0.2928, -0.4158, 0.0558, 0.6270, 0.1605],
  9763. [ 0.1652, -0.7392, 1.9001, -0.0229, -0.1119, 0.4007, 0.5122, 0.2274],
  9764. [ 0.1744, -0.7025, 0.9219, -1.1211, -0.4517, -1.2485, 0.2254, 0.2856],
  9765. [ 0.4882, -0.4664, 1.4131, -0.6096, -0.4171, -0.6723, 0.5737, 0.3040],
  9766. [ 0.2635, -0.6419, 1.5242, -0.7674, -0.6309, -0.7327, 0.2655, 0.2337],
  9767. [ 0.0690, -0.8091, 1.8354, -0.1112, -0.5163, 0.0057, 0.4869, 0.1616]],
  9768. device='cuda:0', grad_fn=<AddmmBackward>)
  9769. landmarks are: tensor([[[ 0.5791, -0.4127, 1.8442, -0.3985, -0.6031, -0.6154, 0.4473,
  9770. 0.2464],
  9771. [ 0.5515, -0.4201, 1.5189, -0.7463, -0.4037, -1.3082, 0.0869,
  9772. 0.0111],
  9773. [ 0.6148, -0.4130, 1.6864, 0.1698, -0.5307, -0.1150, 0.6125,
  9774. 0.0851],
  9775. [ 0.5526, -0.4351, 1.7672, -0.1920, -0.1785, 0.2699, 0.5259,
  9776. 0.2699],
  9777. [ 0.5385, -0.4010, 0.8203, -1.1620, -0.3979, -1.3929, 0.1005,
  9778. 0.2747],
  9779. [ 0.5832, -0.3593, 1.3515, -0.6077, -0.5249, -0.9310, 0.3353,
  9780. 0.3469],
  9781. [ 0.5366, -0.4232, 1.5478, -0.7771, -0.6289, -0.7463, 0.2288,
  9782. 0.3177],
  9783. [ 0.5149, -0.4603, 1.7499, -0.2459, -0.5942, -0.1227, 0.2596,
  9784. 0.2155]]], device='cuda:0')
  9785. loss_train_step before backward: tensor(0.0456, device='cuda:0', grad_fn=<MseLossBackward>)
  9786. loss_train_step after backward: tensor(0.0456, device='cuda:0', grad_fn=<MseLossBackward>)
  9787. loss_train: 2.2748365253210068
  9788. step: 43
  9789. running loss: 0.052903175007465275
  9790. Train Steps: 43/90 Loss: 0.0529 torch.Size([8, 600, 800])
  9791. torch.Size([8, 8])
  9792. tensor([[0.6157, 0.4102, 0.8513, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890],
  9793. [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578],
  9794. [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717],
  9795. [0.6261, 0.4029, 0.8720, 0.3364, 0.3665, 0.3753, 0.6531, 0.5183],
  9796. [0.6268, 0.4052, 0.8175, 0.2250, 0.4688, 0.1917, 0.6375, 0.5267],
  9797. [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726],
  9798. [0.6139, 0.4019, 0.7137, 0.2150, 0.4375, 0.1533, 0.5293, 0.5006],
  9799. [0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533]],
  9800. device='cuda:0', dtype=torch.float64)
  9801. predictions are: tensor([[ 0.5413, -0.4571, 1.9170, 0.0479, -0.6285, 0.0270, 0.2009, 0.2187],
  9802. [-0.0423, -0.8500, 1.2509, -0.7857, -0.3819, -0.8803, 0.3350, 0.2688],
  9803. [ 0.3422, -0.5660, 1.1678, -0.6439, -0.4835, -0.6990, 0.2194, 0.2989],
  9804. [ 0.1666, -0.7358, 1.9808, -0.2059, -0.4849, 0.0364, 0.7351, 0.1599],
  9805. [ 0.1013, -0.7613, 1.7859, -0.6040, -0.0864, -0.9437, 0.7968, 0.1880],
  9806. [ 0.3649, -0.5743, 1.5560, -0.5668, -0.5240, -0.5518, 0.4174, 0.1397],
  9807. [-0.6398, -1.2337, 1.2423, -0.8344, -0.2962, -1.1112, 0.2853, 0.2156],
  9808. [ 0.5759, -0.4420, 1.3873, -0.7166, -0.5027, -0.4597, 0.4913, 0.2565]],
  9809. device='cuda:0', grad_fn=<AddmmBackward>)
  9810. landmarks are: tensor([[[ 0.5581, -0.3912, 1.6460, -0.5230, -0.6173, -0.5923, 0.0681,
  9811. 0.4348],
  9812. [ 0.5232, -0.4189, 0.9358, -1.4006, -0.3921, -1.3698, 0.2555,
  9813. 0.2906],
  9814. [ 0.5633, -0.3883, 0.8648, -1.0288, -0.5192, -1.1081, 0.2776,
  9815. 0.3546],
  9816. [ 0.6059, -0.4249, 1.7420, -0.7321, -0.5931, -0.5523, 0.7307,
  9817. 0.1080],
  9818. [ 0.6092, -0.4143, 1.4901, -1.2467, -0.1208, -1.4006, 0.6587,
  9819. 0.1467],
  9820. [ 0.4997, -0.4446, 1.1610, -0.9772, -0.6058, -1.0311, 0.1404,
  9821. -0.1031],
  9822. [ 0.5496, -0.4295, 1.0109, -1.2928, -0.2651, -1.5777, 0.1589,
  9823. 0.0261],
  9824. [ 0.5878, -0.4052, 1.0229, -1.2855, -0.5596, -0.8232, 0.5316,
  9825. 0.2699]]], device='cuda:0')
  9826. loss_train_step before backward: tensor(0.1343, device='cuda:0', grad_fn=<MseLossBackward>)
  9827. loss_train_step after backward: tensor(0.1343, device='cuda:0', grad_fn=<MseLossBackward>)
  9828. loss_train: 2.4091132134199142
  9829. step: 44
  9830. running loss: 0.054752573032270775
  9831.  
  9832. Train Steps: 44/90 Loss: 0.0548 torch.Size([8, 600, 800])
  9833. torch.Size([8, 8])
  9834. tensor([[0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118],
  9835. [0.6188, 0.4099, 0.7400, 0.2433, 0.3962, 0.2750, 0.6162, 0.5467],
  9836. [0.6097, 0.3988, 0.8650, 0.5250, 0.4213, 0.5200, 0.5675, 0.5050],
  9837. [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656],
  9838. [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301],
  9839. [0.6201, 0.4151, 0.8588, 0.5467, 0.3700, 0.3950, 0.5637, 0.5933],
  9840. [0.6117, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103],
  9841. [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6088, 0.5583]],
  9842. device='cuda:0', dtype=torch.float64)
  9843. predictions are: tensor([[ 0.4334, -0.5116, 1.3480, -0.8113, -0.3142, -1.0172, 0.4388, 0.3419],
  9844. [ 0.4893, -0.5194, 1.3963, -1.1550, -0.5044, -1.0069, 0.5392, 0.1921],
  9845. [ 0.0556, -0.8011, 2.0050, 0.2513, -0.3090, 0.1978, 0.3072, 0.1363],
  9846. [ 0.0547, -0.7857, 1.8429, -0.2424, -0.5380, -0.0357, 0.3545, 0.2218],
  9847. [ 0.0402, -0.8175, 1.9599, 0.0763, -0.3227, 0.1221, 0.5820, 0.1758],
  9848. [ 0.4812, -0.4551, 1.8308, 0.2619, -0.5223, -0.2912, 0.3599, 0.2569],
  9849. [-0.0621, -0.8441, 1.1563, -1.2337, -0.5207, -1.1930, 0.2766, 0.2215],
  9850. [ 0.6473, -0.4163, 1.2970, -1.1117, -0.4648, -1.1489, 0.4737, 0.2342]],
  9851. device='cuda:0', grad_fn=<AddmmBackward>)
  9852. landmarks are: tensor([[[ 0.6077, -0.3226, 0.9993, -0.6462, -0.2651, -1.3082, 0.2946,
  9853. 0.5401],
  9854. [ 0.5725, -0.3926, 1.1321, -1.1620, -0.4557, -1.0157, 0.5605,
  9855. 0.2391],
  9856. [ 0.5303, -0.4440, 1.7095, 0.1390, -0.3402, 0.1159, 0.3353,
  9857. 0.0467],
  9858. [ 0.5283, -0.4429, 1.5940, -0.2844, -0.5827, -0.1458, 0.2823,
  9859. 0.3267],
  9860. [ 0.6010, -0.4562, 1.7198, -0.0090, -0.3464, 0.0108, 0.6294,
  9861. 0.1627],
  9862. [ 0.5785, -0.3687, 1.6806, 0.2391, -0.5769, -0.4614, 0.3180,
  9863. 0.4547],
  9864. [ 0.5393, -0.4299, 0.7453, -1.3775, -0.5596, -1.1081, 0.1527,
  9865. 0.0712],
  9866. [ 0.5992, -0.3968, 0.9307, -1.3497, -0.4730, -1.0465, 0.5259,
  9867. 0.2930]]], device='cuda:0')
  9868. loss_train_step before backward: tensor(0.0466, device='cuda:0', grad_fn=<MseLossBackward>)
  9869. loss_train_step after backward: tensor(0.0466, device='cuda:0', grad_fn=<MseLossBackward>)
  9870. loss_train: 2.45572542399168
  9871. step: 45
  9872. running loss: 0.054571676088704
  9873. Train Steps: 45/90 Loss: 0.0546 torch.Size([8, 600, 800])
  9874. torch.Size([8, 8])
  9875. tensor([[0.6266, 0.4101, 0.8350, 0.2333, 0.3950, 0.2950, 0.6264, 0.4921],
  9876. [0.6236, 0.3967, 0.8675, 0.5400, 0.3862, 0.4517, 0.5825, 0.5200],
  9877. [0.6097, 0.3988, 0.8650, 0.5250, 0.4213, 0.5200, 0.5675, 0.5050],
  9878. [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167],
  9879. [0.6256, 0.4199, 0.8638, 0.5800, 0.3987, 0.4383, 0.5600, 0.5950],
  9880. [0.6286, 0.4086, 0.8408, 0.2801, 0.4163, 0.2800, 0.6725, 0.5393],
  9881. [0.6188, 0.4099, 0.7400, 0.2433, 0.3962, 0.2750, 0.6162, 0.5467],
  9882. [0.6339, 0.4123, 0.8638, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436]],
  9883. device='cuda:0', dtype=torch.float64)
  9884. predictions are: tensor([[ 0.0486, -0.7520, 1.5484, -1.2891, -0.3776, -1.1524, 0.5029, 0.2081],
  9885. [ 0.1276, -0.7314, 1.7045, 0.1468, -0.5277, -0.3232, 0.3239, 0.1693],
  9886. [ 0.1799, -0.6887, 1.7231, -0.0090, -0.3742, 0.0133, 0.1609, 0.1769],
  9887. [ 0.4219, -0.5362, 1.5284, -0.9328, -0.6049, -0.6821, 0.3582, 0.3489],
  9888. [ 0.5471, -0.4507, 1.6436, 0.3289, -0.4051, -0.3247, 0.3472, 0.3031],
  9889. [ 0.4886, -0.4759, 1.6206, -1.0411, -0.2340, -1.1404, 0.6254, 0.2247],
  9890. [ 0.5113, -0.4774, 1.1860, -1.2704, -0.5188, -1.0736, 0.4048, 0.2197],
  9891. [ 0.2363, -0.6918, 1.8009, -0.0398, -0.4113, 0.2251, 0.6258, 0.2378]],
  9892. device='cuda:0', grad_fn=<AddmmBackward>)
  9893. landmarks are: tensor([[[ 0.6081, -0.3918, 1.5709, -1.2082, -0.4614, -0.9233, 0.6072,
  9894. -0.0129],
  9895. [ 0.5944, -0.4538, 1.7210, 0.2083, -0.5018, -0.1997, 0.4046,
  9896. 0.1159],
  9897. [ 0.5303, -0.4440, 1.7095, 0.1390, -0.3402, 0.1159, 0.3353,
  9898. 0.0467],
  9899. [ 0.5768, -0.3899, 1.3861, -0.7771, -0.5885, -0.5461, 0.5028,
  9900. 0.5624],
  9901. [ 0.6038, -0.3464, 1.7037, 0.3931, -0.4441, -0.2613, 0.3007,
  9902. 0.4624],
  9903. [ 0.6174, -0.3984, 1.5975, -0.9921, -0.3633, -0.9926, 0.8205,
  9904. 0.2050],
  9905. [ 0.5725, -0.3926, 1.1321, -1.1620, -0.4557, -1.0157, 0.5605,
  9906. 0.2391],
  9907. [ 0.6421, -0.3816, 1.7037, 0.1929, -0.4037, 0.2391, 1.1861,
  9908. 0.2249]]], device='cuda:0')
  9909. loss_train_step before backward: tensor(0.0342, device='cuda:0', grad_fn=<MseLossBackward>)
  9910. loss_train_step after backward: tensor(0.0342, device='cuda:0', grad_fn=<MseLossBackward>)
  9911. loss_train: 2.489884562790394
  9912. step: 46
  9913. running loss: 0.05412792527805204
  9914. Train Steps: 46/90 Loss: 0.0541 torch.Size([8, 600, 800])
  9915. torch.Size([8, 8])
  9916. tensor([[0.6308, 0.3990, 0.8688, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133],
  9917. [0.6207, 0.4081, 0.7662, 0.2067, 0.3962, 0.3200, 0.6312, 0.5300],
  9918. [0.6216, 0.4099, 0.7225, 0.2033, 0.4188, 0.2217, 0.5975, 0.5283],
  9919. [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283],
  9920. [0.6154, 0.4112, 0.7037, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600],
  9921. [0.6161, 0.4024, 0.8838, 0.4583, 0.3688, 0.3733, 0.5311, 0.5344],
  9922. [0.6058, 0.3986, 0.8324, 0.4626, 0.3838, 0.4983, 0.5147, 0.5466],
  9923. [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986]],
  9924. device='cuda:0', dtype=torch.float64)
  9925. predictions are: tensor([[ 0.2156, -0.7264, 1.8884, 0.1304, -0.4089, 0.0797, 0.6120, 0.1767],
  9926. [ 0.5065, -0.4851, 1.3740, -1.1749, -0.4529, -0.8285, 0.6044, 0.2467],
  9927. [ 0.7207, -0.3375, 1.1234, -1.1218, -0.3663, -1.1669, 0.2888, 0.2777],
  9928. [ 0.5620, -0.4515, 1.7631, -0.4039, -0.5657, -0.5033, 0.4060, 0.2434],
  9929. [ 0.3018, -0.6048, 1.0069, -1.1920, -0.3891, -1.2506, 0.2564, 0.3108],
  9930. [ 0.4506, -0.5106, 1.7971, -0.1529, -0.6035, -0.4831, 0.2787, 0.1326],
  9931. [ 0.1461, -0.7076, 1.6684, -0.1707, -0.4278, 0.0296, 0.3463, 0.2320],
  9932. [ 0.3512, -0.5647, 1.8268, -0.5372, -0.2307, -1.0289, 0.5388, 0.2037]],
  9933. device='cuda:0', grad_fn=<AddmmBackward>)
  9934. landmarks are: tensor([[[ 0.6275, -0.4430, 1.7268, 0.1082, -0.4614, 0.0159, 0.6644,
  9935. 0.0851],
  9936. [ 0.5809, -0.4011, 1.2533, -1.3313, -0.4557, -0.8079, 0.6298,
  9937. 0.1621],
  9938. [ 0.5850, -0.3925, 1.0513, -1.3467, -0.3517, -1.2620, 0.4739,
  9939. 0.1544],
  9940. [ 0.5880, -0.3887, 1.8423, -0.3306, -0.6231, -0.5230, 0.4046,
  9941. 0.1544],
  9942. [ 0.5565, -0.3864, 0.9647, -1.2159, -0.3286, -1.4391, 0.1852,
  9943. 0.3007],
  9944. [ 0.5598, -0.4273, 1.7961, -0.1689, -0.5827, -0.5615, 0.1671,
  9945. 0.1824],
  9946. [ 0.5124, -0.4446, 1.5587, -0.1493, -0.5134, 0.0159, 0.0912,
  9947. 0.2386],
  9948. [ 0.6054, -0.3641, 1.8711, -0.5153, -0.2305, -1.3005, 0.5948,
  9949. 0.0171]]], device='cuda:0')
  9950. loss_train_step before backward: tensor(0.0191, device='cuda:0', grad_fn=<MseLossBackward>)
  9951. loss_train_step after backward: tensor(0.0191, device='cuda:0', grad_fn=<MseLossBackward>)
  9952. loss_train: 2.5090342573821545
  9953. step: 47
  9954. running loss: 0.053383707603875626
  9955. Train Steps: 47/90 Loss: 0.0534 torch.Size([8, 600, 800])
  9956. torch.Size([8, 8])
  9957. tensor([[0.6135, 0.4115, 0.8838, 0.4667, 0.4288, 0.6050, 0.5778, 0.5097],
  9958. [0.6353, 0.4128, 0.8488, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540],
  9959. [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466],
  9960. [0.6133, 0.4066, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358],
  9961. [0.6259, 0.4156, 0.8812, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960],
  9962. [0.6261, 0.4066, 0.8325, 0.2150, 0.4763, 0.2667, 0.7002, 0.5633],
  9963. [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5837, 0.5500],
  9964. [0.6128, 0.4022, 0.8738, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064]],
  9965. device='cuda:0', dtype=torch.float64)
  9966. predictions are: tensor([[ 0.8475, -0.2794, 1.7623, -0.0957, -0.4256, 0.2351, 0.4581, 0.2182],
  9967. [-0.2862, -0.9827, 1.5954, -1.0876, -0.1128, -1.1886, 0.7696, 0.2542],
  9968. [ 0.5297, -0.4046, 1.3062, -0.6987, -0.7591, -0.5133, 0.2072, 0.2336],
  9969. [ 0.7029, -0.3222, 0.8991, -1.0768, -0.5859, -1.2088, 0.0907, 0.2763],
  9970. [ 0.4878, -0.4625, 1.5844, -0.8308, -0.2686, -1.0829, 0.5105, 0.2126],
  9971. [ 0.3900, -0.5851, 1.5981, -1.0304, -0.2450, -1.2939, 0.9065, 0.1962],
  9972. [ 0.4773, -0.5220, 1.6869, -0.1609, -0.5247, 0.0040, 0.3232, 0.2207],
  9973. [ 0.4895, -0.4908, 1.6441, 0.0550, -0.2706, -0.1646, 0.1716, 0.2209]],
  9974. device='cuda:0', grad_fn=<AddmmBackward>)
  9975. landmarks are: tensor([[[ 0.5477, -0.3851, 1.7961, -0.1304, -0.3055, 0.5085, 0.3830,
  9976. 0.0682],
  9977. [ 0.6487, -0.3792, 1.6344, -1.0850, 0.2659, -1.5397, 0.8059,
  9978. 0.2730],
  9979. [ 0.5076, -0.4443, 1.2337, -0.5023, -0.6808, -0.3614, 0.0866,
  9980. 0.2386],
  9981. [ 0.5470, -0.4081, 0.8492, -1.0773, -0.5307, -1.1620, 0.0912,
  9982. 0.1890],
  9983. [ 0.6052, -0.3663, 1.7845, -0.8156, -0.0804, -1.4237, 0.5866,
  9984. 0.0051],
  9985. [ 0.6059, -0.4080, 1.5594, -1.2928, -0.0861, -1.0542, 0.9485,
  9986. 0.3157],
  9987. [ 0.5604, -0.4620, 1.7961, -0.1997, -0.3229, 0.1082, 0.4104,
  9988. 0.2545],
  9989. [ 0.5446, -0.4280, 1.7499, 0.0543, 0.0156, 0.1301, 0.1918,
  9990. 0.0532]]], device='cuda:0')
  9991. loss_train_step before backward: tensor(0.0418, device='cuda:0', grad_fn=<MseLossBackward>)
  9992.  
  9993. loss_train_step after backward: tensor(0.0418, device='cuda:0', grad_fn=<MseLossBackward>)
  9994. loss_train: 2.550849635154009
  9995. step: 48
  9996. running loss: 0.05314270073237518
  9997. Train Steps: 48/90 Loss: 0.0531 torch.Size([8, 600, 800])
  9998. torch.Size([8, 8])
  9999. tensor([[0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927],
  10000. [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760],
  10001. [0.6202, 0.4064, 0.7879, 0.2179, 0.4567, 0.1725, 0.5955, 0.5478],
  10002. [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167],
  10003. [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973],
  10004. [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103],
  10005. [0.6185, 0.4098, 0.8838, 0.4900, 0.4537, 0.5800, 0.6288, 0.5400],
  10006. [0.6204, 0.4055, 0.8438, 0.5733, 0.4574, 0.4801, 0.5487, 0.5617]],
  10007. device='cuda:0', dtype=torch.float64)
  10008. predictions are: tensor([[ 0.5748, -0.4237, 1.6063, -0.8094, -0.6753, -0.6006, 0.4739, 0.1426],
  10009. [ 0.6394, -0.3932, 1.5600, -0.2398, -0.4541, -0.4561, 0.6120, 0.2591],
  10010. [ 0.7896, -0.2924, 1.2072, -1.4725, -0.2467, -1.6366, 0.4556, 0.2659],
  10011. [ 0.6717, -0.3560, 1.6344, -0.4148, -0.4480, -0.3918, 0.3761, 0.2047],
  10012. [ 0.5630, -0.4586, 1.6663, -0.2600, -0.5572, -0.5696, 0.4639, 0.1523],
  10013. [ 0.3004, -0.6163, 1.5867, -0.7224, -0.5165, -0.5790, 0.2084, 0.2126],
  10014. [ 0.4216, -0.5652, 1.7460, -0.4298, -0.1611, -0.1235, 0.5190, 0.2736],
  10015. [ 0.5407, -0.4184, 1.4882, -0.1265, -0.2424, -0.5134, 0.2740, 0.2863]],
  10016. device='cuda:0', grad_fn=<AddmmBackward>)
  10017. landmarks are: tensor([[[ 0.6010, -0.4345, 1.8480, -0.3537, -0.6462, -0.2613, 0.6524,
  10018. -0.0099],
  10019. [ 0.6421, -0.4008, 1.6055, 0.2160, -0.5076, -0.0534, 1.1020,
  10020. 0.3745],
  10021. [ 0.5787, -0.4086, 1.3535, -1.2794, -0.1764, -1.4891, 0.4645,
  10022. 0.2442],
  10023. [ 0.5828, -0.4066, 1.7557, 0.0774, -0.3691, -0.0226, 0.4277,
  10024. 0.1005],
  10025. [ 0.6240, -0.4321, 1.8423, 0.1852, -0.5885, -0.1612, 0.6962,
  10026. 0.0111],
  10027. [ 0.5253, -0.4392, 1.7730, -0.2305, -0.4268, -0.1381, 0.1651,
  10028. 0.0712],
  10029. [ 0.5709, -0.3933, 1.7961, -0.0226, -0.1901, 0.3931, 0.6182,
  10030. 0.2083],
  10031. [ 0.5795, -0.4129, 1.6113, 0.3623, -0.1733, -0.0684, 0.2487,
  10032. 0.3084]]], device='cuda:0')
  10033. loss_train_step before backward: tensor(0.0571, device='cuda:0', grad_fn=<MseLossBackward>)
  10034. loss_train_step after backward: tensor(0.0571, device='cuda:0', grad_fn=<MseLossBackward>)
  10035. loss_train: 2.607935193926096
  10036. step: 49
  10037. running loss: 0.05322316722298155
  10038. Train Steps: 49/90 Loss: 0.0532 torch.Size([8, 600, 800])
  10039. torch.Size([8, 8])
  10040. tensor([[0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656],
  10041. [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483],
  10042. [ nan, nan, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729],
  10043. [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5988, 0.5667],
  10044. [0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153],
  10045. [0.6207, 0.4081, 0.7662, 0.2067, 0.3962, 0.3200, 0.6312, 0.5300],
  10046. [0.6184, 0.4079, 0.8350, 0.3700, 0.3675, 0.2883, 0.5312, 0.5783],
  10047. [0.6264, 0.3972, 0.8853, 0.4771, 0.3853, 0.4511, 0.6293, 0.5334]],
  10048. device='cuda:0', dtype=torch.float64)
  10049. predictions are: tensor([[ 0.6072, -0.3902, 1.6371, -0.2984, -0.5227, -0.1801, 0.3244, 0.2093],
  10050. [ 0.7488, -0.2995, 1.2322, -1.1851, -0.3223, -1.2072, 0.5751, 0.2034],
  10051. [-0.4687, -1.0840, 0.9513, -1.2575, -0.3393, -1.3697, 0.1948, 0.2901],
  10052. [ 1.0454, -0.1078, 1.6811, -0.6875, -0.4405, -0.7529, 0.4998, 0.2144],
  10053. [ 0.6249, -0.4240, 1.7408, 0.0835, -0.1700, 0.0287, 0.4339, 0.1851],
  10054. [ 0.5893, -0.3850, 1.3182, -1.1520, -0.3977, -0.9190, 0.5694, 0.2199],
  10055. [ 1.0067, -0.1159, 1.5522, -0.5882, -0.5079, -0.9046, 0.2085, 0.2066],
  10056. [ 0.7296, -0.3669, 1.8127, -0.0611, -0.4475, -0.2764, 0.6066, 0.1349]],
  10057. device='cuda:0', grad_fn=<AddmmBackward>)
  10058. landmarks are: tensor([[[ 0.5283, -0.4429, 1.5940, -0.2844, -0.5827, -0.1458, 0.2823,
  10059. 0.3267],
  10060. [ 0.5784, -0.4085, 1.0859, -1.3929, -0.4037, -1.1158, 0.5605,
  10061. 0.2468],
  10062. [-2.2859, -2.2859, 0.8824, -1.2727, -0.3691, -1.4237, 0.0943,
  10063. 0.3604],
  10064. [ 0.5783, -0.3933, 1.6748, -0.6154, -0.5769, -0.6462, 0.4797,
  10065. 0.3315],
  10066. [ 0.5548, -0.4682, 1.7309, 0.0966, -0.1394, 0.1757, 0.5045,
  10067. 0.0942],
  10068. [ 0.5809, -0.4011, 1.2533, -1.3313, -0.4557, -0.8079, 0.6298,
  10069. 0.1621],
  10070. [ 0.5704, -0.4019, 1.5709, -0.5769, -0.5885, -0.9541, 0.1679,
  10071. 0.3854],
  10072. [ 0.6075, -0.4514, 1.8032, -0.0822, -0.5061, -0.2023, 0.6208,
  10073. 0.1779]]], device='cuda:0')
  10074. loss_train_step before backward: tensor(0.0894, device='cuda:0', grad_fn=<MseLossBackward>)
  10075. loss_train_step after backward: tensor(0.0894, device='cuda:0', grad_fn=<MseLossBackward>)
  10076. loss_train: 2.6973820067942142
  10077. step: 50
  10078. running loss: 0.05394764013588429
  10079. Train Steps: 50/90 Loss: 0.0539 torch.Size([8, 600, 800])
  10080. torch.Size([8, 8])
  10081. tensor([[0.6097, 0.3988, 0.8650, 0.5250, 0.4213, 0.5200, 0.5675, 0.5050],
  10082. [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575],
  10083. [0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400],
  10084. [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268],
  10085. [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717],
  10086. [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600],
  10087. [0.6157, 0.4102, 0.8513, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890],
  10088. [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466]],
  10089. device='cuda:0', dtype=torch.float64)
  10090. predictions are: tensor([[ 0.7791, -0.2850, 1.6467, -0.0968, -0.3366, -0.3221, 0.2981, 0.1525],
  10091. [ 0.5286, -0.4658, 1.6174, -0.9831, -0.3909, -0.7397, 0.8328, 0.1719],
  10092. [ 0.4610, -0.5079, 1.6431, -0.6665, -0.5766, -0.4751, 0.6097, 0.1839],
  10093. [ 0.6320, -0.3769, 1.6806, -0.2920, -0.2413, -0.3868, 0.5874, 0.1817],
  10094. [ 0.5876, -0.3599, 0.9087, -1.2984, -0.4308, -1.4225, 0.2108, 0.2846],
  10095. [ 0.6575, -0.3667, 1.7049, -0.3855, -0.1634, -0.2377, 0.4880, 0.2273],
  10096. [ 0.9322, -0.1638, 1.4860, -0.7307, -0.6229, -0.9279, 0.2114, 0.2237],
  10097. [ 0.5982, -0.3987, 1.6030, -0.1148, -0.2391, -0.3081, 0.3186, 0.1958]],
  10098. device='cuda:0', grad_fn=<AddmmBackward>)
  10099. landmarks are: tensor([[[ 0.5303, -0.4440, 1.7095, 0.1390, -0.3402, 0.1159, 0.3353,
  10100. 0.0467],
  10101. [ 0.6158, -0.4249, 1.8654, -0.9002, -0.3229, -0.3537, 0.9667,
  10102. 0.2890],
  10103. [ 0.5778, -0.4393, 1.8018, -0.4614, -0.6693, -0.1381, 0.5490,
  10104. 0.2083],
  10105. [ 0.5828, -0.4682, 1.7031, -0.0497, -0.2458, 0.0818, 0.6381,
  10106. 0.1474],
  10107. [ 0.5633, -0.3883, 0.8648, -1.0288, -0.5192, -1.1081, 0.2776,
  10108. 0.3546],
  10109. [ 0.5652, -0.4325, 1.7388, -0.1429, -0.0688, 0.3469, 0.4970,
  10110. 0.3007],
  10111. [ 0.5581, -0.3912, 1.6460, -0.5230, -0.6173, -0.5923, 0.0681,
  10112. 0.4348],
  10113. [ 0.5178, -0.4480, 1.6748, 0.0620, -0.2767, 0.2083, 0.1067,
  10114. 0.2386]]], device='cuda:0')
  10115. loss_train_step before backward: tensor(0.0397, device='cuda:0', grad_fn=<MseLossBackward>)
  10116. loss_train_step after backward: tensor(0.0397, device='cuda:0', grad_fn=<MseLossBackward>)
  10117. loss_train: 2.7370548360049725
  10118. step: 51
  10119. running loss: 0.05366774188245044
  10120. Train Steps: 51/90 Loss: 0.0537 torch.Size([8, 600, 800])
  10121. torch.Size([8, 8])
  10122. tensor([[0.6184, 0.4079, 0.8350, 0.3700, 0.3675, 0.2883, 0.5312, 0.5783],
  10123. [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051],
  10124. [0.6339, 0.4102, 0.9088, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390],
  10125. [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609],
  10126. [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400],
  10127. [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150],
  10128. [0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494],
  10129. [0.6289, 0.4024, 0.9088, 0.4567, 0.3937, 0.5633, 0.7058, 0.5609]],
  10130. device='cuda:0', dtype=torch.float64)
  10131. predictions are: tensor([[ 0.8567, -0.1718, 1.3547, -0.7763, -0.4955, -0.9446, 0.0881, 0.2427],
  10132. [ 0.5651, -0.4153, 1.5431, -0.3089, -0.4450, -0.5726, 0.3844, 0.1504],
  10133. [ 0.6591, -0.3766, 1.6432, -0.2126, -0.3674, 0.1996, 0.5802, 0.2341],
  10134. [ 0.4659, -0.4568, 1.6246, -1.0287, -0.2502, -1.0492, 0.6522, 0.2027],
  10135. [ 0.4960, -0.4170, 1.1506, -1.3285, -0.1680, -1.3687, 0.4071, 0.2298],
  10136. [ 0.8298, -0.2185, 1.5897, -0.2481, -0.4702, -0.4676, 0.2961, 0.1988],
  10137. [ 0.5850, -0.4000, 1.5434, -0.1594, -0.4074, -0.6108, 0.4559, 0.1936],
  10138. [ 0.7825, -0.3099, 1.6470, -0.3506, -0.3493, 0.2103, 0.7165, 0.2178]],
  10139. device='cuda:0', grad_fn=<AddmmBackward>)
  10140. landmarks are: tensor([[[ 0.5704, -0.4019, 1.5709, -0.5769, -0.5885, -0.9541, 0.1679,
  10141. 0.3854],
  10142. [ 0.6204, -0.4336, 1.8654, -0.0688, -0.6058, -0.5230, 0.6503,
  10143. 0.0472],
  10144. [ 0.6421, -0.3912, 1.9115, -0.0842, -0.4730, 0.1544, 1.1824,
  10145. 0.2035],
  10146. [ 0.6026, -0.4032, 1.7326, -0.7771, -0.3691, -0.8617, 0.9704,
  10147. 0.3050],
  10148. [ 0.5842, -0.3848, 1.2938, -1.2159, -0.2132, -1.4160, 0.5778,
  10149. 0.2083],
  10150. [ 0.6096, -0.4046, 1.8249, -0.0072, -0.6115, -0.3537, 0.6182,
  10151. 0.0928],
  10152. [ 0.6224, -0.4032, 1.5940, 0.2930, -0.5769, -0.2690, 0.8900,
  10153. 0.2516],
  10154. [ 0.6191, -0.4273, 1.9115, -0.1766, -0.4672, 0.3161, 0.9741,
  10155. 0.3050]]], device='cuda:0')
  10156. loss_train_step before backward: tensor(0.0385, device='cuda:0', grad_fn=<MseLossBackward>)
  10157. loss_train_step after backward: tensor(0.0385, device='cuda:0', grad_fn=<MseLossBackward>)
  10158. loss_train: 2.775506906211376
  10159. step: 52
  10160. running loss: 0.05337513281175724
  10161.  
  10162. Train Steps: 52/90 Loss: 0.0534 torch.Size([8, 600, 800])
  10163. torch.Size([8, 8])
  10164. tensor([[0.6207, 0.4110, 0.8738, 0.5000, 0.4800, 0.5633, 0.6300, 0.5433],
  10165. [0.6189, 0.4049, 0.8888, 0.4417, 0.4213, 0.5200, 0.5988, 0.5633],
  10166. [0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804],
  10167. [ nan, nan, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737],
  10168. [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317],
  10169. [ nan, nan, 0.6469, 0.1943, 0.4025, 0.2000, 0.5125, 0.5533],
  10170. [0.6109, 0.4041, 0.6975, 0.3167, 0.3513, 0.3383, 0.5153, 0.5319],
  10171. [0.6260, 0.4153, 0.9000, 0.4533, 0.4025, 0.2633, 0.6223, 0.4967]],
  10172. device='cuda:0', dtype=torch.float64)
  10173. predictions are: tensor([[ 1.2758, -0.0035, 1.7499, -0.0705, -0.1819, 0.2553, 0.6336, 0.2368],
  10174. [ 1.0472, -0.1138, 1.7491, -0.2063, -0.3777, 0.2831, 0.5343, 0.2114],
  10175. [ 1.0555, -0.0865, 1.6605, -0.5979, -0.5562, -0.4504, 0.5985, 0.0961],
  10176. [-0.9829, -1.4003, 1.5955, -1.0695, 0.0910, -1.2272, 0.8067, 0.2447],
  10177. [ 1.1451, -0.0027, 1.6130, 0.1069, -0.5170, -0.5456, 0.3886, 0.1717],
  10178. [-0.4768, -1.0166, 0.8050, -1.3387, -0.3480, -1.3481, 0.2209, 0.2965],
  10179. [ 0.7727, -0.2133, 1.0725, -1.0082, -0.5601, -0.6336, 0.1684, 0.2660],
  10180. [ 1.1011, 0.0046, 1.6445, -0.3508, -0.3602, -0.9335, 0.4613, 0.1539]],
  10181. device='cuda:0', grad_fn=<AddmmBackward>)
  10182. landmarks are: tensor([[[ 0.5811, -0.3878, 1.7499, 0.0236, -0.0688, 0.3161, 0.6240,
  10183. 0.2237],
  10184. [ 0.5726, -0.4159, 1.8192, -0.2459, -0.3402, 0.1159, 0.4797,
  10185. 0.3161],
  10186. [ 0.6071, -0.4119, 1.7788, -0.5153, -0.6000, -0.5692, 0.6586,
  10187. -0.0670],
  10188. [-2.2859, -2.2859, 1.6979, -1.1081, 0.2487, -1.2697, 1.1020,
  10189. 0.3638],
  10190. [ 0.5901, -0.4157, 1.7557, 0.1929, -0.5423, -0.5923, 0.3584,
  10191. 0.1698],
  10192. [-2.2859, -2.2859, 0.7023, -1.3883, -0.4268, -1.3621, 0.0813,
  10193. 0.2699],
  10194. [ 0.5359, -0.4193, 0.9358, -0.8232, -0.6635, -0.7232, 0.0943,
  10195. 0.1710],
  10196. [ 0.6055, -0.3676, 1.8711, -0.1920, -0.4268, -1.0696, 0.5887,
  10197. 0.0081]]], device='cuda:0')
  10198. loss_train_step before backward: tensor(0.1587, device='cuda:0', grad_fn=<MseLossBackward>)
  10199. loss_train_step after backward: tensor(0.1587, device='cuda:0', grad_fn=<MseLossBackward>)
  10200. loss_train: 2.9342208728194237
  10201. step: 53
  10202. running loss: 0.05536265797772497
  10203. Train Steps: 53/90 Loss: 0.0554 torch.Size([8, 600, 800])
  10204. torch.Size([8, 8])
  10205. tensor([[0.6200, 0.4118, 0.8287, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799],
  10206. [0.6307, 0.4045, 0.8025, 0.5833, 0.3775, 0.4867, 0.6892, 0.5459],
  10207. [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364],
  10208. [0.6254, 0.3993, 0.8988, 0.4767, 0.3987, 0.5517, 0.6955, 0.5285],
  10209. [0.6339, 0.4149, 0.8800, 0.5000, 0.3900, 0.5283, 0.7541, 0.5424],
  10210. [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650],
  10211. [0.6109, 0.4015, 0.7668, 0.3639, 0.3513, 0.3667, 0.5200, 0.5641],
  10212. [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183]],
  10213. device='cuda:0', dtype=torch.float64)
  10214. predictions are: tensor([[ 0.3557, -0.4607, 1.3432, -0.7182, -0.4009, -1.1316, 0.1919, 0.2739],
  10215. [ 0.6425, -0.3514, 1.5266, -0.1332, -0.3826, -0.3711, 0.5698, 0.1939],
  10216. [ 0.3685, -0.5062, 1.3575, -1.2130, -0.1406, -1.5512, 0.6332, 0.1769],
  10217. [ 0.6780, -0.3683, 1.8085, -0.1522, -0.3207, 0.1319, 0.8006, 0.1867],
  10218. [ 0.9286, -0.1924, 1.7689, 0.0137, -0.3892, 0.0995, 0.7076, 0.1731],
  10219. [ 0.7650, -0.2433, 1.5615, -0.4926, -0.5390, -0.5534, 0.2164, 0.2737],
  10220. [ 0.5087, -0.3998, 1.2703, -0.8557, -0.5282, -0.6876, 0.2457, 0.2515],
  10221. [ 0.8064, -0.2767, 1.7967, -0.1745, -0.3430, 0.2104, 0.6670, 0.1899]],
  10222. device='cuda:0', grad_fn=<AddmmBackward>)
  10223. landmarks are: tensor([[[ 0.5779, -0.3840, 1.5420, -0.4306, -0.5423, -0.9772, 0.2041,
  10224. 0.3928],
  10225. [ 0.6273, -0.4177, 1.4208, 0.4085, -0.5423, -0.0380, 0.8973,
  10226. 0.2356],
  10227. [ 0.6158, -0.3960, 1.4092, -1.2774, -0.2074, -1.1851, 0.8491,
  10228. 0.1917],
  10229. [ 0.6026, -0.4417, 1.8654, -0.0842, -0.4441, 0.2622, 0.9265,
  10230. 0.1554],
  10231. [ 0.6421, -0.3695, 1.7788, 0.0236, -0.4845, 0.1544, 1.1971,
  10232. 0.2196],
  10233. [ 0.5757, -0.3917, 1.7095, -0.4768, -0.6346, -0.4229, 0.3931,
  10234. 0.3238],
  10235. [ 0.5357, -0.4315, 1.2557, -0.6051, -0.6635, -0.5923, 0.1159,
  10236. 0.3198],
  10237. [ 0.5934, -0.4276, 1.8192, -0.1458, -0.4499, 0.3777, 0.6702,
  10238. 0.1082]]], device='cuda:0')
  10239. loss_train_step before backward: tensor(0.0291, device='cuda:0', grad_fn=<MseLossBackward>)
  10240. loss_train_step after backward: tensor(0.0291, device='cuda:0', grad_fn=<MseLossBackward>)
  10241. loss_train: 2.9632861856371164
  10242. step: 54
  10243. running loss: 0.054875670104391046
  10244. Train Steps: 54/90 Loss: 0.0549 torch.Size([8, 600, 800])
  10245. torch.Size([8, 8])
  10246. tensor([[0.6270, 0.4267, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183],
  10247. [0.6151, 0.4085, 0.8750, 0.4367, 0.3887, 0.4367, 0.5066, 0.5846],
  10248. [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749],
  10249. [0.6166, 0.4008, 0.8563, 0.5667, 0.4388, 0.4933, 0.5575, 0.5567],
  10250. [0.6145, 0.4008, 0.8750, 0.5383, 0.3975, 0.4650, 0.5563, 0.5533],
  10251. [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274],
  10252. [ nan, nan, 0.8300, 0.3150, 0.3588, 0.3383, 0.5208, 0.5194],
  10253. [0.6213, 0.4001, 0.7712, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183]],
  10254. device='cuda:0', dtype=torch.float64)
  10255. predictions are: tensor([[ 0.4483, -0.4180, 1.0823, -1.1483, -0.1529, -1.1096, 0.4683, 0.3558],
  10256. [ 0.5588, -0.3981, 1.7192, -0.2750, -0.4433, -0.2098, 0.4805, 0.2380],
  10257. [ 0.4985, -0.4167, 1.3104, -0.8960, -0.5438, -0.5640, 0.4597, 0.2455],
  10258. [ 0.8661, -0.2295, 1.7761, 0.4190, -0.2540, 0.1397, 0.5771, 0.2036],
  10259. [ 0.8158, -0.2632, 1.8079, 0.3212, -0.3979, 0.0774, 0.5858, 0.1922],
  10260. [ 0.8015, -0.2651, 1.9091, -0.0199, -0.4642, -0.2573, 0.7678, 0.1320],
  10261. [ 0.0030, -0.7220, 1.4453, -0.7867, -0.5837, -0.7954, 0.2402, 0.1659],
  10262. [ 0.5133, -0.4323, 1.2798, -1.3046, -0.2583, -1.3138, 0.6630, 0.1580]],
  10263. device='cuda:0', grad_fn=<AddmmBackward>)
  10264. landmarks are: tensor([[[ 0.6101, -0.3152, 1.0166, -0.7540, -0.0226, -1.4468, 0.2823,
  10265. 0.5702],
  10266. [ 0.5553, -0.3992, 1.7557, -0.2690, -0.4903, -0.2690, 0.0542,
  10267. 0.4145],
  10268. [ 0.5419, -0.4160, 1.1810, -0.8939, -0.6808, -0.7463, 0.2485,
  10269. 0.3694],
  10270. [ 0.5620, -0.4346, 1.6691, 0.3315, -0.2594, -0.0072, 0.2891,
  10271. 0.2853],
  10272. [ 0.5526, -0.4347, 1.7557, 0.2006, -0.4499, -0.1381, 0.2834,
  10273. 0.2699],
  10274. [ 0.6108, -0.4008, 1.9088, -0.0253, -0.5769, -0.3075, 0.7905,
  10275. 0.1499],
  10276. [-2.2859, -2.2859, 1.5478, -0.8309, -0.6289, -0.7232, 0.1198,
  10277. 0.1133],
  10278. [ 0.5838, -0.4379, 1.2764, -1.3082, -0.2824, -1.4545, 0.4162,
  10279. 0.1082]]], device='cuda:0')
  10280. loss_train_step before backward: tensor(0.1432, device='cuda:0', grad_fn=<MseLossBackward>)
  10281. loss_train_step after backward: tensor(0.1432, device='cuda:0', grad_fn=<MseLossBackward>)
  10282. loss_train: 3.106526607647538
  10283. step: 55
  10284. running loss: 0.05648230195722797
  10285. Train Steps: 55/90 Loss: 0.0565 torch.Size([8, 600, 800])
  10286. torch.Size([8, 8])
  10287. tensor([[0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183],
  10288. [0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317],
  10289. [0.6260, 0.4153, 0.9000, 0.4533, 0.4025, 0.2633, 0.6223, 0.4967],
  10290. [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717],
  10291. [0.6250, 0.3993, 0.9138, 0.4333, 0.3763, 0.5217, 0.6995, 0.5320],
  10292. [0.6193, 0.4050, 0.7313, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656],
  10293. [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837],
  10294. [ nan, nan, 0.7725, 0.2611, 0.3675, 0.2733, 0.5413, 0.5167]],
  10295. device='cuda:0', dtype=torch.float64)
  10296. predictions are: tensor([[ 1.0553e+00, -1.5796e-01, 2.0803e+00, 3.1226e-01, -3.8757e-01,
  10297. 6.7862e-01, 8.9026e-01, 1.5629e-01],
  10298. [ 1.4193e-01, -6.1169e-01, 1.3132e+00, -9.7362e-01, -2.6433e-01,
  10299. -1.0044e+00, 5.2136e-01, 2.8896e-01],
  10300. [ 9.9363e-01, -7.4630e-02, 1.8583e+00, 3.2327e-02, -3.5001e-01,
  10301. -7.8548e-01, 6.0284e-01, 1.4803e-01],
  10302. [ 5.2145e-01, -3.7143e-01, 1.2435e+00, -8.4906e-01, -4.3652e-01,
  10303. -8.5257e-01, 4.4860e-01, 3.1134e-01],
  10304. [ 1.3241e+00, -6.1965e-04, 2.0147e+00, 5.9596e-02, -4.7098e-01,
  10305. 5.4668e-01, 9.9321e-01, 1.6055e-01],
  10306. [ 4.0170e-01, -4.5806e-01, 1.1629e+00, -9.4310e-01, -3.3796e-01,
  10307. -9.8315e-01, 4.4373e-01, 2.4660e-01],
  10308. [ 5.9641e-01, -3.5358e-01, 1.0719e+00, -7.0736e-01, -5.0569e-01,
  10309. -7.0185e-01, 2.7299e-01, 3.3588e-01],
  10310. [-1.0682e+00, -1.4444e+00, 1.2887e+00, -8.3251e-01, -4.0688e-01,
  10311. -9.9458e-01, 2.4550e-01, 2.2370e-01]], device='cuda:0',
  10312. grad_fn=<AddmmBackward>)
  10313. landmarks are: tensor([[[ 0.5934, -0.4276, 1.8192, -0.1458, -0.4499, 0.3777, 0.6702,
  10314. 0.1082],
  10315. [ 0.5849, -0.3836, 1.2649, -1.2236, -0.3460, -1.2313, 0.4508,
  10316. 0.1698],
  10317. [ 0.6055, -0.3676, 1.8711, -0.1920, -0.4268, -1.0696, 0.5887,
  10318. 0.0081],
  10319. [ 0.5746, -0.3882, 1.1436, -1.2005, -0.4903, -1.0157, 0.4393,
  10320. 0.3546],
  10321. [ 0.6010, -0.4417, 1.9346, -0.2844, -0.5480, 0.1236, 0.9448,
  10322. 0.1715],
  10323. [ 0.5746, -0.4153, 1.0917, -1.1620, -0.4037, -1.3082, 0.3234,
  10324. 0.3267],
  10325. [ 0.5629, -0.3871, 0.7243, -0.9581, -0.5827, -0.9849, 0.1288,
  10326. 0.4103],
  10327. [-2.2859, -2.2859, 1.2820, -1.0801, -0.5885, -1.0234, 0.2141,
  10328. 0.1005]]], device='cuda:0')
  10329. loss_train_step before backward: tensor(0.0860, device='cuda:0', grad_fn=<MseLossBackward>)
  10330. loss_train_step after backward: tensor(0.0860, device='cuda:0', grad_fn=<MseLossBackward>)
  10331.  
  10332. loss_train: 3.1925710681825876
  10333. step: 56
  10334. running loss: 0.05701019764611764
  10335. Train Steps: 56/90 Loss: 0.0570 torch.Size([8, 600, 800])
  10336. torch.Size([8, 8])
  10337. tensor([[0.6272, 0.4045, 0.8538, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649],
  10338. [0.6270, 0.4267, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183],
  10339. [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383],
  10340. [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413],
  10341. [0.6274, 0.4003, 0.8638, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661],
  10342. [0.6144, 0.4032, 0.8563, 0.3283, 0.3525, 0.4200, 0.5775, 0.5583],
  10343. [0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133],
  10344. [0.6132, 0.4037, 0.6963, 0.2217, 0.4100, 0.1950, 0.5395, 0.5175]],
  10345. device='cuda:0', dtype=torch.float64)
  10346. predictions are: tensor([[ 0.7179, -0.3238, 1.7458, 0.3163, -0.5806, -0.1069, 0.5669, 0.1582],
  10347. [ 0.0706, -0.6531, 0.9820, -1.0993, -0.2456, -1.0041, 0.3436, 0.3978],
  10348. [ 0.5253, -0.4076, 1.7337, 0.1074, -0.3820, 0.2137, 0.4023, 0.2355],
  10349. [ 0.5145, -0.4474, 2.0043, -0.2206, -0.2732, -0.6411, 0.8709, 0.1705],
  10350. [ 0.6235, -0.4090, 1.7221, 0.4188, -0.5392, 0.1507, 0.6613, 0.1973],
  10351. [ 0.6203, -0.3780, 1.6743, -0.5646, -0.6705, -0.1017, 0.4486, 0.2820],
  10352. [ 0.3270, -0.5656, 1.5257, -1.1182, -0.2016, -1.1584, 0.8714, 0.2187],
  10353. [ 0.1576, -0.6544, 1.0196, -1.1413, -0.4234, -1.1670, 0.3124, 0.2487]],
  10354. device='cuda:0', grad_fn=<AddmmBackward>)
  10355. landmarks are: tensor([[[ 6.1085e-01, -4.1771e-01, 1.6575e+00, 4.3926e-01, -5.5381e-01,
  10356. -2.4588e-01, 4.8055e-01, -1.3847e-01],
  10357. [ 6.1010e-01, -3.1524e-01, 1.0166e+00, -7.5396e-01, -2.2633e-02,
  10358. -1.4468e+00, 2.8226e-01, 5.7018e-01],
  10359. [ 5.3603e-01, -4.6490e-01, 1.6517e+00, 4.6189e-04, -2.8245e-01,
  10360. -6.8822e-02, 2.3086e-01, 2.0046e-01],
  10361. [ 6.2236e-01, -4.3453e-01, 1.9404e+00, -2.9207e-01, -3.1709e-01,
  10362. -8.7714e-01, 1.0655e+00, 2.1421e-01],
  10363. [ 6.1201e-01, -4.3711e-01, 1.7037e+00, 4.7005e-01, -5.8268e-01,
  10364. -2.2633e-02, 5.3538e-01, -1.3313e-01],
  10365. [ 5.5196e-01, -4.2371e-01, 1.6691e+00, -7.6936e-01, -6.5774e-01,
  10366. -3.4596e-01, 3.8152e-01, 2.9299e-01],
  10367. [ 6.0855e-01, -4.0839e-01, 1.5536e+00, -1.1466e+00, -7.4596e-02,
  10368. -1.4853e+00, 6.2979e-01, 8.5142e-02],
  10369. [ 5.4648e-01, -4.2140e-01, 9.3002e-01, -1.2620e+00, -3.9215e-01,
  10370. -1.3852e+00, 2.0618e-01, 1.0428e-01]]], device='cuda:0')
  10371. loss_train_step before backward: tensor(0.0330, device='cuda:0', grad_fn=<MseLossBackward>)
  10372. loss_train_step after backward: tensor(0.0330, device='cuda:0', grad_fn=<MseLossBackward>)
  10373. loss_train: 3.2255628500133753
  10374. step: 57
  10375. running loss: 0.056588821930059215
  10376. Train Steps: 57/90 Loss: 0.0566 torch.Size([8, 600, 800])
  10377. torch.Size([8, 8])
  10378. tensor([[0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285],
  10379. [0.6222, 0.4169, 0.8638, 0.5650, 0.4313, 0.4783, 0.5637, 0.5633],
  10380. [0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272],
  10381. [0.6277, 0.4118, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938],
  10382. [0.6201, 0.4151, 0.8588, 0.5467, 0.3700, 0.3950, 0.5637, 0.5933],
  10383. [0.6266, 0.4067, 0.8588, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267],
  10384. [0.6276, 0.4120, 0.8738, 0.3133, 0.4225, 0.2217, 0.6203, 0.4892],
  10385. [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235]],
  10386. device='cuda:0', dtype=torch.float64)
  10387. predictions are: tensor([[ 0.6005, -0.4291, 1.6331, 0.1981, -0.4927, 0.1141, 0.6502, 0.2158],
  10388. [ 0.3800, -0.5244, 1.5364, 0.2919, -0.3730, 0.0861, 0.3863, 0.3102],
  10389. [ 0.3911, -0.4867, 1.4536, -0.7373, -0.5116, -0.8875, 0.3204, 0.2352],
  10390. [ 0.3081, -0.5687, 1.7045, -0.6232, -0.4642, -0.9397, 0.6238, 0.1902],
  10391. [ 0.3725, -0.4843, 1.4808, 0.0960, -0.5485, -0.2804, 0.3564, 0.3270],
  10392. [ 0.3914, -0.4865, 1.5987, -1.1421, -0.3147, -0.9816, 0.8374, 0.2485],
  10393. [ 0.4910, -0.4554, 1.6362, -0.9180, -0.3677, -1.0994, 0.7497, 0.2050],
  10394. [ 0.3051, -0.6135, 1.6622, -0.1353, -0.3918, 0.2559, 0.6588, 0.2314]],
  10395. device='cuda:0', grad_fn=<AddmmBackward>)
  10396. landmarks are: tensor([[[ 6.5365e-01, -3.9601e-01, 1.6517e+00, 3.1609e-01, -4.9607e-01,
  10397. 4.6189e-04, 7.6203e-01, 1.5543e-01],
  10398. [ 5.8799e-01, -3.6051e-01, 1.7037e+00, 3.2379e-01, -2.9400e-01,
  10399. -7.6520e-02, 3.1801e-01, 3.1609e-01],
  10400. [ 5.7113e-01, -4.0146e-01, 1.6979e+00, -6.7698e-01, -5.3649e-01,
  10401. -1.0619e+00, 1.7122e-01, 1.4937e-01],
  10402. [ 6.1351e-01, -3.8406e-01, 1.8654e+00, -5.1532e-01, -4.6143e-01,
  10403. -1.0619e+00, 6.1946e-01, -4.8817e-03],
  10404. [ 5.7852e-01, -3.6867e-01, 1.6806e+00, 2.3911e-01, -5.7691e-01,
  10405. -4.6143e-01, 3.1801e-01, 4.5466e-01],
  10406. [ 6.0837e-01, -4.0762e-01, 1.6806e+00, -9.6182e-01, -2.9977e-01,
  10407. -9.6952e-01, 6.3557e-01, 1.4673e-01],
  10408. [ 6.1282e-01, -3.8283e-01, 1.7499e+00, -8.3865e-01, -3.3441e-01,
  10409. -1.2620e+00, 5.7925e-01, -2.6256e-02],
  10410. [ 5.7625e-01, -4.7064e-01, 1.7754e+00, -9.8417e-02, -3.6803e-01,
  10411. 2.3803e-01, 6.2770e-01, 1.3223e-01]]], device='cuda:0')
  10412. loss_train_step before backward: tensor(0.0179, device='cuda:0', grad_fn=<MseLossBackward>)
  10413. loss_train_step after backward: tensor(0.0179, device='cuda:0', grad_fn=<MseLossBackward>)
  10414. loss_train: 3.2434706911444664
  10415. step: 58
  10416. running loss: 0.05592190846800804
  10417. Train Steps: 58/90 Loss: 0.0559 torch.Size([8, 600, 800])
  10418. torch.Size([8, 8])
  10419. tensor([[0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202],
  10420. [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650],
  10421. [0.6329, 0.4196, 0.9238, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748],
  10422. [0.6289, 0.4019, 0.8113, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332],
  10423. [0.6175, 0.3957, 0.8700, 0.4817, 0.4662, 0.5133, 0.5800, 0.5517],
  10424. [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383],
  10425. [0.6113, 0.4088, 0.6859, 0.2208, 0.4363, 0.1700, 0.5188, 0.5533],
  10426. [0.6125, 0.3999, 0.8750, 0.4883, 0.4750, 0.4700, 0.5533, 0.5617]],
  10427. device='cuda:0', dtype=torch.float64)
  10428. predictions are: tensor([[ 0.6097, -0.3595, 1.6705, -0.0305, -0.4489, -0.0061, 0.5127, 0.2028],
  10429. [ 0.6639, -0.3332, 1.6342, -0.4861, -0.7198, -0.5016, 0.3668, 0.2877],
  10430. [ 0.5705, -0.3734, 1.9512, -0.2589, -0.3449, -1.1658, 0.9691, 0.1807],
  10431. [ 0.6820, -0.3269, 1.5413, -0.0517, -0.5332, -0.0984, 0.6546, 0.2289],
  10432. [ 0.2860, -0.6007, 1.7352, -0.1885, -0.3268, -0.1205, 0.5123, 0.2239],
  10433. [ 0.6428, -0.3875, 1.7646, -0.3212, -0.3645, 0.1299, 0.7426, 0.2294],
  10434. [-1.0394, -1.4764, 0.8833, -1.3383, -0.4541, -1.6894, 0.2561, 0.2452],
  10435. [ 0.6094, -0.3865, 1.7440, -0.0939, -0.2608, -0.1809, 0.4635, 0.2568]],
  10436. device='cuda:0', grad_fn=<AddmmBackward>)
  10437. landmarks are: tensor([[[ 0.5614, -0.4586, 1.7587, 0.0641, -0.2998, 0.1712, 0.4958,
  10438. 0.1170],
  10439. [ 0.5757, -0.3917, 1.7095, -0.4768, -0.6346, -0.4229, 0.3931,
  10440. 0.3238],
  10441. [ 0.6372, -0.3479, 1.9808, -0.1689, -0.3171, -0.9310, 1.1057,
  10442. 0.3692],
  10443. [ 0.6191, -0.4297, 1.4612, 0.2391, -0.4961, 0.0313, 1.1166,
  10444. 0.1768],
  10445. [ 0.5662, -0.4581, 1.7326, -0.0611, -0.1323, 0.0851, 0.3931,
  10446. 0.2622],
  10447. [ 0.5664, -0.4321, 1.8249, -0.2074, -0.2651, 0.4162, 0.5663,
  10448. 0.2006],
  10449. [ 0.5377, -0.3978, 0.8824, -1.2663, -0.2709, -1.5007, 0.1102,
  10450. 0.2699],
  10451. [ 0.5432, -0.4388, 1.7557, -0.0303, -0.0919, -0.1150, 0.2699,
  10452. 0.3087]]], device='cuda:0')
  10453. loss_train_step before backward: tensor(0.0746, device='cuda:0', grad_fn=<MseLossBackward>)
  10454. loss_train_step after backward: tensor(0.0746, device='cuda:0', grad_fn=<MseLossBackward>)
  10455. loss_train: 3.3181116357445717
  10456. step: 59
  10457. running loss: 0.05623918026685715
  10458. Train Steps: 59/90 Loss: 0.0562 torch.Size([8, 600, 800])
  10459. torch.Size([8, 8])
  10460. tensor([[0.6043, 0.4022, 0.6887, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136],
  10461. [0.6193, 0.4034, 0.7757, 0.2347, 0.3733, 0.2919, 0.5930, 0.4926],
  10462. [0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411],
  10463. [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683],
  10464. [0.6201, 0.4064, 0.8688, 0.5050, 0.4225, 0.5100, 0.6138, 0.5500],
  10465. [0.6147, 0.4026, 0.6600, 0.2467, 0.4088, 0.2150, 0.5489, 0.5773],
  10466. [ nan, nan, 0.6412, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700],
  10467. [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837]],
  10468. device='cuda:0', dtype=torch.float64)
  10469. predictions are: tensor([[ 0.5722, -0.3972, 1.2615, -0.8737, -0.5309, -0.9137, 0.4895, 0.1950],
  10470. [ 0.2688, -0.6062, 1.5072, -0.8471, -0.5606, -0.7359, 0.5700, 0.1372],
  10471. [ 0.5582, -0.3715, 1.6745, -0.5818, -0.3719, -1.0065, 0.5898, 0.1881],
  10472. [ 0.4482, -0.5203, 2.0914, 0.4037, -0.2245, 0.3096, 0.6900, 0.1714],
  10473. [ 0.6675, -0.3792, 2.0147, 0.5004, -0.4290, 0.4757, 0.7432, 0.1545],
  10474. [ 0.7220, -0.3007, 1.1824, -0.7761, -0.4194, -1.0112, 0.4273, 0.3017],
  10475. [-1.0870, -1.5275, 1.1445, -0.9036, -0.3555, -1.3402, 0.3323, 0.2497],
  10476. [ 0.6270, -0.3712, 1.2397, -0.6168, -0.5985, -0.6370, 0.3945, 0.3137]],
  10477. device='cuda:0', grad_fn=<AddmmBackward>)
  10478. landmarks are: tensor([[[ 0.5053, -0.4281, 0.8954, -1.3698, -0.5423, -1.1389, 0.2453,
  10479. 0.0862],
  10480. [ 0.5746, -0.4225, 1.2968, -1.2019, -0.5615, -0.9374, 0.4533,
  10481. -0.0108],
  10482. [ 0.5713, -0.4321, 1.4975, -0.8134, -0.3094, -1.3345, 0.3779,
  10483. 0.2134],
  10484. [ 0.5359, -0.4370, 1.7095, -0.0303, -0.0804, -0.0380, 0.1044,
  10485. 0.3392],
  10486. [ 0.5784, -0.4088, 1.7268, 0.0467, -0.3344, 0.0697, 0.5490,
  10487. 0.2545],
  10488. [ 0.5532, -0.4264, 0.7626, -1.1466, -0.3979, -1.2928, 0.2494,
  10489. 0.3808],
  10490. [-2.2859, -2.2859, 0.6760, -1.4083, -0.3286, -1.4160, 0.2487,
  10491. 0.3469],
  10492. [ 0.5629, -0.3871, 0.7243, -0.9581, -0.5827, -0.9849, 0.1288,
  10493. 0.4103]]], device='cuda:0')
  10494. loss_train_step before backward: tensor(0.0950, device='cuda:0', grad_fn=<MseLossBackward>)
  10495.  
  10496. loss_train_step after backward: tensor(0.0950, device='cuda:0', grad_fn=<MseLossBackward>)
  10497. loss_train: 3.413125194609165
  10498. step: 60
  10499. running loss: 0.05688541991015275
  10500. Train Steps: 60/90 Loss: 0.0569 torch.Size([8, 600, 800])
  10501. torch.Size([8, 8])
  10502. tensor([[0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507],
  10503. [0.6286, 0.4055, 0.9000, 0.4717, 0.3763, 0.4683, 0.7018, 0.5494],
  10504. [0.6205, 0.4016, 0.8350, 0.2717, 0.3987, 0.2550, 0.5787, 0.5133],
  10505. [0.6329, 0.4196, 0.9238, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748],
  10506. [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466],
  10507. [0.6060, 0.3924, 0.8450, 0.5717, 0.4200, 0.5217, 0.5253, 0.4752],
  10508. [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012],
  10509. [0.6229, 0.4198, 0.7662, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783]],
  10510. device='cuda:0', dtype=torch.float64)
  10511. predictions are: tensor([[ 0.4477, -0.4989, 1.0241, -1.1866, -0.4538, -1.1541, 0.3545, 0.2349],
  10512. [ 0.5304, -0.4797, 1.9840, -0.0666, -0.5513, -0.0591, 0.9308, 0.1303],
  10513. [ 0.3024, -0.5637, 1.5836, -0.9812, -0.3989, -1.0223, 0.4467, 0.1653],
  10514. [ 0.1767, -0.6606, 1.9765, -0.0760, -0.2892, -0.9039, 0.8481, 0.2020],
  10515. [ 0.3131, -0.5527, 1.3676, -0.6091, -0.6676, -0.2343, 0.2590, 0.2547],
  10516. [ 0.3597, -0.5725, 1.6503, 0.4191, -0.3494, 0.0670, 0.4046, 0.2217],
  10517. [ 0.6185, -0.4473, 1.8138, 0.3166, -0.5707, -0.0937, 0.6552, 0.1352],
  10518. [ 0.1686, -0.6116, 1.1855, -1.0051, -0.2213, -1.1281, 0.3396, 0.3605]],
  10519. device='cuda:0', grad_fn=<AddmmBackward>)
  10520. landmarks are: tensor([[[ 0.5185, -0.4252, 0.9647, -1.2928, -0.4788, -1.2390, 0.2617,
  10521. 0.2576],
  10522. [ 0.6174, -0.4129, 1.8711, -0.1073, -0.5480, -0.1227, 0.9558,
  10523. 0.2516],
  10524. [ 0.5800, -0.4312, 1.5709, -1.0311, -0.4441, -1.1081, 0.3873,
  10525. 0.0851],
  10526. [ 0.6372, -0.3479, 1.9808, -0.1689, -0.3171, -0.9310, 1.1057,
  10527. 0.3692],
  10528. [ 0.5076, -0.4443, 1.2337, -0.5023, -0.6808, -0.3614, 0.0866,
  10529. 0.2386],
  10530. [ 0.5132, -0.4736, 1.6171, 0.3546, -0.3460, 0.1236, 0.1404,
  10531. -0.0911],
  10532. [ 0.6249, -0.4352, 1.8018, 0.2545, -0.6173, -0.1997, 0.6401,
  10533. 0.0291],
  10534. [ 0.5912, -0.3467, 1.2533, -1.0388, -0.1150, -1.3005, 0.3353,
  10535. 0.3854]]], device='cuda:0')
  10536. loss_train_step before backward: tensor(0.0200, device='cuda:0', grad_fn=<MseLossBackward>)
  10537. loss_train_step after backward: tensor(0.0200, device='cuda:0', grad_fn=<MseLossBackward>)
  10538. loss_train: 3.433107813820243
  10539. step: 61
  10540. running loss: 0.05628045596426628
  10541. Train Steps: 61/90 Loss: 0.0563 torch.Size([8, 600, 800])
  10542. torch.Size([8, 8])
  10543. tensor([[0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355],
  10544. [0.6202, 0.4064, 0.7879, 0.2179, 0.4567, 0.1725, 0.5955, 0.5478],
  10545. [0.6147, 0.4026, 0.6600, 0.2467, 0.4088, 0.2150, 0.5489, 0.5773],
  10546. [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153],
  10547. [0.6147, 0.4112, 0.7988, 0.3200, 0.3775, 0.2767, 0.5150, 0.5550],
  10548. [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183],
  10549. [0.6151, 0.4085, 0.8750, 0.4367, 0.3887, 0.4367, 0.5066, 0.5846],
  10550. [0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689]],
  10551. device='cuda:0', dtype=torch.float64)
  10552. predictions are: tensor([[ 0.1825, -0.7061, 1.7562, -0.5810, -0.2417, -0.6851, 0.8891, 0.1766],
  10553. [ 0.5891, -0.4146, 1.3659, -0.8705, -0.2362, -1.3972, 0.5131, 0.2131],
  10554. [ 0.7430, -0.3105, 1.0501, -0.8250, -0.3662, -1.2392, 0.3054, 0.2914],
  10555. [ 0.6278, -0.3860, 1.5139, -0.6035, -0.6098, -0.5631, 0.4918, 0.1655],
  10556. [ 0.0770, -0.7091, 1.4412, -0.5819, -0.5632, -0.9197, 0.1810, 0.1949],
  10557. [ 0.3116, -0.6728, 1.9799, 0.2008, -0.3828, 0.4249, 0.8613, 0.1679],
  10558. [-0.1072, -0.8606, 1.7340, -0.0695, -0.4591, -0.2890, 0.4195, 0.2378],
  10559. [ 0.4073, -0.5107, 1.4580, -0.4866, -0.6220, -0.3523, 0.3635, 0.2498]],
  10560. device='cuda:0', grad_fn=<AddmmBackward>)
  10561. landmarks are: tensor([[[ 0.6224, -0.4105, 1.9173, -0.7771, -0.1030, -0.7308, 1.1532,
  10562. 0.1875],
  10563. [ 0.5787, -0.4086, 1.3535, -1.2794, -0.1764, -1.4891, 0.4645,
  10564. 0.2442],
  10565. [ 0.5532, -0.4264, 0.7626, -1.1466, -0.3979, -1.2928, 0.2494,
  10566. 0.3808],
  10567. [ 0.5614, -0.4514, 1.3252, -0.7240, -0.6237, -0.7215, 0.4299,
  10568. 0.0942],
  10569. [ 0.5532, -0.3864, 1.4035, -0.8079, -0.5423, -1.0080, 0.0928,
  10570. 0.2776],
  10571. [ 0.5934, -0.4276, 1.8192, -0.1458, -0.4499, 0.3777, 0.6702,
  10572. 0.1082],
  10573. [ 0.5553, -0.3992, 1.7557, -0.2690, -0.4903, -0.2690, 0.0542,
  10574. 0.4145],
  10575. [ 0.5327, -0.4381, 1.2880, -0.6308, -0.6866, -0.5307, 0.2658,
  10576. 0.3417]]], device='cuda:0')
  10577. loss_train_step before backward: tensor(0.0419, device='cuda:0', grad_fn=<MseLossBackward>)
  10578. loss_train_step after backward: tensor(0.0419, device='cuda:0', grad_fn=<MseLossBackward>)
  10579. loss_train: 3.4750487077981234
  10580. step: 62
  10581. running loss: 0.056049172706421346
  10582. Train Steps: 62/90 Loss: 0.0560 torch.Size([8, 600, 800])
  10583. torch.Size([8, 8])
  10584. tensor([[0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5787, 0.5600],
  10585. [0.6264, 0.4049, 0.8988, 0.4633, 0.3813, 0.4983, 0.6326, 0.4843],
  10586. [0.6203, 0.4021, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405],
  10587. [ nan, nan, 0.6412, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700],
  10588. [0.6064, 0.4019, 0.8650, 0.4517, 0.4037, 0.5367, 0.5703, 0.5609],
  10589. [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833],
  10590. [0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266],
  10591. [0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272]],
  10592. device='cuda:0', dtype=torch.float64)
  10593. predictions are: tensor([[ 0.5372, -0.4427, 1.7632, 0.0029, -0.1842, -0.0286, 0.4543, 0.2477],
  10594. [ 0.7022, -0.3753, 2.0274, -0.0909, -0.5449, -0.0211, 0.7920, 0.0436],
  10595. [ 0.8202, -0.2692, 1.8586, 0.2130, -0.6447, -0.3823, 0.6089, 0.0845],
  10596. [-1.2660, -1.6953, 0.9615, -1.1883, -0.3203, -1.5984, 0.1894, 0.2762],
  10597. [ 0.4728, -0.4804, 1.6838, -0.2210, -0.3768, 0.2081, 0.5723, 0.2581],
  10598. [ 0.0046, -0.7779, 1.0222, -0.8906, -0.4931, -1.0298, 0.2306, 0.3229],
  10599. [ 1.1376, -0.0465, 1.2655, -0.9522, -0.3285, -1.1190, 0.5421, 0.3597],
  10600. [ 0.4906, -0.4482, 1.6257, -0.5826, -0.4980, -1.0188, 0.3590, 0.1793]],
  10601. device='cuda:0', grad_fn=<AddmmBackward>)
  10602. landmarks are: tensor([[[ 0.5505, -0.4207, 1.7095, -0.0534, -0.0509, 0.1050, 0.3873,
  10603. 0.3007],
  10604. [ 0.6075, -0.4159, 1.8654, -0.1458, -0.5249, 0.0159, 0.6359,
  10605. -0.0490],
  10606. [ 0.5791, -0.4289, 1.7694, 0.0379, -0.5923, -0.4927, 0.4126,
  10607. 0.2107],
  10608. [-2.2859, -2.2859, 0.6760, -1.4083, -0.3286, -1.4160, 0.2487,
  10609. 0.3469],
  10610. [ 0.5151, -0.4296, 1.7095, -0.1997, -0.4210, 0.1929, 0.3484,
  10611. 0.3047],
  10612. [ 0.5614, -0.4080, 0.7741, -0.8848, -0.5423, -0.9156, 0.3584,
  10613. 0.4085],
  10614. [ 0.5746, -0.4021, 1.0801, -1.1312, -0.3229, -1.1081, 0.4803,
  10615. 0.6084],
  10616. [ 0.5711, -0.4015, 1.6979, -0.6770, -0.5365, -1.0619, 0.1712,
  10617. 0.1494]]], device='cuda:0')
  10618. loss_train_step before backward: tensor(0.0488, device='cuda:0', grad_fn=<MseLossBackward>)
  10619. loss_train_step after backward: tensor(0.0488, device='cuda:0', grad_fn=<MseLossBackward>)
  10620. loss_train: 3.523830173537135
  10621. step: 63
  10622. running loss: 0.055933812278367225
  10623. Train Steps: 63/90 Loss: 0.0559 torch.Size([8, 600, 800])
  10624. torch.Size([8, 8])
  10625. tensor([[0.6153, 0.4117, 0.8688, 0.5167, 0.4895, 0.5647, 0.5524, 0.5136],
  10626. [0.6197, 0.4118, 0.8688, 0.5517, 0.4037, 0.5233, 0.5875, 0.5600],
  10627. [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183],
  10628. [0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133],
  10629. [0.6128, 0.4084, 0.8738, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397],
  10630. [0.6115, 0.3998, 0.7063, 0.2383, 0.4037, 0.1950, 0.5320, 0.4993],
  10631. [0.6273, 0.4143, 0.8750, 0.5700, 0.3987, 0.4717, 0.6013, 0.5467],
  10632. [0.6161, 0.4024, 0.8838, 0.4583, 0.3688, 0.3733, 0.5311, 0.5344]],
  10633. device='cuda:0', dtype=torch.float64)
  10634. predictions are: tensor([[ 0.2926, -0.6040, 1.6742, -0.2913, -0.0972, -0.4117, 0.4652, 0.2531],
  10635. [ 0.4361, -0.5205, 1.5725, 0.0129, -0.3745, -0.1130, 0.4582, 0.2891],
  10636. [ 0.4407, -0.5717, 1.6980, -0.5028, -0.3800, 0.0268, 0.6957, 0.2434],
  10637. [ 0.6617, -0.4029, 1.7236, -0.1642, -0.4997, -0.3592, 0.7090, 0.1786],
  10638. [ 0.1602, -0.6668, 1.5842, -0.4179, -0.6784, -0.8125, 0.2578, 0.2007],
  10639. [ 0.6501, -0.3948, 0.9773, -1.4541, -0.3942, -1.6491, 0.1857, 0.2270],
  10640. [ 0.2614, -0.6123, 1.6131, -0.0798, -0.4499, -0.5095, 0.4385, 0.2353],
  10641. [ 0.2661, -0.6166, 1.6872, -0.4090, -0.5893, -0.7739, 0.3292, 0.1672]],
  10642. device='cuda:0', grad_fn=<AddmmBackward>)
  10643. landmarks are: tensor([[[ 0.5564, -0.3842, 1.7268, 0.1005, -0.0250, 0.3225, 0.2658,
  10644. 0.0862],
  10645. [ 0.5762, -0.3840, 1.7268, 0.2622, -0.4210, 0.1313, 0.4277,
  10646. 0.3007],
  10647. [ 0.5934, -0.4276, 1.8192, -0.1458, -0.4499, 0.3777, 0.6702,
  10648. 0.1082],
  10649. [ 0.6148, -0.4130, 1.6864, 0.1698, -0.5307, -0.1150, 0.6125,
  10650. 0.0851],
  10651. [ 0.5443, -0.3994, 1.7499, -0.1227, -0.6173, -0.5769, 0.0049,
  10652. 0.2071],
  10653. [ 0.5384, -0.4393, 0.9762, -1.1851, -0.4210, -1.3852, 0.1712,
  10654. 0.0201],
  10655. [ 0.6115, -0.3724, 1.7557, 0.3469, -0.4441, -0.1073, 0.4912,
  10656. 0.2391],
  10657. [ 0.5598, -0.4273, 1.7961, -0.1689, -0.5827, -0.5615, 0.1671,
  10658. 0.1824]]], device='cuda:0')
  10659. loss_train_step before backward: tensor(0.0479, device='cuda:0', grad_fn=<MseLossBackward>)
  10660. loss_train_step after backward: tensor(0.0479, device='cuda:0', grad_fn=<MseLossBackward>)
  10661. loss_train: 3.5717637445777655
  10662. step: 64
  10663. running loss: 0.055808808509027585
  10664.  
  10665. Train Steps: 64/90 Loss: 0.0558 torch.Size([8, 600, 800])
  10666. torch.Size([8, 8])
  10667. tensor([[0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578],
  10668. [0.6110, 0.4047, 0.8700, 0.4483, 0.3713, 0.3967, 0.5088, 0.5517],
  10669. [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5837, 0.5500],
  10670. [0.6203, 0.4096, 0.8862, 0.4267, 0.3538, 0.4117, 0.6025, 0.5650],
  10671. [0.6030, 0.3969, 0.7988, 0.3917, 0.3450, 0.3667, 0.5266, 0.4700],
  10672. [0.6346, 0.4092, 0.7712, 0.5917, 0.4037, 0.4767, 0.7343, 0.5725],
  10673. [0.6134, 0.4090, 0.6926, 0.2819, 0.3538, 0.3233, 0.5563, 0.5667],
  10674. [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791]],
  10675. device='cuda:0', dtype=torch.float64)
  10676. predictions are: tensor([[ 0.1179, -0.7341, 0.9864, -1.3143, -0.3100, -1.6252, 0.2862, 0.2487],
  10677. [-0.0459, -0.8347, 1.5879, -0.2601, -0.5802, -0.5579, 0.1200, 0.2064],
  10678. [ 0.1300, -0.7234, 1.6999, -0.2421, -0.2803, -0.0950, 0.4022, 0.2162],
  10679. [ 0.7086, -0.3531, 1.8816, -0.3375, -0.4554, -0.5277, 0.5610, 0.2529],
  10680. [ 0.6942, -0.3544, 1.5433, -0.5519, -0.6117, -0.6795, 0.2996, 0.1721],
  10681. [ 0.4877, -0.4784, 1.4619, -0.0209, -0.3364, -0.2908, 0.5970, 0.2885],
  10682. [ 0.5985, -0.3888, 1.0590, -1.0507, -0.5366, -0.9573, 0.2380, 0.2933],
  10683. [ 0.4173, -0.5434, 1.8164, -0.0607, -0.3924, -0.0882, 0.5622, 0.1204]],
  10684. device='cuda:0', grad_fn=<AddmmBackward>)
  10685. landmarks are: tensor([[[ 0.5232, -0.4189, 0.9358, -1.4006, -0.3921, -1.3698, 0.2555,
  10686. 0.2906],
  10687. [ 0.5363, -0.4168, 1.7326, -0.2151, -0.5711, -0.4537, 0.0640,
  10688. 0.2622],
  10689. [ 0.5604, -0.4620, 1.7961, -0.1997, -0.3229, 0.1082, 0.4104,
  10690. 0.2545],
  10691. [ 0.5790, -0.3940, 1.8076, -0.3152, -0.6520, -0.3844, 0.4970,
  10692. 0.3238],
  10693. [ 0.4992, -0.4525, 1.4035, -0.4768, -0.6924, -0.5923, 0.1465,
  10694. -0.1151],
  10695. [ 0.6454, -0.3960, 1.2764, 0.4470, -0.4210, -0.0842, 1.1057,
  10696. 0.3585],
  10697. [ 0.5473, -0.3966, 0.9131, -0.9838, -0.6520, -0.7925, 0.2834,
  10698. 0.3315],
  10699. [ 0.6074, -0.4223, 1.8192, -0.0303, -0.4152, 0.1236, 0.6524,
  10700. -0.0731]]], device='cuda:0')
  10701. loss_train_step before backward: tensor(0.0361, device='cuda:0', grad_fn=<MseLossBackward>)
  10702. loss_train_step after backward: tensor(0.0361, device='cuda:0', grad_fn=<MseLossBackward>)
  10703. loss_train: 3.6078634057193995
  10704. step: 65
  10705. running loss: 0.05550559085722153
  10706. Train Steps: 65/90 Loss: 0.0555 torch.Size([8, 600, 800])
  10707. torch.Size([8, 8])
  10708. tensor([[0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150],
  10709. [0.6171, 0.4127, 0.8900, 0.4800, 0.4325, 0.5783, 0.5769, 0.5090],
  10710. [0.6128, 0.4118, 0.8638, 0.5333, 0.4625, 0.5267, 0.5193, 0.5475],
  10711. [0.6122, 0.4006, 0.8850, 0.4217, 0.4088, 0.5517, 0.6063, 0.5517],
  10712. [0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719],
  10713. [0.6182, 0.4058, 0.8738, 0.4350, 0.3563, 0.3400, 0.5290, 0.5822],
  10714. [0.6273, 0.4143, 0.8750, 0.5700, 0.3987, 0.4717, 0.6013, 0.5467],
  10715. [0.6350, 0.4043, 0.8738, 0.5650, 0.3850, 0.4750, 0.6401, 0.4950]],
  10716. device='cuda:0', dtype=torch.float64)
  10717. predictions are: tensor([[ 0.6163, -0.4138, 1.7250, -0.3640, -0.5013, -0.7299, 0.5247, 0.1902],
  10718. [ 0.5455, -0.4279, 1.6308, -0.4267, -0.2563, -0.1842, 0.3521, 0.2184],
  10719. [ 0.4818, -0.4494, 1.5292, -0.1557, -0.1983, -0.3982, 0.2413, 0.2642],
  10720. [ 0.6494, -0.3782, 1.5309, -0.6403, -0.3890, -0.0985, 0.4624, 0.2583],
  10721. [ 0.2372, -0.6260, 1.3226, -0.8351, -0.6079, -0.6038, 0.2684, 0.2611],
  10722. [-0.0703, -0.8515, 1.4496, -0.6823, -0.6232, -1.0105, 0.1126, 0.2541],
  10723. [ 0.3390, -0.5573, 1.5586, -0.0887, -0.4338, -0.5858, 0.3697, 0.2282],
  10724. [ 0.6466, -0.4126, 1.6265, -0.0297, -0.4755, -0.4721, 0.5771, 0.1254]],
  10725. device='cuda:0', grad_fn=<AddmmBackward>)
  10726. landmarks are: tensor([[[ 6.0964e-01, -4.0462e-01, 1.8249e+00, -7.2363e-03, -6.1155e-01,
  10727. -3.5366e-01, 6.1824e-01, 9.2841e-02],
  10728. [ 5.6449e-01, -3.7968e-01, 1.8249e+00, -6.8822e-02, -2.8822e-01,
  10729. 3.8537e-01, 3.7891e-01, 6.5205e-02],
  10730. [ 5.4475e-01, -3.8383e-01, 1.7037e+00, 1.7752e-01, -1.4965e-01,
  10731. 1.4673e-01, 1.1283e-01, 2.4313e-01],
  10732. [ 5.4169e-01, -4.3549e-01, 1.8018e+00, -3.3826e-01, -3.9792e-01,
  10733. 2.6220e-01, 5.1432e-01, 2.6220e-01],
  10734. [ 5.3672e-01, -4.2941e-01, 1.5709e+00, -4.9992e-01, -6.6928e-01,
  10735. -3.0747e-01, 2.4546e-01, 3.5585e-01],
  10736. [ 5.6969e-01, -4.1132e-01, 1.7499e+00, -2.7667e-01, -6.4042e-01,
  10737. -7.1547e-01, 1.5756e-01, 4.0319e-01],
  10738. [ 6.1149e-01, -3.7244e-01, 1.7557e+00, 3.4688e-01, -4.4411e-01,
  10739. -1.0731e-01, 4.9122e-01, 2.3911e-01],
  10740. [ 6.4706e-01, -4.1832e-01, 1.7499e+00, 3.2379e-01, -5.0762e-01,
  10741. -9.1917e-02, 6.7064e-01, 4.6189e-04]]], device='cuda:0')
  10742. loss_train_step before backward: tensor(0.0605, device='cuda:0', grad_fn=<MseLossBackward>)
  10743. loss_train_step after backward: tensor(0.0605, device='cuda:0', grad_fn=<MseLossBackward>)
  10744. loss_train: 3.6683297883719206
  10745. step: 66
  10746. running loss: 0.055580754369271526
  10747. Train Steps: 66/90 Loss: 0.0556 torch.Size([8, 600, 800])
  10748. torch.Size([8, 8])
  10749. tensor([[0.6126, 0.4067, 0.8638, 0.5383, 0.4188, 0.4850, 0.5016, 0.5392],
  10750. [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467],
  10751. [0.6131, 0.4064, 0.8638, 0.5200, 0.4788, 0.4783, 0.5258, 0.5867],
  10752. [0.6212, 0.4159, 0.8675, 0.5783, 0.4088, 0.4317, 0.5613, 0.5917],
  10753. [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426],
  10754. [0.6034, 0.4011, 0.7350, 0.2533, 0.3438, 0.3367, 0.5516, 0.5084],
  10755. [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817],
  10756. [0.6224, 0.4061, 0.8988, 0.4300, 0.3838, 0.4750, 0.6112, 0.5483]],
  10757. device='cuda:0', dtype=torch.float64)
  10758. predictions are: tensor([[ 0.3850, -0.5190, 1.6111, 0.1055, -0.3750, -0.2330, 0.2022, 0.1973],
  10759. [ 0.3577, -0.5619, 1.4013, -1.0461, -0.6647, -0.5698, 0.4761, 0.2041],
  10760. [ 0.6324, -0.3384, 1.5729, -0.1080, -0.1719, -0.2944, 0.2511, 0.2831],
  10761. [ 0.1516, -0.6485, 1.5047, 0.1610, -0.4048, -0.4003, 0.2617, 0.3252],
  10762. [ 0.3335, -0.5744, 1.6998, -0.5424, -0.5618, -0.9373, 0.3913, 0.1944],
  10763. [ 0.6777, -0.3749, 1.0782, -1.4071, -0.6146, -1.0976, 0.3684, 0.1518],
  10764. [ 0.3836, -0.5650, 1.5613, 0.1353, -0.2767, -0.2953, 0.2244, 0.2116],
  10765. [ 0.7497, -0.3079, 1.7789, -0.4326, -0.5525, -0.0923, 0.4460, 0.1365]],
  10766. device='cuda:0', grad_fn=<AddmmBackward>)
  10767. landmarks are: tensor([[[ 0.5436, -0.4076, 1.7037, 0.2006, -0.3517, -0.0457, 0.0311,
  10768. 0.2048],
  10769. [ 0.5484, -0.4591, 1.5940, -0.7771, -0.6693, -0.3460, 0.4219,
  10770. 0.2391],
  10771. [ 0.5462, -0.4090, 1.7037, 0.1159, -0.0746, -0.0765, 0.1427,
  10772. 0.4239],
  10773. [ 0.5836, -0.3649, 1.7210, 0.3854, -0.3979, -0.2921, 0.3065,
  10774. 0.4470],
  10775. [ 0.5788, -0.4135, 1.8214, -0.2408, -0.6039, -0.7854, 0.4115,
  10776. 0.2203],
  10777. [ 0.5009, -0.4333, 1.1090, -1.1158, -0.6982, -0.7309, 0.2617,
  10778. 0.0622],
  10779. [ 0.5193, -0.4699, 1.5767, 0.4008, -0.2420, 0.0774, 0.1178,
  10780. -0.0610],
  10781. [ 0.5888, -0.4101, 1.8654, -0.2998, -0.5134, -0.0919, 0.5374,
  10782. 0.2468]]], device='cuda:0')
  10783. loss_train_step before backward: tensor(0.0281, device='cuda:0', grad_fn=<MseLossBackward>)
  10784. loss_train_step after backward: tensor(0.0281, device='cuda:0', grad_fn=<MseLossBackward>)
  10785. loss_train: 3.6964302863925695
  10786. step: 67
  10787. running loss: 0.05517060128944134
  10788. Train Steps: 67/90 Loss: 0.0552 torch.Size([8, 600, 800])
  10789. torch.Size([8, 8])
  10790. tensor([[0.6212, 0.4171, 0.7875, 0.3633, 0.3813, 0.2933, 0.5675, 0.5700],
  10791. [0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767],
  10792. [ nan, nan, 0.7512, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617],
  10793. [ nan, nan, 0.8938, 0.2850, 0.4662, 0.3117, 0.7406, 0.5528],
  10794. [0.6197, 0.4050, 0.7527, 0.2000, 0.4042, 0.2249, 0.5895, 0.4995],
  10795. [ nan, nan, 0.8300, 0.3150, 0.3588, 0.3383, 0.5208, 0.5194],
  10796. [0.6198, 0.4115, 0.7762, 0.2717, 0.3713, 0.3200, 0.5837, 0.5683],
  10797. [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611]],
  10798. device='cuda:0', dtype=torch.float64)
  10799. predictions are: tensor([[ 1.0962, -0.0146, 1.2800, -0.4273, -0.4927, -0.5402, 0.3793, 0.2654],
  10800. [ 1.1303, 0.0088, 1.7883, 0.1866, -0.6208, -0.2242, 0.3961, 0.1948],
  10801. [-0.9555, -1.3945, 0.9320, -1.0741, -0.2958, -1.0923, 0.1046, 0.3117],
  10802. [-0.2414, -0.9752, 1.8767, -0.6888, -0.1368, -0.7230, 0.9422, 0.2153],
  10803. [ 1.2111, 0.0026, 1.0614, -0.9498, -0.4365, -1.0711, 0.2465, 0.1665],
  10804. [-0.9628, -1.4400, 1.3093, -0.5955, -0.5826, -0.5954, -0.0528, 0.1944],
  10805. [ 0.7971, -0.2340, 1.3919, -0.7441, -0.5827, -0.6364, 0.2316, 0.2503],
  10806. [ 1.2873, 0.0838, 1.7751, 0.3542, -0.3632, 0.3964, 0.1676, 0.1421]],
  10807. device='cuda:0', grad_fn=<AddmmBackward>)
  10808. landmarks are: tensor([[[ 0.5832, -0.3593, 1.3515, -0.6077, -0.5249, -0.9310, 0.3353,
  10809. 0.3469],
  10810. [ 0.6058, -0.3216, 1.8423, -0.2536, -0.5885, -0.6000, 0.3353,
  10811. 0.3777],
  10812. [-2.2859, -2.2859, 1.1841, -1.3082, -0.3055, -1.3621, 0.3007,
  10813. 0.3084],
  10814. [-2.2859, -2.2859, 1.8423, -0.9695, -0.1323, -0.8463, 1.1349,
  10815. 0.2676],
  10816. [ 0.5762, -0.4153, 1.1908, -1.3622, -0.4190, -1.2471, 0.4368,
  10817. 0.0213],
  10818. [-2.2859, -2.2859, 1.5478, -0.8309, -0.6289, -0.7232, 0.1198,
  10819. 0.1133],
  10820. [ 0.5768, -0.3852, 1.2995, -1.0311, -0.5711, -0.8079, 0.4104,
  10821. 0.3392],
  10822. [ 0.5491, -0.4290, 1.7788, -0.1073, -0.2651, -0.1073, 0.2555,
  10823. 0.3057]]], device='cuda:0')
  10824. loss_train_step before backward: tensor(0.2376, device='cuda:0', grad_fn=<MseLossBackward>)
  10825. loss_train_step after backward: tensor(0.2376, device='cuda:0', grad_fn=<MseLossBackward>)
  10826. loss_train: 3.9340319838374853
  10827. step: 68
  10828. running loss: 0.05785341152702184
  10829.  
  10830. Train Steps: 68/90 Loss: 0.0579 torch.Size([8, 600, 800])
  10831. torch.Size([8, 8])
  10832. tensor([[0.6273, 0.4100, 0.7137, 0.2133, 0.4000, 0.2650, 0.6075, 0.5633],
  10833. [0.6126, 0.4067, 0.8638, 0.5383, 0.4188, 0.4850, 0.5016, 0.5392],
  10834. [0.6200, 0.4098, 0.8237, 0.2917, 0.4012, 0.2967, 0.6000, 0.5683],
  10835. [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767],
  10836. [ nan, nan, 0.7648, 0.2722, 0.3962, 0.2183, 0.5060, 0.5422],
  10837. [0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400],
  10838. [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5513, 0.5750],
  10839. [0.6308, 0.3990, 0.8688, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133]],
  10840. device='cuda:0', dtype=torch.float64)
  10841. predictions are: tensor([[ 0.9220, -0.1580, 1.0716, -1.1080, -0.4721, -0.9665, 0.4439, 0.2881],
  10842. [ 0.9092, -0.1540, 1.7848, 0.5763, -0.4562, 0.3608, 0.1721, 0.1350],
  10843. [ 1.0781, -0.0502, 1.5693, -0.8404, -0.4444, -0.5808, 0.4128, 0.2579],
  10844. [ 0.7694, -0.1921, 1.6814, 0.1707, -0.6954, -0.4038, 0.1699, 0.1999],
  10845. [-1.3740, -1.7090, 1.0674, -0.9417, -0.4480, -0.9951, 0.0507, 0.2347],
  10846. [ 0.7294, -0.2306, 1.4294, -0.7451, -0.2152, -0.9191, 0.3202, 0.2228],
  10847. [-0.9718, -1.4156, 0.8144, -1.1385, -0.3027, -1.2208, 0.0989, 0.3416],
  10848. [ 1.0325, -0.1115, 1.8781, 0.3365, -0.4912, 0.4340, 0.5514, 0.0747]],
  10849. device='cuda:0', grad_fn=<AddmmBackward>)
  10850. landmarks are: tensor([[[ 0.6115, -0.3924, 1.0109, -1.3005, -0.4383, -1.0619, 0.5201,
  10851. 0.3161],
  10852. [ 0.5436, -0.4076, 1.7037, 0.2006, -0.3517, -0.0457, 0.0311,
  10853. 0.2048],
  10854. [ 0.5778, -0.3930, 1.5189, -0.9387, -0.4326, -0.9156, 0.4855,
  10855. 0.3392],
  10856. [ 0.5781, -0.3848, 1.7441, -0.0996, -0.5769, -0.8002, 0.2021,
  10857. 0.3778],
  10858. [-2.2859, -2.2859, 1.2469, -1.0288, -0.4557, -1.2774, 0.0511,
  10859. 0.2183],
  10860. [ 0.5911, -0.3888, 1.4727, -0.9541, -0.0919, -1.4930, 0.3988,
  10861. 0.2083],
  10862. [ 0.5663, -0.4396, 0.8261, -1.1312, -0.2940, -1.3929, 0.2603,
  10863. 0.3700],
  10864. [ 0.6275, -0.4430, 1.7268, 0.1082, -0.4614, 0.0159, 0.6644,
  10865. 0.0851]]], device='cuda:0')
  10866. loss_train_step before backward: tensor(0.1133, device='cuda:0', grad_fn=<MseLossBackward>)
  10867. loss_train_step after backward: tensor(0.1133, device='cuda:0', grad_fn=<MseLossBackward>)
  10868. loss_train: 4.047340488061309
  10869. step: 69
  10870. running loss: 0.05865710852262766
  10871. Train Steps: 69/90 Loss: 0.0587 torch.Size([8, 600, 800])
  10872. torch.Size([8, 8])
  10873. tensor([[0.6085, 0.4008, 0.8588, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283],
  10874. [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5988, 0.5283],
  10875. [0.6095, 0.3970, 0.8688, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940],
  10876. [0.6183, 0.4076, 0.8838, 0.4517, 0.3813, 0.4483, 0.5775, 0.5633],
  10877. [0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585],
  10878. [0.6262, 0.4163, 0.8850, 0.5183, 0.3763, 0.4150, 0.6025, 0.5500],
  10879. [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609],
  10880. [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317]],
  10881. device='cuda:0', dtype=torch.float64)
  10882. predictions are: tensor([[ 0.5268, -0.4157, 1.5812, -0.0265, -0.1701, -0.2196, 0.0827, 0.2199],
  10883. [ 0.4279, -0.4902, 1.4951, 0.0641, -0.5407, -0.0932, 0.2282, 0.2088],
  10884. [ 0.0935, -0.7143, 1.5847, -0.2925, -0.2550, -0.1841, 0.1801, 0.1760],
  10885. [ 0.5612, -0.4071, 1.5233, -0.3984, -0.5938, -0.1783, 0.2074, 0.2572],
  10886. [ 0.4294, -0.4534, 1.4494, -0.5346, -0.3546, 0.1353, 0.2257, 0.2259],
  10887. [ 0.4408, -0.4619, 1.5718, -0.1829, -0.6354, -0.6182, 0.2701, 0.2355],
  10888. [ 0.2905, -0.6043, 1.6711, -1.0866, -0.3326, -1.1208, 0.8237, 0.2227],
  10889. [ 0.6131, -0.3707, 1.5938, 0.0196, -0.6678, -0.7217, 0.3395, 0.1753]],
  10890. device='cuda:0', grad_fn=<AddmmBackward>)
  10891. landmarks are: tensor([[[ 0.5245, -0.4347, 1.6806, 0.1159, 0.0046, 0.0129, 0.1044,
  10892. 0.1544],
  10893. [ 0.5603, -0.4319, 1.7788, 0.1775, -0.5538, -0.0611, 0.4797,
  10894. 0.1544],
  10895. [ 0.5292, -0.4520, 1.7268, -0.0842, -0.0413, -0.0324, 0.1116,
  10896. -0.0039],
  10897. [ 0.5700, -0.4032, 1.7961, -0.1997, -0.5249, -0.2151, 0.3815,
  10898. 0.3161],
  10899. [ 0.5115, -0.4332, 1.7557, -0.3152, -0.2536, 0.3931, 0.4139,
  10900. 0.2936],
  10901. [ 0.6066, -0.3632, 1.8018, 0.1082, -0.5480, -0.3691, 0.4970,
  10902. 0.2545],
  10903. [ 0.6026, -0.4032, 1.7326, -0.7771, -0.3691, -0.8617, 0.9704,
  10904. 0.3050],
  10905. [ 0.5901, -0.4157, 1.7557, 0.1929, -0.5423, -0.5923, 0.3584,
  10906. 0.1698]]], device='cuda:0')
  10907. loss_train_step before backward: tensor(0.0292, device='cuda:0', grad_fn=<MseLossBackward>)
  10908. loss_train_step after backward: tensor(0.0292, device='cuda:0', grad_fn=<MseLossBackward>)
  10909. loss_train: 4.076580006629229
  10910. step: 70
  10911. running loss: 0.05823685723756041
  10912. Train Steps: 70/90 Loss: 0.0582 torch.Size([8, 600, 800])
  10913. torch.Size([8, 8])
  10914. tensor([[0.6147, 0.4107, 0.8137, 0.3333, 0.3750, 0.2683, 0.5006, 0.5412],
  10915. [0.6126, 0.3954, 0.8538, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350],
  10916. [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103],
  10917. [0.6151, 0.4085, 0.8750, 0.4367, 0.3887, 0.4367, 0.5066, 0.5846],
  10918. [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5988, 0.5283],
  10919. [0.6126, 0.4073, 0.8750, 0.5133, 0.3800, 0.4333, 0.4986, 0.5378],
  10920. [0.6336, 0.4086, 0.8900, 0.3950, 0.3900, 0.2950, 0.6504, 0.5066],
  10921. [0.6127, 0.4066, 0.8550, 0.5567, 0.4662, 0.5141, 0.5070, 0.5412]],
  10922. device='cuda:0', dtype=torch.float64)
  10923. predictions are: tensor([[ 0.0387, -0.7622, 1.2751, -0.9462, -0.5724, -1.0365, 0.1144, 0.2447],
  10924. [ 0.3928, -0.5237, 1.6429, -0.0807, -0.3716, -0.0862, 0.2124, 0.1810],
  10925. [ 0.3224, -0.5755, 1.7140, -0.3758, -0.4884, -0.1473, 0.1929, 0.1669],
  10926. [ 0.5737, -0.3796, 1.5390, -0.4548, -0.5041, -0.2859, 0.1710, 0.2812],
  10927. [ 0.4217, -0.4986, 1.6079, 0.0801, -0.4330, -0.0176, 0.3749, 0.2277],
  10928. [ 0.2513, -0.5825, 1.6658, 0.0256, -0.5307, -0.3036, 0.2110, 0.2185],
  10929. [ 0.7274, -0.3245, 1.5881, -0.7377, -0.4621, -1.0521, 0.7483, 0.1522],
  10930. [ 0.5020, -0.4279, 1.5960, 0.1977, -0.1573, -0.0532, 0.2000, 0.2492]],
  10931. device='cuda:0', grad_fn=<AddmmBackward>)
  10932. landmarks are: tensor([[[ 0.5532, -0.3888, 1.4727, -0.7463, -0.5538, -1.0465, 0.0265,
  10933. 0.2138],
  10934. [ 0.5435, -0.4597, 1.6575, 0.0159, -0.3229, -0.1150, 0.1877,
  10935. 0.1854],
  10936. [ 0.5253, -0.4392, 1.7730, -0.2305, -0.4268, -0.1381, 0.1651,
  10937. 0.0712],
  10938. [ 0.5553, -0.3992, 1.7557, -0.2690, -0.4903, -0.2690, 0.0542,
  10939. 0.4145],
  10940. [ 0.5603, -0.4319, 1.7788, 0.1775, -0.5538, -0.0611, 0.4797,
  10941. 0.1544],
  10942. [ 0.5435, -0.4045, 1.7557, 0.0851, -0.5307, -0.2844, 0.0172,
  10943. 0.1980],
  10944. [ 0.6405, -0.3984, 1.8249, -0.4614, -0.4845, -0.9233, 0.7182,
  10945. 0.0539],
  10946. [ 0.5443, -0.4079, 1.6633, 0.2853, -0.1325, 0.0888, 0.0558,
  10947. 0.2138]]], device='cuda:0')
  10948. loss_train_step before backward: tensor(0.0193, device='cuda:0', grad_fn=<MseLossBackward>)
  10949. loss_train_step after backward: tensor(0.0193, device='cuda:0', grad_fn=<MseLossBackward>)
  10950. loss_train: 4.095911048352718
  10951. step: 71
  10952. running loss: 0.05768888800496787
  10953. Train Steps: 71/90 Loss: 0.0577 torch.Size([8, 600, 800])
  10954. torch.Size([8, 8])
  10955. tensor([[0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672],
  10956. [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500],
  10957. [0.6224, 0.4179, 0.8700, 0.5683, 0.4037, 0.4683, 0.5650, 0.5600],
  10958. [0.6151, 0.4058, 0.7068, 0.2680, 0.3400, 0.4083, 0.5775, 0.5733],
  10959. [0.6222, 0.4169, 0.8638, 0.5650, 0.4313, 0.4783, 0.5637, 0.5633],
  10960. [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879],
  10961. [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167],
  10962. [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012]],
  10963. device='cuda:0', dtype=torch.float64)
  10964. predictions are: tensor([[ 0.5969, -0.4272, 1.8214, -0.2437, -0.6127, -0.7385, 0.4143, 0.1130],
  10965. [ 0.2856, -0.6078, 1.3131, -1.0711, -0.6239, -0.5276, 0.3054, 0.2557],
  10966. [ 0.5606, -0.4006, 1.7808, 0.4420, -0.3751, -0.0319, 0.2513, 0.2444],
  10967. [ 0.3458, -0.5758, 1.1444, -1.0937, -0.6510, -0.5539, 0.1815, 0.3004],
  10968. [ 0.5171, -0.4272, 1.7789, 0.4450, -0.2649, 0.0206, 0.1874, 0.2545],
  10969. [ 0.3801, -0.5404, 1.8297, -0.2141, -0.4082, 0.2344, 0.2679, 0.1173],
  10970. [ 0.0454, -0.7213, 1.2479, -0.9528, -0.0633, -1.2849, 0.2315, 0.3592],
  10971. [ 0.6147, -0.4189, 1.8836, 0.2746, -0.5885, -0.2440, 0.4363, 0.1093]],
  10972. device='cuda:0', grad_fn=<AddmmBackward>)
  10973. landmarks are: tensor([[[ 0.6132, -0.4324, 1.8192, -0.0842, -0.6231, -0.6385, 0.5537,
  10974. -0.1278],
  10975. [ 0.5763, -0.4147, 1.3226, -1.0619, -0.6635, -0.4152, 0.5374,
  10976. 0.2545],
  10977. [ 0.5891, -0.3557, 1.7326, 0.3392, -0.4210, -0.1227, 0.3238,
  10978. 0.3007],
  10979. [ 0.5552, -0.4113, 0.9790, -1.0480, -0.7155, -0.3998, 0.3815,
  10980. 0.3623],
  10981. [ 0.5880, -0.3605, 1.7037, 0.3238, -0.2940, -0.0765, 0.3180,
  10982. 0.3161],
  10983. [ 0.5676, -0.4327, 1.8252, -0.2095, -0.4886, 0.3084, 0.4727,
  10984. -0.0322],
  10985. [ 0.6161, -0.3075, 1.1678, -0.6500, 0.0813, -1.4006, 0.2545,
  10986. 0.5624],
  10987. [ 0.6249, -0.4352, 1.8018, 0.2545, -0.6173, -0.1997, 0.6401,
  10988. 0.0291]]], device='cuda:0')
  10989. loss_train_step before backward: tensor(0.0222, device='cuda:0', grad_fn=<MseLossBackward>)
  10990. loss_train_step after backward: tensor(0.0222, device='cuda:0', grad_fn=<MseLossBackward>)
  10991. loss_train: 4.11814627237618
  10992. step: 72
  10993. running loss: 0.05719647600522472
  10994.  
  10995. Train Steps: 72/90 Loss: 0.0572 torch.Size([8, 600, 800])
  10996. torch.Size([8, 8])
  10997. tensor([[0.6280, 0.4101, 0.9050, 0.4533, 0.3775, 0.3217, 0.6338, 0.4915],
  10998. [0.6143, 0.4034, 0.8800, 0.4833, 0.4512, 0.5367, 0.5289, 0.5097],
  10999. [0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6138, 0.5450],
  11000. [ nan, nan, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733],
  11001. [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5463, 0.5800],
  11002. [0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272],
  11003. [0.6124, 0.4069, 0.8314, 0.5001, 0.3738, 0.4650, 0.5167, 0.5402],
  11004. [0.6346, 0.4165, 0.9138, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471]],
  11005. device='cuda:0', dtype=torch.float64)
  11006. predictions are: tensor([[ 0.6401, -0.3275, 1.9039, 0.0135, -0.5674, -0.7047, 0.5180, 0.0962],
  11007. [ 0.5311, -0.3679, 1.7335, 0.1201, -0.2505, 0.2096, 0.0463, 0.1490],
  11008. [ 0.7563, -0.2695, 1.7467, -0.0610, -0.3387, 0.3851, 0.3857, 0.1715],
  11009. [-1.4279, -1.7449, 0.9868, -1.1764, -0.1548, -1.2947, 0.2669, 0.3564],
  11010. [ 0.2403, -0.5642, 1.0672, -0.7588, -0.5522, -0.9347, 0.0379, 0.3111],
  11011. [ 0.7850, -0.2328, 1.5708, -0.5475, -0.5102, -0.8937, 0.2295, 0.2070],
  11012. [ 0.9283, -0.1462, 1.5625, 0.1236, -0.4814, -0.0100, 0.1466, 0.2467],
  11013. [ 0.2391, -0.6283, 1.9438, -0.2441, -0.3081, -0.1375, 0.7502, 0.2023]],
  11014. device='cuda:0', grad_fn=<AddmmBackward>)
  11015. landmarks are: tensor([[[ 0.6148, -0.3918, 1.8942, -0.1920, -0.5423, -0.8002, 0.6414,
  11016. -0.0156],
  11017. [ 0.5515, -0.4225, 1.7788, -0.0534, -0.2016, 0.1929, 0.1568,
  11018. 0.0682],
  11019. [ 0.5960, -0.4102, 1.8018, -0.1612, -0.3344, 0.1159, 0.5490,
  11020. 0.2314],
  11021. [-2.2859, -2.2859, 0.7106, -1.4468, -0.2882, -1.4237, 0.2430,
  11022. 0.3623],
  11023. [ 0.5548, -0.3936, 1.1634, -0.8105, -0.5192, -1.0696, 0.2372,
  11024. 0.3931],
  11025. [ 0.5711, -0.4015, 1.6979, -0.6770, -0.5365, -1.0619, 0.1712,
  11026. 0.1494],
  11027. [ 0.5425, -0.4067, 1.5543, 0.0241, -0.5596, -0.1381, 0.1005,
  11028. 0.2093],
  11029. [ 0.6454, -0.3623, 1.9346, -0.4460, -0.4961, -0.2921, 1.1642,
  11030. 0.2409]]], device='cuda:0')
  11031. loss_train_step before backward: tensor(0.0410, device='cuda:0', grad_fn=<MseLossBackward>)
  11032. loss_train_step after backward: tensor(0.0410, device='cuda:0', grad_fn=<MseLossBackward>)
  11033. loss_train: 4.15917850472033
  11034. step: 73
  11035. running loss: 0.05697504800986754
  11036. Train Steps: 73/90 Loss: 0.0570 torch.Size([8, 600, 800])
  11037. torch.Size([8, 8])
  11038. tensor([[0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517],
  11039. [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717],
  11040. [0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835],
  11041. [0.6218, 0.4137, 0.7263, 0.2233, 0.4075, 0.2650, 0.6212, 0.5783],
  11042. [0.6200, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148],
  11043. [0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540],
  11044. [0.6264, 0.3972, 0.8853, 0.4771, 0.3853, 0.4511, 0.6293, 0.5334],
  11045. [0.6153, 0.4117, 0.8688, 0.5167, 0.4895, 0.5647, 0.5524, 0.5136]],
  11046. device='cuda:0', dtype=torch.float64)
  11047. predictions are: tensor([[ 0.0255, -0.7414, 1.0145, -1.0360, -0.3244, -1.0102, 0.2996, 0.3519],
  11048. [ 0.1935, -0.6416, 0.8982, -0.9045, -0.4853, -0.9969, 0.0657, 0.2896],
  11049. [ 0.5508, -0.4163, 1.8753, -0.2968, -0.3653, -0.9057, 0.3999, 0.1318],
  11050. [ 0.9391, -0.2031, 1.1864, -0.7964, -0.4275, -0.8836, 0.3701, 0.3034],
  11051. [ 0.3174, -0.5763, 1.8704, 0.0961, -0.6383, -0.0635, 0.1696, 0.1344],
  11052. [ 0.0848, -0.7070, 2.0174, -0.0102, -0.3566, -0.5104, 0.5992, 0.1564],
  11053. [ 0.4316, -0.5112, 1.8690, 0.0852, -0.5426, 0.1211, 0.4263, 0.1277],
  11054. [ 0.3346, -0.5371, 1.7998, 0.2516, -0.0975, 0.4123, 0.1748, 0.1796]],
  11055. device='cuda:0', grad_fn=<AddmmBackward>)
  11056. landmarks are: tensor([[[ 0.5713, -0.3671, 0.8665, -1.0696, -0.3691, -1.2236, 0.3527,
  11057. 0.2622],
  11058. [ 0.5633, -0.3883, 0.8648, -1.0288, -0.5192, -1.1081, 0.2776,
  11059. 0.3546],
  11060. [ 0.6117, -0.3877, 1.8249, -0.5230, -0.3517, -1.2851, 0.4200,
  11061. -0.0529],
  11062. [ 0.5863, -0.3749, 1.0686, -1.2543, -0.4037, -1.0619, 0.5836,
  11063. 0.3854],
  11064. [ 0.5781, -0.4205, 1.8159, -0.0691, -0.6116, -0.3069, 0.4236,
  11065. 0.0919],
  11066. [ 0.6240, -0.3912, 1.9115, -0.2382, -0.3979, -0.8694, 0.8644,
  11067. 0.2730],
  11068. [ 0.6075, -0.4514, 1.8032, -0.0822, -0.5061, -0.2023, 0.6208,
  11069. 0.1779],
  11070. [ 0.5564, -0.3842, 1.7268, 0.1005, -0.0250, 0.3225, 0.2658,
  11071. 0.0862]]], device='cuda:0')
  11072. loss_train_step before backward: tensor(0.0431, device='cuda:0', grad_fn=<MseLossBackward>)
  11073. loss_train_step after backward: tensor(0.0431, device='cuda:0', grad_fn=<MseLossBackward>)
  11074. loss_train: 4.202301474288106
  11075. step: 74
  11076. running loss: 0.05678785776065008
  11077. Train Steps: 74/90 Loss: 0.0568 torch.Size([8, 600, 800])
  11078. torch.Size([8, 8])
  11079. tensor([[0.6282, 0.4034, 0.7830, 0.2080, 0.4532, 0.2080, 0.6404, 0.5323],
  11080. [0.6339, 0.4149, 0.8800, 0.5000, 0.3900, 0.5283, 0.7541, 0.5424],
  11081. [0.6224, 0.4179, 0.8700, 0.5683, 0.4037, 0.4683, 0.5650, 0.5600],
  11082. [0.6182, 0.4058, 0.8738, 0.4350, 0.3563, 0.3400, 0.5290, 0.5822],
  11083. [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390],
  11084. [ nan, nan, 0.6900, 0.1917, 0.3937, 0.2367, 0.5240, 0.5246],
  11085. [0.6129, 0.4069, 0.8750, 0.5067, 0.3875, 0.4233, 0.5235, 0.5881],
  11086. [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383]],
  11087. device='cuda:0', dtype=torch.float64)
  11088. predictions are: tensor([[ 0.7725, -0.2617, 1.3868, -1.0722, -0.2079, -1.3805, 0.5797, 0.2321],
  11089. [ 0.7794, -0.2949, 1.9241, 0.0387, -0.4490, 0.2553, 0.8479, 0.1340],
  11090. [ 0.8670, -0.1966, 1.7308, 0.3913, -0.4143, -0.0618, 0.3734, 0.2194],
  11091. [ 0.5301, -0.4200, 1.6865, -0.3245, -0.6355, -0.6120, 0.1663, 0.2268],
  11092. [ 0.5905, -0.3883, 1.7943, -0.0852, -0.4299, 0.0460, 0.2930, 0.1580],
  11093. [-1.6482, -1.9028, 1.1660, -1.1783, -0.2913, -1.2421, 0.1438, 0.2935],
  11094. [ 0.5850, -0.3715, 1.7286, 0.1431, -0.5558, -0.2252, 0.1153, 0.2125],
  11095. [ 0.5957, -0.3554, 1.7312, -0.0205, -0.2870, -0.0232, 0.1938, 0.1764]],
  11096. device='cuda:0', grad_fn=<AddmmBackward>)
  11097. landmarks are: tensor([[[ 6.1577e-01, -4.2249e-01, 1.3307e+00, -1.3253e+00, -1.9244e-01,
  11098. -1.3252e+00, 6.7213e-01, 1.7271e-01],
  11099. [ 6.4212e-01, -3.6953e-01, 1.7788e+00, 2.3557e-02, -4.8453e-01,
  11100. 1.5443e-01, 1.1971e+00, 2.1955e-01],
  11101. [ 5.8909e-01, -3.5574e-01, 1.7326e+00, 3.3918e-01, -4.2102e-01,
  11102. -1.2271e-01, 3.2379e-01, 3.0069e-01],
  11103. [ 5.6969e-01, -4.1132e-01, 1.7499e+00, -2.7667e-01, -6.4042e-01,
  11104. -7.1547e-01, 1.5756e-01, 4.0319e-01],
  11105. [ 5.2500e-01, -4.6613e-01, 1.7383e+00, -7.6520e-02, -4.2679e-01,
  11106. -2.2633e-02, 2.5348e-01, 2.0347e-01],
  11107. [-2.2859e+00, -2.2859e+00, 9.0115e-01, -1.4006e+00, -4.6721e-01,
  11108. -1.1928e+00, 1.3421e-01, 1.3734e-01],
  11109. [ 5.4515e-01, -4.0670e-01, 1.7557e+00, 5.4350e-02, -4.9607e-01,
  11110. -3.3056e-01, 1.3228e-01, 4.3063e-01],
  11111. [ 5.3603e-01, -4.6490e-01, 1.6517e+00, 4.6189e-04, -2.8245e-01,
  11112. -6.8822e-02, 2.3086e-01, 2.0046e-01]]], device='cuda:0')
  11113. loss_train_step before backward: tensor(0.0209, device='cuda:0', grad_fn=<MseLossBackward>)
  11114. loss_train_step after backward: tensor(0.0209, device='cuda:0', grad_fn=<MseLossBackward>)
  11115. loss_train: 4.223247792571783
  11116. step: 75
  11117. running loss: 0.056309970567623775
  11118. Train Steps: 75/90 Loss: 0.0563 torch.Size([8, 600, 800])
  11119. torch.Size([8, 8])
  11120. tensor([[0.6199, 0.4065, 0.7598, 0.2385, 0.4317, 0.1981, 0.5933, 0.5221],
  11121. [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118],
  11122. [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167],
  11123. [0.6277, 0.4118, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938],
  11124. [ nan, nan, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552],
  11125. [0.6276, 0.4120, 0.8738, 0.3133, 0.4225, 0.2217, 0.6203, 0.4892],
  11126. [0.6224, 0.4097, 0.7438, 0.2267, 0.3850, 0.2850, 0.5988, 0.5250],
  11127. [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333]],
  11128. device='cuda:0', dtype=torch.float64)
  11129. predictions are: tensor([[ 0.7325, -0.3244, 1.1943, -0.8618, -0.3379, -0.9226, 0.3813, 0.1958],
  11130. [ 0.6158, -0.3913, 1.6816, 0.5930, -0.4862, 0.2030, 0.3024, 0.2457],
  11131. [ 0.6307, -0.3547, 1.6683, -0.4211, -0.6285, -0.4675, 0.1638, 0.1631],
  11132. [ 0.6086, -0.3802, 1.9739, -0.0922, -0.4960, -0.5134, 0.4618, 0.1108],
  11133. [-1.6262, -1.8678, 1.7056, -0.6897, 0.0506, -0.6270, 0.7561, 0.2796],
  11134. [ 0.5644, -0.3997, 1.8604, -0.3902, -0.3610, -0.6911, 0.5889, 0.1362],
  11135. [ 0.6110, -0.4007, 1.1670, -0.8371, -0.5049, -0.6734, 0.3403, 0.2276],
  11136. [ 0.6246, -0.3696, 1.0815, -0.8941, -0.3092, -1.0043, 0.2052, 0.3066]],
  11137. device='cuda:0', grad_fn=<AddmmBackward>)
  11138. landmarks are: tensor([[[ 0.5774, -0.4082, 1.2235, -1.1844, -0.2919, -1.3709, 0.4544,
  11139. 0.1256],
  11140. [ 0.6121, -0.3844, 1.4556, 0.3936, -0.4383, -0.1689, 0.2925,
  11141. 0.5401],
  11142. [ 0.5483, -0.4249, 1.5305, -0.7386, -0.6115, -0.8694, 0.3353,
  11143. 0.1005],
  11144. [ 0.6135, -0.3841, 1.8654, -0.5153, -0.4614, -1.0619, 0.6195,
  11145. -0.0049],
  11146. [-2.2859, -2.2859, 1.7557, -1.1466, 0.0871, -1.0773, 1.1239,
  11147. 0.2783],
  11148. [ 0.6128, -0.3828, 1.7499, -0.8386, -0.3344, -1.2620, 0.5792,
  11149. -0.0263],
  11150. [ 0.5891, -0.3937, 1.1494, -1.2390, -0.5076, -0.9695, 0.4797,
  11151. 0.1390],
  11152. [ 0.5853, -0.3920, 1.1090, -1.3313, -0.2882, -1.3390, 0.4624,
  11153. 0.1775]]], device='cuda:0')
  11154. loss_train_step before backward: tensor(0.0613, device='cuda:0', grad_fn=<MseLossBackward>)
  11155.  
  11156. loss_train_step after backward: tensor(0.0613, device='cuda:0', grad_fn=<MseLossBackward>)
  11157. loss_train: 4.284526634961367
  11158. step: 76
  11159. running loss: 0.05637535046001798
  11160. Train Steps: 76/90 Loss: 0.0564 torch.Size([8, 600, 800])
  11161. torch.Size([8, 8])
  11162. tensor([[0.6248, 0.4185, 0.8500, 0.5767, 0.4463, 0.4550, 0.5613, 0.5917],
  11163. [0.6147, 0.4026, 0.6600, 0.2467, 0.4088, 0.2150, 0.5489, 0.5773],
  11164. [0.6225, 0.4116, 0.8662, 0.3517, 0.3663, 0.3233, 0.5837, 0.5317],
  11165. [0.6117, 0.4019, 0.8538, 0.4067, 0.3513, 0.3583, 0.5663, 0.5133],
  11166. [0.6148, 0.3996, 0.8488, 0.3867, 0.3488, 0.4067, 0.5863, 0.5000],
  11167. [0.6304, 0.4024, 0.8925, 0.4800, 0.3937, 0.4817, 0.7485, 0.5297],
  11168. [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6038, 0.4833],
  11169. [0.6179, 0.3993, 0.8925, 0.4789, 0.3879, 0.4900, 0.6041, 0.5279]],
  11170. device='cuda:0', dtype=torch.float64)
  11171. predictions are: tensor([[ 0.5418, -0.4318, 1.7813, 0.4587, -0.0794, -0.2210, 0.4923, 0.2903],
  11172. [ 0.6229, -0.4087, 0.9010, -1.1260, -0.3065, -1.4490, 0.2739, 0.3452],
  11173. [ 0.7011, -0.3479, 1.7164, -0.6893, -0.4533, -0.8815, 0.4622, 0.2529],
  11174. [ 0.3350, -0.5665, 1.7980, -0.4537, -0.5948, -0.7768, 0.2783, 0.1598],
  11175. [ 0.2740, -0.6352, 1.6905, -0.5705, -0.6383, -0.4217, 0.5188, 0.1312],
  11176. [ 0.2169, -0.6774, 1.7397, -0.1561, -0.3435, -0.0342, 0.6219, 0.1825],
  11177. [ 0.1279, -0.7503, 1.7811, -0.0669, -0.4797, -0.1236, 0.4468, 0.1145],
  11178. [ 0.4340, -0.5311, 1.8522, -0.1038, -0.4461, -0.0288, 0.4235, 0.1248]],
  11179. device='cuda:0', grad_fn=<AddmmBackward>)
  11180. landmarks are: tensor([[[ 0.5999, -0.3530, 1.6402, 0.3777, -0.2247, -0.1843, 0.3065,
  11181. 0.4470],
  11182. [ 0.5532, -0.4264, 0.7626, -1.1466, -0.3979, -1.2928, 0.2494,
  11183. 0.3808],
  11184. [ 0.5893, -0.3847, 1.7152, -0.6616, -0.5942, -0.7925, 0.4104,
  11185. 0.1698],
  11186. [ 0.5393, -0.4294, 1.6575, -0.4075, -0.6635, -0.6308, 0.3296,
  11187. 0.0851],
  11188. [ 0.5538, -0.4400, 1.6344, -0.4999, -0.6751, -0.4075, 0.4219,
  11189. 0.0236],
  11190. [ 0.6257, -0.4273, 1.8365, -0.0688, -0.4672, -0.0611, 1.1715,
  11191. 0.1608],
  11192. [ 0.5726, -0.4795, 1.7788, -0.0149, -0.5711, -0.0688, 0.5028,
  11193. -0.0534],
  11194. [ 0.5680, -0.4417, 1.8365, -0.0740, -0.4941, -0.0227, 0.5045,
  11195. 0.1525]]], device='cuda:0')
  11196. loss_train_step before backward: tensor(0.0220, device='cuda:0', grad_fn=<MseLossBackward>)
  11197. loss_train_step after backward: tensor(0.0220, device='cuda:0', grad_fn=<MseLossBackward>)
  11198. loss_train: 4.306571505963802
  11199. step: 77
  11200. running loss: 0.055929500077451975
  11201. Train Steps: 77/90 Loss: 0.0559 torch.Size([8, 600, 800])
  11202. torch.Size([8, 8])
  11203. tensor([[0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391],
  11204. [0.6257, 0.4167, 0.8775, 0.3433, 0.3563, 0.4133, 0.6200, 0.5667],
  11205. [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817],
  11206. [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170],
  11207. [0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533],
  11208. [0.6250, 0.4146, 0.8838, 0.3933, 0.3588, 0.4283, 0.6162, 0.5367],
  11209. [0.6197, 0.3986, 0.8800, 0.4617, 0.4188, 0.4783, 0.5687, 0.5550],
  11210. [0.6188, 0.4099, 0.7400, 0.2433, 0.3962, 0.2750, 0.6162, 0.5467]],
  11211. device='cuda:0', dtype=torch.float64)
  11212. predictions are: tensor([[ 0.2123, -0.6681, 1.4267, -0.7008, -0.3961, -1.0789, 0.4470, 0.1575],
  11213. [ 0.6287, -0.4280, 1.9480, -0.2461, -0.6259, -0.2741, 0.5778, 0.1985],
  11214. [ 0.2384, -0.6014, 1.5249, -0.6004, -0.1790, -0.7868, 0.4314, 0.3348],
  11215. [ 0.2662, -0.6620, 1.6252, -0.7981, -0.0459, -1.2227, 0.8659, 0.1341],
  11216. [ 0.3869, -0.5725, 1.1212, -0.8591, -0.5330, -0.8021, 0.4060, 0.2322],
  11217. [ 0.6979, -0.3889, 1.9113, -0.0834, -0.5975, -0.2396, 0.4525, 0.2011],
  11218. [ 0.1241, -0.7364, 1.8760, 0.2413, -0.3607, 0.2421, 0.2625, 0.1373],
  11219. [ 0.4837, -0.5007, 1.1885, -0.8307, -0.4771, -0.9443, 0.4297, 0.1917]],
  11220. device='cuda:0', grad_fn=<AddmmBackward>)
  11221. landmarks are: tensor([[[ 0.5785, -0.4259, 1.4228, -1.0261, -0.4190, -1.2189, 0.4763,
  11222. 0.2043],
  11223. [ 0.6040, -0.3614, 1.7672, -0.7001, -0.6404, -0.3768, 0.5778,
  11224. 0.3315],
  11225. [ 0.5964, -0.3380, 1.4785, -0.8386, -0.2420, -1.0619, 0.3238,
  11226. 0.4008],
  11227. [ 0.6471, -0.3840, 1.5767, -1.0311, -0.0457, -1.5007, 0.6889,
  11228. 0.1020],
  11229. [ 0.5878, -0.4052, 1.0229, -1.2855, -0.5596, -0.8232, 0.5316,
  11230. 0.2699],
  11231. [ 0.6009, -0.3710, 1.7961, -0.4691, -0.6289, -0.3075, 0.5605,
  11232. 0.1929],
  11233. [ 0.5763, -0.4450, 1.7788, -0.1535, -0.3517, -0.0765, 0.3411,
  11234. 0.2776],
  11235. [ 0.5725, -0.3926, 1.1321, -1.1620, -0.4557, -1.0157, 0.5605,
  11236. 0.2391]]], device='cuda:0')
  11237. loss_train_step before backward: tensor(0.0397, device='cuda:0', grad_fn=<MseLossBackward>)
  11238. loss_train_step after backward: tensor(0.0397, device='cuda:0', grad_fn=<MseLossBackward>)
  11239. loss_train: 4.346283700317144
  11240. step: 78
  11241. running loss: 0.05572158590150185
  11242. Train Steps: 78/90 Loss: 0.0557 torch.Size([8, 600, 800])
  11243. torch.Size([8, 8])
  11244. tensor([[0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585],
  11245. [0.6135, 0.4115, 0.8838, 0.4667, 0.4288, 0.6050, 0.5778, 0.5097],
  11246. [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567],
  11247. [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274],
  11248. [0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204],
  11249. [0.6165, 0.4106, 0.7575, 0.1733, 0.3838, 0.2650, 0.5680, 0.5116],
  11250. [0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938],
  11251. [0.6321, 0.4048, 0.8738, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927]],
  11252. device='cuda:0', dtype=torch.float64)
  11253. predictions are: tensor([[ 0.2503, -0.6194, 1.6771, -0.5663, -0.2092, -0.0134, 0.4559, 0.2668],
  11254. [ 0.4025, -0.5215, 1.7636, -0.3499, -0.2321, 0.0510, 0.4189, 0.2570],
  11255. [ 0.3624, -0.5574, 1.6009, -0.4521, -0.5776, -0.7097, 0.4002, 0.2726],
  11256. [ 0.4997, -0.5109, 1.9008, -0.2579, -0.4983, -0.6274, 0.6899, 0.2027],
  11257. [ 0.3547, -0.6100, 1.7118, -0.0586, -0.3042, -0.1724, 0.6228, 0.1929],
  11258. [ 0.8087, -0.3421, 1.1945, -1.4183, -0.5040, -1.5953, 0.5418, 0.1857],
  11259. [ 0.5330, -0.4932, 1.7886, 0.0924, -0.4707, -0.4936, 0.5452, 0.1295],
  11260. [ 0.4743, -0.5207, 1.8230, 0.0724, -0.5959, -0.6447, 0.4366, 0.1072]],
  11261. device='cuda:0', grad_fn=<AddmmBackward>)
  11262. landmarks are: tensor([[[ 0.5115, -0.4332, 1.7557, -0.3152, -0.2536, 0.3931, 0.4139,
  11263. 0.2936],
  11264. [ 0.5477, -0.3851, 1.7961, -0.1304, -0.3055, 0.5085, 0.3830,
  11265. 0.0682],
  11266. [ 0.5432, -0.4336, 1.7095, -0.1766, -0.5942, -0.4845, 0.3007,
  11267. 0.2853],
  11268. [ 0.6108, -0.4008, 1.9088, -0.0253, -0.5769, -0.3075, 0.7905,
  11269. 0.1499],
  11270. [ 0.6042, -0.4273, 1.7198, 0.2184, -0.3478, 0.1149, 0.8062,
  11271. 0.1176],
  11272. [ 0.5618, -0.3895, 1.2129, -1.4853, -0.5134, -1.0619, 0.3378,
  11273. 0.0772],
  11274. [ 0.6395, -0.4213, 1.7037, 0.3623, -0.4326, -0.1073, 0.6560,
  11275. -0.0049],
  11276. [ 0.6335, -0.4162, 1.7499, 0.3084, -0.4961, -0.2459, 0.6524,
  11277. -0.0102]]], device='cuda:0')
  11278. loss_train_step before backward: tensor(0.0373, device='cuda:0', grad_fn=<MseLossBackward>)
  11279. loss_train_step after backward: tensor(0.0373, device='cuda:0', grad_fn=<MseLossBackward>)
  11280. loss_train: 4.3835613541305065
  11281. step: 79
  11282. running loss: 0.055488118406715276
  11283. Train Steps: 79/90 Loss: 0.0555 torch.Size([8, 600, 800])
  11284. torch.Size([8, 8])
  11285. tensor([[0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494],
  11286. [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646],
  11287. [0.6233, 0.4091, 0.8100, 0.2950, 0.3563, 0.3883, 0.6013, 0.5200],
  11288. [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683],
  11289. [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466],
  11290. [0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117],
  11291. [0.6207, 0.4081, 0.7662, 0.2067, 0.3962, 0.3200, 0.6312, 0.5300],
  11292. [0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585]],
  11293. device='cuda:0', dtype=torch.float64)
  11294. predictions are: tensor([[ 0.2519, -0.6683, 1.8736, -0.7073, -0.3280, -0.7509, 1.0281, 0.1443],
  11295. [ 0.2991, -0.6456, 1.4643, -0.8791, -0.6811, -1.1794, 0.2277, 0.1704],
  11296. [ 0.4103, -0.5468, 1.4818, -1.0196, -0.6254, -0.7361, 0.6622, 0.1772],
  11297. [ 0.6202, -0.4099, 1.7969, 0.1244, -0.5959, -0.7569, 0.3795, 0.1859],
  11298. [ 0.4177, -0.5366, 1.7767, 0.0455, -0.1893, -0.0683, 0.3880, 0.1860],
  11299. [ 0.6376, -0.4223, 1.7051, 0.2728, -0.3978, -0.2228, 0.6107, 0.2835],
  11300. [ 0.7116, -0.3657, 1.2793, -1.2706, -0.4624, -1.1501, 0.7176, 0.1937],
  11301. [ 0.3611, -0.5519, 1.7906, -0.4244, -0.2106, 0.0961, 0.5442, 0.1985]],
  11302. device='cuda:0', grad_fn=<AddmmBackward>)
  11303. landmarks are: tensor([[[ 0.6388, -0.3792, 1.9635, -0.6616, -0.2536, -0.5153, 1.1605,
  11304. 0.2516],
  11305. [ 0.5483, -0.4105, 1.4208, -0.8002, -0.6000, -0.9002, 0.0511,
  11306. 0.3220],
  11307. [ 0.5932, -0.3962, 1.4554, -0.9233, -0.6404, -0.4922, 0.4912,
  11308. 0.1159],
  11309. [ 0.5777, -0.3915, 1.7961, 0.1698, -0.5192, -0.5307, 0.2141,
  11310. 0.3392],
  11311. [ 0.5178, -0.4480, 1.6748, 0.0620, -0.2767, 0.2083, 0.1067,
  11312. 0.2386],
  11313. [ 0.6118, -0.3983, 1.5824, 0.3469, -0.4268, -0.0688, 0.3469,
  11314. 0.5393],
  11315. [ 0.5809, -0.4011, 1.2533, -1.3313, -0.4557, -0.8079, 0.6298,
  11316. 0.1621],
  11317. [ 0.5115, -0.4332, 1.7557, -0.3152, -0.2536, 0.3931, 0.4139,
  11318. 0.2936]]], device='cuda:0')
  11319. loss_train_step before backward: tensor(0.0244, device='cuda:0', grad_fn=<MseLossBackward>)
  11320.  
  11321. loss_train_step after backward: tensor(0.0244, device='cuda:0', grad_fn=<MseLossBackward>)
  11322. loss_train: 4.407923875376582
  11323. step: 80
  11324. running loss: 0.055099048442207275
  11325. Train Steps: 80/90 Loss: 0.0551 torch.Size([8, 600, 800])
  11326. torch.Size([8, 8])
  11327. tensor([[0.6261, 0.3987, 0.8688, 0.4917, 0.4300, 0.5333, 0.7010, 0.5309],
  11328. [0.6185, 0.4098, 0.8838, 0.4900, 0.4537, 0.5800, 0.6288, 0.5400],
  11329. [0.6218, 0.4185, 0.7338, 0.2650, 0.4625, 0.1950, 0.5687, 0.5800],
  11330. [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461],
  11331. [0.6100, 0.4071, 0.7601, 0.3444, 0.3400, 0.4117, 0.5625, 0.5617],
  11332. [0.6179, 0.4082, 0.6688, 0.2667, 0.3588, 0.3317, 0.5750, 0.5783],
  11333. [0.6205, 0.4062, 0.8337, 0.2683, 0.3675, 0.4283, 0.6338, 0.5250],
  11334. [0.6185, 0.4129, 0.8900, 0.4567, 0.3937, 0.5417, 0.5734, 0.5110]],
  11335. device='cuda:0', dtype=torch.float64)
  11336. predictions are: tensor([[ 3.5290e-01, -6.1397e-01, 1.9549e+00, -5.3336e-03, -2.9136e-01,
  11337. -7.2324e-02, 9.3207e-01, 1.6320e-01],
  11338. [ 2.8469e-01, -6.2809e-01, 1.9445e+00, -2.0434e-04, -1.3079e-01,
  11339. -2.2070e-02, 6.3742e-01, 2.1544e-01],
  11340. [ 3.5060e-01, -5.5138e-01, 1.1136e+00, -1.1509e+00, -2.4896e-01,
  11341. -1.5926e+00, 4.1498e-01, 2.8779e-01],
  11342. [ 6.6889e-01, -3.9197e-01, 1.8542e+00, 2.6315e-01, -2.2090e-01,
  11343. -2.3001e-01, 3.9280e-01, 1.7063e-01],
  11344. [ 4.6131e-01, -4.7106e-01, 1.4068e+00, -8.2631e-01, -7.4783e-01,
  11345. -6.5008e-01, 4.4228e-01, 2.2298e-01],
  11346. [ 8.3234e-01, -2.7670e-01, 1.0542e+00, -1.0518e+00, -7.1007e-01,
  11347. -1.0961e+00, 3.9934e-01, 2.9696e-01],
  11348. [ 2.8947e-01, -6.0541e-01, 1.5527e+00, -9.9893e-01, -6.6533e-01,
  11349. -6.2536e-01, 7.8111e-01, 1.6534e-01],
  11350. [ 6.4313e-01, -4.2306e-01, 2.0376e+00, -1.2549e-01, -4.8717e-01,
  11351. -1.5416e-01, 5.7382e-01, 1.1506e-01]], device='cuda:0',
  11352. grad_fn=<AddmmBackward>)
  11353. landmarks are: tensor([[[ 0.6059, -0.4442, 1.7268, -0.0149, -0.2998, 0.1775, 0.9521,
  11354. 0.1661],
  11355. [ 0.5709, -0.3933, 1.7961, -0.0226, -0.1901, 0.3931, 0.6182,
  11356. 0.2083],
  11357. [ 0.5862, -0.3530, 1.1032, -1.0619, -0.1497, -1.3852, 0.3411,
  11358. 0.3931],
  11359. [ 0.5443, -0.3831, 1.7095, 0.1621, -0.2016, 0.1390, 0.1437,
  11360. 0.2364],
  11361. [ 0.5318, -0.4056, 1.2249, -0.6949, -0.7155, -0.3844, 0.3122,
  11362. 0.3084],
  11363. [ 0.5680, -0.4005, 0.8030, -1.0542, -0.6289, -0.7540, 0.3700,
  11364. 0.3854],
  11365. [ 0.5799, -0.4099, 1.5651, -1.0465, -0.5885, -0.3075, 0.6413,
  11366. 0.1390],
  11367. [ 0.5711, -0.3788, 1.8249, -0.1766, -0.4672, 0.2160, 0.3625,
  11368. 0.0742]]], device='cuda:0')
  11369. loss_train_step before backward: tensor(0.0296, device='cuda:0', grad_fn=<MseLossBackward>)
  11370. loss_train_step after backward: tensor(0.0296, device='cuda:0', grad_fn=<MseLossBackward>)
  11371. loss_train: 4.4374978970736265
  11372. step: 81
  11373. running loss: 0.05478392465522996
  11374. Train Steps: 81/90 Loss: 0.0548 torch.Size([8, 600, 800])
  11375. torch.Size([8, 8])
  11376. tensor([[ nan, nan, 0.9088, 0.3783, 0.4562, 0.2617, 0.6741, 0.5575],
  11377. [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834],
  11378. [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901],
  11379. [0.6212, 0.4171, 0.7875, 0.3633, 0.3813, 0.2933, 0.5675, 0.5700],
  11380. [0.6152, 0.4131, 0.6863, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305],
  11381. [0.6128, 0.4022, 0.8738, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064],
  11382. [0.6075, 0.4000, 0.8513, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280],
  11383. [0.6117, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103]],
  11384. device='cuda:0', dtype=torch.float64)
  11385. predictions are: tensor([[ 0.1934, -0.6862, 2.0995, -0.3841, -0.2928, -0.8127, 1.0653, 0.1940],
  11386. [ 0.3868, -0.5774, 1.4899, -1.2743, -0.2030, -1.3073, 0.9704, 0.1167],
  11387. [ 0.1129, -0.7319, 1.4276, -0.8536, -0.6659, -0.9510, 0.3433, 0.1963],
  11388. [ 0.9127, -0.1787, 1.4410, -0.5508, -0.5565, -0.6560, 0.6148, 0.2726],
  11389. [ 0.6915, -0.3389, 1.0534, -1.0535, -0.6616, -0.7224, 0.4801, 0.2981],
  11390. [ 0.5534, -0.4577, 1.8698, 0.1757, -0.1685, 0.1346, 0.3258, 0.1852],
  11391. [ 0.5698, -0.4565, 1.8043, 0.2670, -0.2506, 0.2621, 0.3408, 0.1853],
  11392. [ 0.3866, -0.5505, 0.9395, -1.1604, -0.5641, -1.0599, 0.3823, 0.2258]],
  11393. device='cuda:0', grad_fn=<AddmmBackward>)
  11394. landmarks are: tensor([[[-2.2859, -2.2859, 1.9115, -0.5384, -0.1785, -1.0773, 0.8278,
  11395. 0.2890],
  11396. [ 0.6126, -0.4057, 1.4439, -1.3159, -0.1150, -1.5777, 0.5537,
  11397. -0.0530],
  11398. [ 0.5515, -0.4129, 1.4785, -1.0080, -0.5192, -1.1004, 0.1034,
  11399. -0.0220],
  11400. [ 0.5832, -0.3593, 1.3515, -0.6077, -0.5249, -0.9310, 0.3353,
  11401. 0.3469],
  11402. [ 0.5557, -0.3779, 0.8838, -1.1004, -0.6115, -0.7617, 0.3769,
  11403. 0.1644],
  11404. [ 0.5446, -0.4280, 1.7499, 0.0543, 0.0156, 0.1301, 0.1918,
  11405. 0.0532],
  11406. [ 0.5202, -0.4382, 1.6460, 0.1082, -0.2029, 0.1754, 0.1067,
  11407. 0.1530],
  11408. [ 0.5393, -0.4299, 0.7453, -1.3775, -0.5596, -1.1081, 0.1527,
  11409. 0.0712]]], device='cuda:0')
  11410. loss_train_step before backward: tensor(0.1651, device='cuda:0', grad_fn=<MseLossBackward>)
  11411. loss_train_step after backward: tensor(0.1651, device='cuda:0', grad_fn=<MseLossBackward>)
  11412. loss_train: 4.602609543129802
  11413. step: 82
  11414. running loss: 0.05612938467231465
  11415. Train Steps: 82/90 Loss: 0.0561 torch.Size([8, 600, 800])
  11416. torch.Size([8, 8])
  11417. tensor([[0.6219, 0.4114, 0.8175, 0.2817, 0.3925, 0.2783, 0.5900, 0.5350],
  11418. [0.6153, 0.4117, 0.8688, 0.5167, 0.4895, 0.5647, 0.5524, 0.5136],
  11419. [ nan, nan, 0.8463, 0.2550, 0.5850, 0.2133, 0.7129, 0.6072],
  11420. [0.6182, 0.3972, 0.8552, 0.5914, 0.3683, 0.4181, 0.5688, 0.5378],
  11421. [0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117],
  11422. [0.6286, 0.4078, 0.8063, 0.2267, 0.4788, 0.1533, 0.5953, 0.4913],
  11423. [0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131],
  11424. [ nan, nan, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702]],
  11425. device='cuda:0', dtype=torch.float64)
  11426. predictions are: tensor([[ 0.6205, -0.3294, 1.2714, -1.1643, -0.4718, -0.8882, 0.4508, 0.2994],
  11427. [ 0.8339, -0.2433, 1.6666, -0.0028, -0.2328, 0.1562, 0.3314, 0.2164],
  11428. [-0.2367, -0.9609, 1.5820, -1.1476, -0.0225, -0.9568, 1.1855, 0.2878],
  11429. [ 1.3021, 0.0271, 1.6371, 0.3666, -0.7279, -0.3856, 0.4049, 0.1173],
  11430. [ 0.8664, -0.2550, 1.7417, -0.1421, -0.6307, -0.0862, 0.6780, 0.1225],
  11431. [ 0.2749, -0.5678, 1.1647, -1.2503, -0.3124, -1.2384, 0.3588, 0.2341],
  11432. [ 1.0003, -0.1206, 1.5467, -0.5714, -0.7366, -0.1394, 0.3882, 0.1260],
  11433. [-0.8779, -1.3588, 1.7350, -0.7833, -0.1671, -0.9548, 0.8598, 0.2712]],
  11434. device='cuda:0', grad_fn=<AddmmBackward>)
  11435. landmarks are: tensor([[[ 0.5868, -0.3858, 1.4901, -0.9849, -0.4730, -1.0003, 0.4393,
  11436. 0.1852],
  11437. [ 0.5564, -0.3842, 1.7268, 0.1005, -0.0250, 0.3225, 0.2658,
  11438. 0.0862],
  11439. [-2.2859, -2.2859, 1.6229, -1.1081, 0.4162, -1.3005, 1.0070,
  11440. 0.5188],
  11441. [ 0.5697, -0.4514, 1.6642, 0.4459, -0.5850, -0.3546, 0.3414,
  11442. 0.1982],
  11443. [ 0.6262, -0.4461, 1.8480, -0.0534, -0.5827, -0.1227, 0.6587,
  11444. 0.0774],
  11445. [ 0.6177, -0.4022, 1.4381, -1.2390, -0.0746, -1.5777, 0.4639,
  11446. -0.0168],
  11447. [ 0.5581, -0.4586, 1.5586, -0.3747, -0.6792, -0.2391, 0.4455,
  11448. 0.0840],
  11449. [-2.2859, -2.2859, 1.8942, -0.6693, 0.0871, -1.2236, 1.1130,
  11450. 0.3478]]], device='cuda:0')
  11451. loss_train_step before backward: tensor(0.1799, device='cuda:0', grad_fn=<MseLossBackward>)
  11452. loss_train_step after backward: tensor(0.1799, device='cuda:0', grad_fn=<MseLossBackward>)
  11453. loss_train: 4.782462088391185
  11454. step: 83
  11455. running loss: 0.05762002516133958
  11456. Train Steps: 83/90 Loss: 0.0576 torch.Size([8, 600, 800])
  11457. torch.Size([8, 8])
  11458. tensor([[ nan, nan, 0.7225, 0.2167, 0.3987, 0.2283, 0.5427, 0.5181],
  11459. [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478],
  11460. [0.6168, 0.4029, 0.8523, 0.3417, 0.3588, 0.5000, 0.6125, 0.5400],
  11461. [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467],
  11462. [0.6261, 0.3987, 0.8688, 0.4917, 0.4300, 0.5333, 0.7010, 0.5309],
  11463. [0.6275, 0.4081, 0.8063, 0.2017, 0.4825, 0.1583, 0.6156, 0.4869],
  11464. [0.6246, 0.4008, 0.8757, 0.5088, 0.4101, 0.5392, 0.6644, 0.5133],
  11465. [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103]],
  11466. device='cuda:0', dtype=torch.float64)
  11467. predictions are: tensor([[-1.6758, -1.8790, 0.9515, -1.1832, -0.3921, -1.2005, 0.1190, 0.2377],
  11468. [ 0.6928, -0.3176, 1.7793, -0.5673, -0.5725, -0.8589, 0.5203, 0.2073],
  11469. [ 0.6950, -0.3005, 1.5411, -0.6932, -0.5098, -0.0067, 0.7635, 0.2678],
  11470. [ 0.8057, -0.2322, 1.5274, -0.7015, -0.6557, -0.4211, 0.5909, 0.1877],
  11471. [ 0.8430, -0.2581, 1.7463, 0.0786, -0.2258, 0.1675, 0.8384, 0.2404],
  11472. [ 0.8383, -0.2479, 1.3780, -1.2788, -0.1072, -1.4723, 0.6900, 0.1429],
  11473. [ 0.7669, -0.3051, 1.6815, 0.1584, -0.2706, 0.1765, 0.6218, 0.2041],
  11474. [ 0.6534, -0.3386, 1.4837, -0.7632, -0.6079, -0.6089, 0.3371, 0.2506]],
  11475. device='cuda:0', grad_fn=<AddmmBackward>)
  11476. landmarks are: tensor([[[-2.2859, -2.2859, 1.0513, -1.2851, -0.4441, -1.2313, 0.2206,
  11477. 0.1073],
  11478. [ 0.5792, -0.4052, 1.8214, -0.6587, -0.5384, -0.8924, 0.4381,
  11479. 0.2442],
  11480. [ 0.5631, -0.4249, 1.6509, -0.7078, -0.6289, 0.0236, 0.5432,
  11481. 0.2083],
  11482. [ 0.5484, -0.4591, 1.5940, -0.7771, -0.6693, -0.3460, 0.4219,
  11483. 0.2391],
  11484. [ 0.6059, -0.4442, 1.7268, -0.0149, -0.2998, 0.1775, 0.9521,
  11485. 0.1661],
  11486. [ 0.6124, -0.4010, 1.4381, -1.3544, -0.0573, -1.5546, 0.5573,
  11487. -0.0369],
  11488. [ 0.5993, -0.4345, 1.7587, 0.0641, -0.3917, 0.2048, 0.7827,
  11489. 0.0852],
  11490. [ 0.5886, -0.3784, 1.4554, -0.9079, -0.6577, -0.4845, 0.3440,
  11491. 0.0712]]], device='cuda:0')
  11492. loss_train_step before backward: tensor(0.0214, device='cuda:0', grad_fn=<MseLossBackward>)
  11493. loss_train_step after backward: tensor(0.0214, device='cuda:0', grad_fn=<MseLossBackward>)
  11494. loss_train: 4.803826930001378
  11495. step: 84
  11496. running loss: 0.057188415833349736
  11497.  
  11498. Train Steps: 84/90 Loss: 0.0572 torch.Size([8, 600, 800])
  11499. torch.Size([8, 8])
  11500. tensor([[0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892],
  11501. [0.6270, 0.4267, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183],
  11502. [0.6160, 0.4093, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808],
  11503. [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400],
  11504. [0.6128, 0.4115, 0.7934, 0.3778, 0.3450, 0.4033, 0.5337, 0.5456],
  11505. [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817],
  11506. [0.6265, 0.4071, 0.8875, 0.3367, 0.3975, 0.3350, 0.6312, 0.5250],
  11507. [0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496]],
  11508. device='cuda:0', dtype=torch.float64)
  11509. predictions are: tensor([[ 0.5288, -0.4258, 1.6598, 0.0234, -0.6744, -0.2807, 0.2428, 0.2266],
  11510. [ 0.9124, -0.1514, 0.9119, -1.0870, -0.1578, -1.0845, 0.3587, 0.3047],
  11511. [ 0.5259, -0.4405, 1.5958, -0.1026, -0.4962, 0.0710, 0.4900, 0.2501],
  11512. [ 0.4467, -0.4996, 1.2111, -1.1375, -0.1979, -1.1565, 0.5828, 0.1334],
  11513. [ 0.8034, -0.2387, 1.4822, -0.5721, -0.6348, -0.1920, 0.4950, 0.1476],
  11514. [ 0.0750, -0.6879, 1.3695, -0.9216, -0.2341, -0.8327, 0.4966, 0.2875],
  11515. [-0.2534, -0.9799, 1.8873, -0.7545, -0.4268, -0.6145, 1.0349, 0.0990],
  11516. [ 0.2832, -0.5986, 1.6411, -0.9750, -0.3327, -0.7753, 0.6956, 0.1492]],
  11517. device='cuda:0', grad_fn=<AddmmBackward>)
  11518. landmarks are: tensor([[[ 0.5454, -0.4053, 1.6633, -0.1766, -0.6058, -0.5923, 0.1577,
  11519. 0.4357],
  11520. [ 0.6101, -0.3152, 1.0166, -0.7540, -0.0226, -1.4468, 0.2823,
  11521. 0.5702],
  11522. [ 0.5592, -0.3956, 1.5543, -0.2456, -0.5885, -0.1689, 0.1392,
  11523. 0.3968],
  11524. [ 0.5842, -0.3848, 1.2938, -1.2159, -0.2132, -1.4160, 0.5778,
  11525. 0.2083],
  11526. [ 0.5444, -0.3852, 1.3786, -0.5409, -0.6924, -0.4229, 0.1791,
  11527. 0.2341],
  11528. [ 0.5964, -0.3380, 1.4785, -0.8386, -0.2420, -1.0619, 0.3238,
  11529. 0.4008],
  11530. [ 0.6078, -0.4057, 1.8134, -0.7309, -0.4499, -0.7386, 0.6298,
  11531. 0.1390],
  11532. [ 0.5790, -0.4031, 1.6915, -0.9564, -0.4152, -1.1063, 0.4425,
  11533. 0.2528]]], device='cuda:0')
  11534. loss_train_step before backward: tensor(0.0526, device='cuda:0', grad_fn=<MseLossBackward>)
  11535. loss_train_step after backward: tensor(0.0526, device='cuda:0', grad_fn=<MseLossBackward>)
  11536. loss_train: 4.856406262144446
  11537. step: 85
  11538. running loss: 0.057134191319346425
  11539. Train Steps: 85/90 Loss: 0.0571 torch.Size([8, 600, 800])
  11540. torch.Size([8, 8])
  11541. tensor([[0.6161, 0.4024, 0.8662, 0.4683, 0.4935, 0.5364, 0.6063, 0.5567],
  11542. [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400],
  11543. [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284],
  11544. [0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528],
  11545. [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938],
  11546. [0.6187, 0.4104, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683],
  11547. [0.6171, 0.4127, 0.8900, 0.4800, 0.4325, 0.5783, 0.5769, 0.5090],
  11548. [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617]],
  11549. device='cuda:0', dtype=torch.float64)
  11550. predictions are: tensor([[ 1.3328e-01, -6.7352e-01, 1.6691e+00, -2.4416e-01, -1.3594e-01,
  11551. -3.2876e-04, 4.3535e-01, 2.3712e-01],
  11552. [ 5.2176e-01, -4.1727e-01, 1.4926e+00, -8.6348e-01, -5.5333e-01,
  11553. -1.8318e-01, 7.7127e-01, 2.5558e-01],
  11554. [ 9.9327e-01, -1.1433e-01, 1.0954e+00, -1.3476e+00, -2.7377e-01,
  11555. -1.3365e+00, 3.8144e-01, 1.8049e-01],
  11556. [-8.2795e-01, -1.3359e+00, 1.9219e+00, -7.7653e-01, -3.3671e-01,
  11557. -8.4426e-01, 9.5188e-01, 1.5540e-01],
  11558. [ 4.9690e-01, -4.4354e-01, 1.7715e+00, 3.1300e-02, -6.5161e-01,
  11559. -6.7309e-01, 5.1132e-01, 5.7911e-02],
  11560. [ 7.3540e-01, -2.5406e-01, 9.1125e-01, -1.3418e+00, -4.7364e-01,
  11561. -1.1851e+00, 3.1099e-01, 2.4200e-01],
  11562. [ 4.7442e-01, -4.5498e-01, 1.7000e+00, -1.5574e-01, -2.9064e-01,
  11563. 1.7565e-01, 4.3889e-01, 2.0958e-01],
  11564. [ 6.5780e-01, -3.2612e-01, 1.6180e+00, -1.8878e-01, -3.5431e-01,
  11565. 1.8572e-02, 3.9815e-01, 2.2789e-01]], device='cuda:0',
  11566. grad_fn=<AddmmBackward>)
  11567. landmarks are: tensor([[[ 0.5598, -0.4273, 1.7152, -0.1227, -0.0065, 0.1917, 0.5143,
  11568. 0.2853],
  11569. [ 0.5631, -0.4008, 1.6113, -0.7309, -0.6693, -0.1304, 0.5836,
  11570. 0.2083],
  11571. [ 0.5761, -0.4070, 1.2452, -1.2541, -0.1725, -1.4835, 0.4511,
  11572. 0.1545],
  11573. [ 0.6388, -0.3623, 1.9173, -0.7386, -0.3517, -0.5846, 1.1495,
  11574. 0.2676],
  11575. [ 0.6273, -0.4105, 1.8480, 0.1082, -0.5538, -0.5076, 0.6414,
  11576. -0.0049],
  11577. [ 0.5718, -0.3905, 1.0053, -1.3305, -0.4614, -1.1235, 0.4450,
  11578. 0.3392],
  11579. [ 0.5645, -0.3797, 1.8249, -0.0688, -0.2882, 0.3854, 0.3789,
  11580. 0.0652],
  11581. [ 0.5637, -0.4143, 1.7519, -0.0787, -0.3055, -0.0149, 0.3758,
  11582. 0.3084]]], device='cuda:0')
  11583. loss_train_step before backward: tensor(0.0671, device='cuda:0', grad_fn=<MseLossBackward>)
  11584. loss_train_step after backward: tensor(0.0671, device='cuda:0', grad_fn=<MseLossBackward>)
  11585. loss_train: 4.923488324508071
  11586. step: 86
  11587. running loss: 0.05724986423846594
  11588. Train Steps: 86/90 Loss: 0.0572 torch.Size([8, 600, 800])
  11589. torch.Size([8, 8])
  11590. tensor([[0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123],
  11591. [0.6048, 0.3928, 0.8538, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719],
  11592. [0.6257, 0.4034, 0.8287, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901],
  11593. [0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266],
  11594. [0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343],
  11595. [0.6200, 0.4098, 0.8237, 0.2917, 0.4012, 0.2967, 0.6000, 0.5683],
  11596. [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268],
  11597. [0.6134, 0.4090, 0.6926, 0.2819, 0.3538, 0.3233, 0.5563, 0.5667]],
  11598. device='cuda:0', dtype=torch.float64)
  11599. predictions are: tensor([[ 0.6679, -0.3393, 1.1044, -1.2029, -0.5126, -0.9637, 0.2172, 0.1473],
  11600. [ 0.2462, -0.6525, 1.7013, 0.3707, -0.3928, 0.1506, 0.3154, 0.2182],
  11601. [ 0.2769, -0.6023, 1.5248, -1.0627, -0.4206, -0.9832, 0.6722, 0.0666],
  11602. [ 0.6292, -0.3429, 1.2376, -1.0654, -0.3068, -0.9931, 0.4548, 0.3508],
  11603. [-0.0074, -0.7957, 1.7611, -0.6247, -0.3023, -0.7269, 0.8707, 0.0755],
  11604. [ 0.5369, -0.4352, 1.6622, -0.9727, -0.4015, -0.7603, 0.5024, 0.2316],
  11605. [ 0.2221, -0.6411, 1.7975, -0.0278, -0.1875, 0.1898, 0.6727, 0.1995],
  11606. [ 0.5766, -0.3768, 1.0787, -1.0031, -0.5580, -0.6710, 0.2390, 0.2642]],
  11607. device='cuda:0', grad_fn=<AddmmBackward>)
  11608. landmarks are: tensor([[[ 0.5342, -0.4280, 1.0051, -1.0619, -0.5711, -1.0388, 0.2226,
  11609. 0.0802],
  11610. [ 0.5079, -0.4714, 1.6575, 0.2237, -0.4961, 0.0774, 0.1465,
  11611. -0.1061],
  11612. [ 0.6042, -0.4225, 1.5420, -1.2082, -0.4730, -1.0311, 0.6380,
  11613. -0.0220],
  11614. [ 0.5746, -0.4021, 1.0801, -1.1312, -0.3229, -1.1081, 0.4803,
  11615. 0.6084],
  11616. [ 0.6125, -0.4369, 1.9173, -0.5384, -0.2594, -0.8386, 0.9741,
  11617. 0.1821],
  11618. [ 0.5778, -0.3930, 1.5189, -0.9387, -0.4326, -0.9156, 0.4855,
  11619. 0.3392],
  11620. [ 0.5828, -0.4682, 1.7031, -0.0497, -0.2458, 0.0818, 0.6381,
  11621. 0.1474],
  11622. [ 0.5473, -0.3966, 0.9131, -0.9838, -0.6520, -0.7925, 0.2834,
  11623. 0.3315]]], device='cuda:0')
  11624. loss_train_step before backward: tensor(0.0238, device='cuda:0', grad_fn=<MseLossBackward>)
  11625. loss_train_step after backward: tensor(0.0238, device='cuda:0', grad_fn=<MseLossBackward>)
  11626. loss_train: 4.947322266176343
  11627. step: 87
  11628. running loss: 0.05686577317444073
  11629. Train Steps: 87/90 Loss: 0.0569 torch.Size([8, 600, 800])
  11630. torch.Size([8, 8])
  11631. tensor([[0.6307, 0.4029, 0.8650, 0.5200, 0.3763, 0.4017, 0.7311, 0.5366],
  11632. [0.6357, 0.4097, 0.9038, 0.3883, 0.4213, 0.2950, 0.6686, 0.5390],
  11633. [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117],
  11634. [0.6159, 0.4085, 0.6900, 0.2283, 0.4088, 0.1950, 0.5123, 0.5397],
  11635. [0.6266, 0.4067, 0.8588, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267],
  11636. [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350],
  11637. [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183],
  11638. [ nan, nan, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533]],
  11639. device='cuda:0', dtype=torch.float64)
  11640. predictions are: tensor([[ 0.7241, -0.3271, 1.7330, 0.0022, -0.4415, -0.4059, 0.5559, 0.1113],
  11641. [ 0.8021, -0.2346, 1.5974, -0.6172, -0.2946, -0.7826, 0.7004, 0.1298],
  11642. [ 0.5272, -0.4370, 1.6671, -0.8570, -0.4961, -0.7450, 0.4616, 0.1232],
  11643. [-0.0215, -0.7598, 0.7726, -1.3853, -0.3094, -1.2399, 0.1006, 0.2363],
  11644. [ 0.6110, -0.3637, 1.6312, -1.1276, -0.2190, -0.9056, 0.7891, 0.1843],
  11645. [ 0.5471, -0.3935, 1.6186, -0.3462, -0.4808, -0.3456, 0.2056, 0.2678],
  11646. [ 1.0903, -0.1401, 1.7994, -0.1475, -0.3498, 0.5473, 0.7149, 0.1603],
  11647. [-1.3033, -1.6128, 1.1779, -0.9719, -0.4504, -0.8305, 0.2302, 0.2690]],
  11648. device='cuda:0', grad_fn=<AddmmBackward>)
  11649. landmarks are: tensor([[[ 0.6273, -0.4249, 1.7095, 0.1159, -0.5480, -0.4306, 1.0910,
  11650. 0.1928],
  11651. [ 0.6504, -0.3936, 1.8885, -0.4922, -0.3402, -0.9233, 0.8022,
  11652. 0.2035],
  11653. [ 0.5799, -0.4329, 1.7210, -0.7694, -0.5711, -0.8771, 0.3988,
  11654. 0.0774],
  11655. [ 0.5591, -0.3990, 0.9012, -1.2313, -0.3979, -1.3852, 0.0804,
  11656. 0.2071],
  11657. [ 0.6084, -0.4076, 1.6806, -0.9618, -0.2998, -0.9695, 0.6356,
  11658. 0.1467],
  11659. [ 0.6072, -0.3239, 1.8423, -0.3537, -0.4961, -0.3921, 0.2083,
  11660. 0.1852],
  11661. [ 0.5934, -0.4276, 1.8192, -0.1458, -0.4499, 0.3777, 0.6702,
  11662. 0.1082],
  11663. [-2.2859, -2.2859, 1.5074, -1.0388, -0.5423, -0.9849, 0.2199,
  11664. 0.2699]]], device='cuda:0')
  11665. loss_train_step before backward: tensor(0.0496, device='cuda:0', grad_fn=<MseLossBackward>)
  11666. loss_train_step after backward: tensor(0.0496, device='cuda:0', grad_fn=<MseLossBackward>)
  11667. loss_train: 4.99695965833962
  11668. step: 88
  11669. running loss: 0.056783632481132044
  11670.  
  11671. Train Steps: 88/90 Loss: 0.0568 torch.Size([8, 600, 800])
  11672. torch.Size([8, 8])
  11673. tensor([[0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5787, 0.5117],
  11674. [0.6129, 0.4069, 0.8750, 0.5067, 0.3875, 0.4233, 0.5235, 0.5881],
  11675. [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466],
  11676. [0.6261, 0.4045, 0.8865, 0.5369, 0.3895, 0.4859, 0.6683, 0.5249],
  11677. [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833],
  11678. [0.6219, 0.4097, 0.8738, 0.3400, 0.3563, 0.4117, 0.5975, 0.5683],
  11679. [0.6346, 0.4144, 0.9088, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899],
  11680. [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879]],
  11681. device='cuda:0', dtype=torch.float64)
  11682. predictions are: tensor([[ 0.5584, -0.4093, 1.1146, -1.4048, -0.2779, -1.3822, 0.1954, 0.1406],
  11683. [ 0.4821, -0.4613, 1.6472, -0.0517, -0.4324, -0.3837, 0.1814, 0.2442],
  11684. [ 0.4873, -0.4168, 1.3685, -0.8206, -0.5818, -0.4522, 0.2436, 0.1914],
  11685. [ 0.3995, -0.5603, 1.7979, -0.0745, -0.4099, -0.0866, 0.7555, 0.1212],
  11686. [ 0.0065, -0.7383, 0.9568, -1.1590, -0.3999, -1.0661, 0.2975, 0.3177],
  11687. [ 0.6278, -0.4072, 1.7102, -0.8546, -0.5490, -0.6138, 0.4779, 0.2245],
  11688. [ 0.0327, -0.7839, 1.8877, -0.2893, -0.2485, -0.4808, 0.8511, 0.2126],
  11689. [ 0.7731, -0.3011, 1.7908, -0.3452, -0.3399, 0.2271, 0.5904, 0.0849]],
  11690. device='cuda:0', grad_fn=<AddmmBackward>)
  11691. landmarks are: tensor([[[ 0.5598, -0.4201, 1.1898, -1.3005, -0.3806, -1.3313, 0.3873,
  11692. 0.0774],
  11693. [ 0.5452, -0.4067, 1.7557, 0.0543, -0.4961, -0.3306, 0.1323,
  11694. 0.4306],
  11695. [ 0.5076, -0.4443, 1.2337, -0.5023, -0.6808, -0.3614, 0.0866,
  11696. 0.2386],
  11697. [ 0.6059, -0.4177, 1.8087, 0.1941, -0.4868, -0.0414, 0.8010,
  11698. 0.1385],
  11699. [ 0.5614, -0.4080, 0.7741, -0.8848, -0.5423, -0.9156, 0.3584,
  11700. 0.4085],
  11701. [ 0.5867, -0.3937, 1.7499, -0.7155, -0.6404, -0.3844, 0.4739,
  11702. 0.3392],
  11703. [ 0.6454, -0.3719, 1.9115, -0.1304, -0.5076, -0.2844, 1.0033,
  11704. 0.4386],
  11705. [ 0.5676, -0.4327, 1.8252, -0.2095, -0.4886, 0.3084, 0.4727,
  11706. -0.0322]]], device='cuda:0')
  11707. loss_train_step before backward: tensor(0.0316, device='cuda:0', grad_fn=<MseLossBackward>)
  11708. loss_train_step after backward: tensor(0.0316, device='cuda:0', grad_fn=<MseLossBackward>)
  11709. loss_train: 5.028513407334685
  11710. step: 89
  11711. running loss: 0.05650015064420995
  11712. Train Steps: 89/90 Loss: 0.0565 torch.Size([8, 600, 800])
  11713. torch.Size([8, 8])
  11714. tensor([[0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5363, 0.5550],
  11715. [0.6264, 0.4049, 0.8988, 0.4633, 0.3813, 0.4983, 0.6326, 0.4843],
  11716. [0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626],
  11717. [0.6148, 0.4076, 0.8666, 0.4820, 0.4138, 0.5067, 0.5250, 0.5767],
  11718. [0.6339, 0.4123, 0.8638, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436],
  11719. [0.6241, 0.4143, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550],
  11720. [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220],
  11721. [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100]],
  11722. device='cuda:0', dtype=torch.float64)
  11723. predictions are: tensor([[ 0.4825, -0.4272, 0.9730, -1.4118, -0.4009, -1.3373, 0.0767, 0.2495],
  11724. [ 0.4915, -0.4617, 1.9515, -0.4588, -0.5543, -0.2200, 0.6500, 0.0742],
  11725. [ 0.3416, -0.6033, 1.6379, -0.0055, -0.4627, -0.4151, 0.4429, 0.1592],
  11726. [ 0.4801, -0.4831, 1.6780, -0.1948, -0.3423, -0.0419, 0.2214, 0.2790],
  11727. [ 0.6173, -0.4038, 1.7523, -0.1950, -0.3700, 0.1127, 0.7578, 0.2431],
  11728. [ 0.5661, -0.4204, 1.8014, -0.4218, -0.4077, 0.0485, 0.4907, 0.2373],
  11729. [ 0.0555, -0.7193, 0.9680, -1.5541, -0.2692, -1.6329, 0.2577, 0.1827],
  11730. [ 0.4184, -0.5378, 1.6813, -0.1549, -0.4171, -0.3171, 0.5697, 0.1647]],
  11731. device='cuda:0', grad_fn=<AddmmBackward>)
  11732. landmarks are: tensor([[[ 0.5448, -0.3859, 0.9242, -1.1466, -0.4152, -1.3005, 0.1910,
  11733. 0.2776],
  11734. [ 0.6075, -0.4159, 1.8654, -0.1458, -0.5249, 0.0159, 0.6359,
  11735. -0.0490],
  11736. [ 0.6105, -0.4293, 1.5824, 0.5239, -0.4730, -0.0380, 0.5025,
  11737. -0.1492],
  11738. [ 0.5539, -0.4031, 1.7168, -0.0594, -0.3748, 0.0543, 0.1390,
  11739. 0.3777],
  11740. [ 0.6421, -0.3816, 1.7037, 0.1929, -0.4037, 0.2391, 1.1861,
  11741. 0.2249],
  11742. [ 0.5968, -0.3725, 1.8423, -0.1381, -0.4037, 0.1852, 0.6009,
  11743. 0.2776],
  11744. [ 0.5566, -0.4160, 0.9935, -1.3313, -0.2824, -1.5161, 0.2144,
  11745. 0.1253],
  11746. [ 0.6125, -0.4273, 1.6402, 0.2006, -0.4788, -0.0303, 0.6182,
  11747. 0.0697]]], device='cuda:0')
  11748. loss_train_step before backward: tensor(0.0333, device='cuda:0', grad_fn=<MseLossBackward>)
  11749. loss_train_step after backward: tensor(0.0333, device='cuda:0', grad_fn=<MseLossBackward>)
  11750. loss_train: 5.0618322510272264
  11751. step: 90
  11752. running loss: 0.05624258056696918
  11753. Valid Steps: 10/10 Loss: nan 62
  11754. --------------------------------------------------
  11755. Epoch: 3 Train Loss: 0.0562 Valid Loss: nan
  11756. --------------------------------------------------
  11757. size of train loader is: 90
  11758. torch.Size([8, 600, 800])
  11759. torch.Size([8, 8])
  11760. tensor([[0.6136, 0.4117, 0.8700, 0.5167, 0.4188, 0.5083, 0.5147, 0.5495],
  11761. [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5787, 0.5117],
  11762. [0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166],
  11763. [ nan, nan, 0.8300, 0.3150, 0.3588, 0.3383, 0.5208, 0.5194],
  11764. [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355],
  11765. [0.6268, 0.4029, 0.8500, 0.2683, 0.3937, 0.3500, 0.6860, 0.5297],
  11766. [0.6263, 0.4039, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804],
  11767. [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567]],
  11768. device='cuda:0', dtype=torch.float64)
  11769. predictions are: tensor([[ 0.7934, -0.2856, 1.6860, 0.1092, -0.3487, 0.0531, 0.2048, 0.2442],
  11770. [ 0.6889, -0.3477, 1.0870, -1.3416, -0.3528, -1.3638, 0.1548, 0.2013],
  11771. [ 0.5929, -0.4533, 1.6458, -0.0300, -0.2567, 0.1042, 0.5948, 0.2456],
  11772. [-1.2841, -1.6451, 1.3930, -0.9300, -0.5634, -0.8736, 0.1785, 0.2065],
  11773. [ 0.2508, -0.6173, 1.6443, -0.9053, -0.2038, -0.8371, 0.8204, 0.1669],
  11774. [ 0.7218, -0.3298, 1.5397, -1.0182, -0.5050, -0.7588, 0.8013, 0.1396],
  11775. [ 0.9012, -0.2219, 1.7789, -0.3461, -0.6196, -0.1334, 0.4337, 0.0805],
  11776. [ 0.8209, -0.2396, 1.5937, -0.2505, -0.5622, -0.4064, 0.3184, 0.2603]],
  11777. device='cuda:0', grad_fn=<AddmmBackward>)
  11778. landmarks are: tensor([[[ 0.5482, -0.3841, 1.7326, 0.1005, -0.3517, 0.0620, 0.0912,
  11779. 0.2522],
  11780. [ 0.5598, -0.4201, 1.1898, -1.3005, -0.3806, -1.3313, 0.3873,
  11781. 0.0774],
  11782. [ 0.6042, -0.4273, 1.6920, 0.1860, -0.2717, 0.1406, 0.7997,
  11783. 0.1004],
  11784. [-2.2859, -2.2859, 1.5478, -0.8309, -0.6289, -0.7232, 0.1198,
  11785. 0.1133],
  11786. [ 0.6224, -0.4105, 1.9173, -0.7771, -0.1030, -0.7308, 1.1532,
  11787. 0.1875],
  11788. [ 0.6092, -0.4249, 1.6402, -1.0465, -0.4672, -0.6693, 0.8827,
  11789. 0.1608],
  11790. [ 0.6072, -0.4206, 1.8711, -0.2536, -0.6115, -0.1304, 0.6812,
  11791. -0.0670],
  11792. [ 0.5432, -0.4336, 1.7095, -0.1766, -0.5942, -0.4845, 0.3007,
  11793. 0.2853]]], device='cuda:0')
  11794. loss_train_step before backward: tensor(0.0408, device='cuda:0', grad_fn=<MseLossBackward>)
  11795. loss_train_step after backward: tensor(0.0408, device='cuda:0', grad_fn=<MseLossBackward>)
  11796. loss_train: 0.0408310666680336
  11797. step: 1
  11798. running loss: 0.0408310666680336
  11799. Train Steps: 1/90 Loss: 0.0408 torch.Size([8, 600, 800])
  11800. torch.Size([8, 8])
  11801. tensor([[0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6088, 0.5583],
  11802. [0.6203, 0.4096, 0.8862, 0.4267, 0.3538, 0.4117, 0.6025, 0.5650],
  11803. [ nan, nan, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650],
  11804. [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100],
  11805. [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683],
  11806. [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679],
  11807. [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750],
  11808. [0.6199, 0.4102, 0.8950, 0.4417, 0.4012, 0.5367, 0.6112, 0.5967]],
  11809. device='cuda:0', dtype=torch.float64)
  11810. predictions are: tensor([[ 0.8029, -0.2294, 1.0461, -1.3274, -0.4808, -1.3214, 0.4222, 0.1729],
  11811. [ 0.5322, -0.4257, 1.7850, -0.4723, -0.6244, -0.5832, 0.4501, 0.2225],
  11812. [-1.7818, -1.9189, 1.1634, -1.3966, -0.2579, -1.3634, 0.2830, 0.2149],
  11813. [ 0.7616, -0.3080, 1.6486, -0.0197, -0.4939, -0.2841, 0.5924, 0.1066],
  11814. [ 0.7523, -0.2829, 1.6721, -0.1539, -0.2369, -0.1656, 0.1428, 0.1686],
  11815. [ 0.8615, -0.2260, 1.7055, -0.0355, -0.3871, 0.1315, 0.7918, 0.1862],
  11816. [ 0.6541, -0.3519, 1.6196, -0.1458, -0.3515, -0.2593, 0.2668, 0.1777],
  11817. [ 0.7497, -0.3061, 1.7884, -0.3122, -0.4508, 0.1585, 0.5595, 0.2249]],
  11818. device='cuda:0', grad_fn=<AddmmBackward>)
  11819. landmarks are: tensor([[[ 0.5992, -0.3968, 0.9307, -1.3497, -0.4730, -1.0465, 0.5259,
  11820. 0.2930],
  11821. [ 0.5790, -0.3940, 1.8076, -0.3152, -0.6520, -0.3844, 0.4970,
  11822. 0.3238],
  11823. [-2.2859, -2.2859, 0.9175, -1.3947, -0.3691, -1.2467, 0.2314,
  11824. 0.3238],
  11825. [ 0.6125, -0.4273, 1.6402, 0.2006, -0.4788, -0.0303, 0.6182,
  11826. 0.0697],
  11827. [ 0.5359, -0.4370, 1.7095, -0.0303, -0.0804, -0.0380, 0.1044,
  11828. 0.3392],
  11829. [ 0.6257, -0.4249, 1.5998, 0.1236, -0.3806, 0.3084, 0.9887,
  11830. 0.3371],
  11831. [ 0.5525, -0.4463, 1.7326, 0.0313, -0.2536, -0.0688, 0.1968,
  11832. 0.3700],
  11833. [ 0.5772, -0.3913, 1.8480, -0.2459, -0.4326, 0.1929, 0.5374,
  11834. 0.4701]]], device='cuda:0')
  11835. loss_train_step before backward: tensor(0.0242, device='cuda:0', grad_fn=<MseLossBackward>)
  11836. loss_train_step after backward: tensor(0.0242, device='cuda:0', grad_fn=<MseLossBackward>)
  11837. loss_train: 0.06501925364136696
  11838. step: 2
  11839. running loss: 0.03250962682068348
  11840.  
  11841. Train Steps: 2/90 Loss: 0.0325 torch.Size([8, 600, 800])
  11842. torch.Size([8, 8])
  11843. tensor([[0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650],
  11844. [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461],
  11845. [0.6192, 0.4128, 0.8513, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633],
  11846. [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649],
  11847. [0.6268, 0.4102, 0.8938, 0.3667, 0.4025, 0.2833, 0.6275, 0.5183],
  11848. [0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240],
  11849. [0.6161, 0.4099, 0.8738, 0.4383, 0.3788, 0.5483, 0.5605, 0.5019],
  11850. [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268]],
  11851. device='cuda:0', dtype=torch.float64)
  11852. predictions are: tensor([[ 0.2261, -0.6299, 1.5691, 0.0380, -0.4303, -0.4064, 0.3217, 0.2605],
  11853. [ 0.8775, -0.2186, 1.6144, -0.0903, -0.2743, -0.0483, 0.2540, 0.2187],
  11854. [ 0.4941, -0.4669, 1.6542, 0.0592, -0.4179, 0.0067, 0.4888, 0.2468],
  11855. [ 0.6490, -0.3877, 1.5852, 0.0381, -0.5231, -0.2220, 0.4692, 0.1378],
  11856. [ 0.6293, -0.3672, 1.8195, -0.9589, -0.4476, -1.1585, 0.6918, 0.1580],
  11857. [-0.9213, -1.3341, 1.0464, -1.5667, -0.3857, -1.5674, 0.3233, 0.1719],
  11858. [ 0.8097, -0.2308, 1.6492, -0.5242, -0.5474, 0.1092, 0.4527, 0.1488],
  11859. [ 0.6409, -0.3545, 1.6522, -0.3213, -0.3163, -0.0979, 0.5912, 0.1635]],
  11860. device='cuda:0', grad_fn=<AddmmBackward>)
  11861. landmarks are: tensor([[[ 0.5770, -0.4036, 1.7095, 0.3084, -0.3691, -0.2690, 0.2314,
  11862. 0.3238],
  11863. [ 0.5443, -0.3831, 1.7095, 0.1621, -0.2016, 0.1390, 0.1437,
  11864. 0.2364],
  11865. [ 0.5742, -0.3792, 1.6460, 0.3084, -0.3460, 0.1467, 0.4162,
  11866. 0.3161],
  11867. [ 0.6105, -0.4360, 1.6171, 0.5162, -0.5076, 0.0159, 0.5171,
  11868. -0.1385],
  11869. [ 0.6092, -0.3913, 1.8423, -0.5923, -0.4268, -0.9772, 0.6125,
  11870. 0.1082],
  11871. [ 0.5676, -0.4112, 1.1898, -1.2467, -0.2940, -1.4622, 0.2103,
  11872. 0.1343],
  11873. [ 0.5597, -0.3928, 1.7499, -0.2613, -0.5365, 0.2468, 0.3028,
  11874. 0.0321],
  11875. [ 0.5828, -0.4682, 1.7031, -0.0497, -0.2458, 0.0818, 0.6381,
  11876. 0.1474]]], device='cuda:0')
  11877. loss_train_step before backward: tensor(0.0751, device='cuda:0', grad_fn=<MseLossBackward>)
  11878. loss_train_step after backward: tensor(0.0751, device='cuda:0', grad_fn=<MseLossBackward>)
  11879. loss_train: 0.1400858722627163
  11880. step: 3
  11881. running loss: 0.04669529075423876
  11882. Train Steps: 3/90 Loss: 0.0467 torch.Size([8, 600, 800])
  11883. torch.Size([8, 8])
  11884. tensor([[0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343],
  11885. [0.6201, 0.4102, 0.7288, 0.2417, 0.4150, 0.2383, 0.6100, 0.5500],
  11886. [0.6213, 0.4001, 0.7712, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183],
  11887. [0.6250, 0.4236, 0.8638, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799],
  11888. [0.6189, 0.4033, 0.8650, 0.5267, 0.4487, 0.5150, 0.5925, 0.5050],
  11889. [0.6058, 0.3986, 0.8324, 0.4626, 0.3838, 0.4983, 0.5147, 0.5466],
  11890. [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5687, 0.5567],
  11891. [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750]],
  11892. device='cuda:0', dtype=torch.float64)
  11893. predictions are: tensor([[ 0.3854, -0.5699, 1.7729, -0.4065, -0.3826, -0.5799, 0.8162, 0.0487],
  11894. [ 0.4473, -0.4987, 1.2553, -0.9025, -0.4371, -1.0451, 0.5099, 0.1875],
  11895. [ 0.3850, -0.5415, 1.2425, -1.1285, -0.3311, -1.2141, 0.4639, 0.1034],
  11896. [ 0.4943, -0.4585, 1.7603, -0.3231, -0.4710, -0.5697, 0.3823, 0.2806],
  11897. [ 0.5609, -0.4599, 1.7936, 0.3726, -0.2832, 0.3955, 0.2943, 0.1781],
  11898. [ 0.3167, -0.5832, 1.6743, 0.1032, -0.5132, 0.2972, 0.2819, 0.1813],
  11899. [ 0.0717, -0.6920, 1.2379, -0.9312, -0.3837, -0.9721, 0.3222, 0.3046],
  11900. [ 0.5122, -0.4441, 1.2345, -0.7984, -0.4852, -0.8042, 0.5456, 0.2060]],
  11901. device='cuda:0', grad_fn=<AddmmBackward>)
  11902. landmarks are: tensor([[[ 0.6125, -0.4369, 1.9173, -0.5384, -0.2594, -0.8386, 0.9741,
  11903. 0.1821],
  11904. [ 0.5784, -0.3913, 1.0801, -1.1697, -0.3691, -1.1851, 0.5316,
  11905. 0.2545],
  11906. [ 0.5838, -0.4379, 1.2764, -1.3082, -0.2824, -1.4545, 0.4162,
  11907. 0.1082],
  11908. [ 0.6008, -0.3293, 1.7037, -0.5461, -0.4152, -0.8309, 0.3234,
  11909. 0.3928],
  11910. [ 0.5726, -0.4230, 1.7095, 0.1467, -0.2132, 0.0928, 0.4508,
  11911. 0.0467],
  11912. [ 0.5124, -0.4446, 1.5587, -0.1493, -0.5134, 0.0159, 0.0912,
  11913. 0.2386],
  11914. [ 0.5824, -0.3625, 1.0405, -0.9325, -0.4268, -1.1389, 0.3411,
  11915. 0.2853],
  11916. [ 0.5805, -0.3818, 1.0282, -1.1774, -0.4903, -0.9310, 0.5894,
  11917. 0.3700]]], device='cuda:0')
  11918. loss_train_step before backward: tensor(0.0287, device='cuda:0', grad_fn=<MseLossBackward>)
  11919. loss_train_step after backward: tensor(0.0287, device='cuda:0', grad_fn=<MseLossBackward>)
  11920. loss_train: 0.16878480091691017
  11921. step: 4
  11922. running loss: 0.04219620022922754
  11923. Train Steps: 4/90 Loss: 0.0422 torch.Size([8, 600, 800])
  11924. torch.Size([8, 8])
  11925. tensor([[0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5988, 0.5667],
  11926. [0.6199, 0.4102, 0.8950, 0.4417, 0.4012, 0.5367, 0.6112, 0.5967],
  11927. [0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626],
  11928. [0.6134, 0.4090, 0.6926, 0.2819, 0.3538, 0.3233, 0.5563, 0.5667],
  11929. [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450],
  11930. [0.6260, 0.4153, 0.9000, 0.4533, 0.4025, 0.2633, 0.6223, 0.4967],
  11931. [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5463, 0.5800],
  11932. [0.6219, 0.4097, 0.8738, 0.3400, 0.3563, 0.4117, 0.5975, 0.5683]],
  11933. device='cuda:0', dtype=torch.float64)
  11934. predictions are: tensor([[ 0.6402, -0.3300, 1.6367, -0.5436, -0.4823, -0.6285, 0.3887, 0.2062],
  11935. [ 0.4304, -0.5247, 1.8311, -0.0857, -0.2697, 0.3819, 0.6370, 0.2235],
  11936. [ 0.3012, -0.6414, 1.6228, 0.3746, -0.3542, -0.0797, 0.5618, 0.0932],
  11937. [ 0.3471, -0.5312, 1.0608, -0.9703, -0.4720, -0.8061, 0.3146, 0.2213],
  11938. [ 0.3248, -0.5650, 1.7027, -0.1101, -0.4417, -0.2336, 0.5059, 0.1623],
  11939. [ 0.3386, -0.5307, 1.7406, -0.2570, -0.3301, -1.0203, 0.5861, 0.1280],
  11940. [ 0.3174, -0.5268, 1.0960, -0.8999, -0.4709, -0.9923, 0.1387, 0.2535],
  11941. [ 0.6184, -0.4022, 1.6616, -0.6118, -0.5818, -0.3922, 0.4269, 0.2213]],
  11942. device='cuda:0', grad_fn=<AddmmBackward>)
  11943. landmarks are: tensor([[[ 0.5783, -0.3933, 1.6748, -0.6154, -0.5769, -0.6462, 0.4797,
  11944. 0.3315],
  11945. [ 0.5772, -0.3913, 1.8480, -0.2459, -0.4326, 0.1929, 0.5374,
  11946. 0.4701],
  11947. [ 0.6105, -0.4293, 1.5824, 0.5239, -0.4730, -0.0380, 0.5025,
  11948. -0.1492],
  11949. [ 0.5473, -0.3966, 0.9131, -0.9838, -0.6520, -0.7925, 0.2834,
  11950. 0.3315],
  11951. [ 0.5785, -0.4252, 1.7676, -0.1602, -0.5845, -0.3446, 0.4566,
  11952. 0.2314],
  11953. [ 0.6055, -0.3676, 1.8711, -0.1920, -0.4268, -1.0696, 0.5887,
  11954. 0.0081],
  11955. [ 0.5548, -0.3936, 1.1634, -0.8105, -0.5192, -1.0696, 0.2372,
  11956. 0.3931],
  11957. [ 0.5867, -0.3937, 1.7499, -0.7155, -0.6404, -0.3844, 0.4739,
  11958. 0.3392]]], device='cuda:0')
  11959. loss_train_step before backward: tensor(0.0165, device='cuda:0', grad_fn=<MseLossBackward>)
  11960. loss_train_step after backward: tensor(0.0165, device='cuda:0', grad_fn=<MseLossBackward>)
  11961. loss_train: 0.18524960055947304
  11962. step: 5
  11963. running loss: 0.037049920111894605
  11964. Train Steps: 5/90 Loss: 0.0370 torch.Size([8, 600, 800])
  11965. torch.Size([8, 8])
  11966. tensor([[0.6234, 0.4179, 0.7825, 0.3450, 0.3813, 0.2867, 0.5675, 0.5617],
  11967. [0.6230, 0.4152, 0.7588, 0.2283, 0.4012, 0.2883, 0.6200, 0.5767],
  11968. [0.6085, 0.4008, 0.8588, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283],
  11969. [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012],
  11970. [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986],
  11971. [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202],
  11972. [0.6200, 0.4071, 0.7338, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517],
  11973. [0.6229, 0.4198, 0.7662, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783]],
  11974. device='cuda:0', dtype=torch.float64)
  11975. predictions are: tensor([[ 0.2791, -0.5424, 1.3593, -0.6483, -0.5082, -0.7606, 0.2373, 0.3004],
  11976. [ 0.5529, -0.4208, 1.3162, -0.7741, -0.5420, -0.8749, 0.5742, 0.2522],
  11977. [ 0.5527, -0.4374, 1.6764, 0.2120, -0.1778, 0.1175, 0.2001, 0.1639],
  11978. [ 0.7011, -0.3902, 1.7922, 0.3820, -0.5591, 0.0326, 0.7921, 0.0431],
  11979. [ 0.1883, -0.6151, 1.7248, -0.4770, -0.3421, -1.0000, 0.5109, 0.1547],
  11980. [ 0.2386, -0.6325, 1.7365, 0.1798, -0.4045, 0.3961, 0.4558, 0.1295],
  11981. [ 0.3358, -0.5460, 1.2014, -1.0744, -0.4078, -1.0060, 0.5095, 0.1908],
  11982. [ 0.3464, -0.4803, 1.1684, -0.9478, -0.2323, -1.0542, 0.3087, 0.2845]],
  11983. device='cuda:0', grad_fn=<AddmmBackward>)
  11984. landmarks are: tensor([[[ 0.5935, -0.3558, 1.3284, -0.6924, -0.5249, -0.9618, 0.3353,
  11985. 0.3084],
  11986. [ 0.5915, -0.3682, 1.2187, -1.2313, -0.4326, -0.9541, 0.5778,
  11987. 0.3777],
  11988. [ 0.5245, -0.4347, 1.6806, 0.1159, 0.0046, 0.0129, 0.1044,
  11989. 0.1544],
  11990. [ 0.6249, -0.4352, 1.8018, 0.2545, -0.6173, -0.1997, 0.6401,
  11991. 0.0291],
  11992. [ 0.6054, -0.3641, 1.8711, -0.5153, -0.2305, -1.3005, 0.5948,
  11993. 0.0171],
  11994. [ 0.5614, -0.4586, 1.7587, 0.0641, -0.2998, 0.1712, 0.4958,
  11995. 0.1170],
  11996. [ 0.5779, -0.4054, 1.1032, -1.4006, -0.3460, -1.1543, 0.5547,
  11997. 0.2622],
  11998. [ 0.5912, -0.3467, 1.2533, -1.0388, -0.1150, -1.3005, 0.3353,
  11999. 0.3854]]], device='cuda:0')
  12000. loss_train_step before backward: tensor(0.0253, device='cuda:0', grad_fn=<MseLossBackward>)
  12001. loss_train_step after backward: tensor(0.0253, device='cuda:0', grad_fn=<MseLossBackward>)
  12002. loss_train: 0.21053649485111237
  12003. step: 6
  12004. running loss: 0.03508941580851873
  12005.  
  12006. Train Steps: 6/90 Loss: 0.0351 torch.Size([8, 600, 800])
  12007. torch.Size([8, 8])
  12008. tensor([[0.6254, 0.3993, 0.8988, 0.4767, 0.3987, 0.5517, 0.6955, 0.5285],
  12009. [0.6095, 0.3970, 0.8688, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940],
  12010. [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960],
  12011. [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285],
  12012. [0.6208, 0.4082, 0.8538, 0.3067, 0.3588, 0.3717, 0.6112, 0.5517],
  12013. [0.6177, 0.4086, 0.8738, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700],
  12014. [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356],
  12015. [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167]],
  12016. device='cuda:0', dtype=torch.float64)
  12017. predictions are: tensor([[ 0.3029, -0.6244, 1.8468, 0.1460, -0.4225, 0.1632, 0.9719, 0.1582],
  12018. [ 0.3994, -0.5348, 1.6714, 0.0397, -0.2388, -0.1771, 0.2345, 0.1581],
  12019. [ 0.2754, -0.6236, 1.7249, 0.0636, -0.3578, -0.0518, 0.3006, 0.1270],
  12020. [ 0.6174, -0.3559, 1.3701, -0.3050, -0.6034, -0.3570, 0.1214, 0.1833],
  12021. [ 0.5716, -0.4054, 1.5687, -0.6477, -0.6942, -0.6264, 0.4524, 0.2152],
  12022. [ 0.2340, -0.6259, 1.7155, -0.2797, -0.4494, 0.2091, 0.7213, 0.2117],
  12023. [ 0.6382, -0.3643, 1.3742, -0.9335, -0.3079, -1.3957, 0.6677, 0.2151],
  12024. [ 0.5772, -0.3455, 1.0543, -0.8894, -0.1617, -1.3255, 0.2601, 0.3533]],
  12025. device='cuda:0', grad_fn=<AddmmBackward>)
  12026. landmarks are: tensor([[[ 0.6026, -0.4417, 1.8654, -0.0842, -0.4441, 0.2622, 0.9265,
  12027. 0.1554],
  12028. [ 0.5292, -0.4520, 1.7268, -0.0842, -0.0413, -0.0324, 0.1116,
  12029. -0.0039],
  12030. [ 0.5205, -0.4577, 1.7326, -0.1304, -0.1862, 0.0447, 0.1301,
  12031. 0.0051],
  12032. [ 0.5253, -0.4296, 1.3084, -0.3098, -0.6115, -0.2767, 0.0928,
  12033. 0.1552],
  12034. [ 0.5814, -0.4003, 1.6575, -0.8694, -0.6289, -0.5692, 0.5374,
  12035. 0.2622],
  12036. [ 0.5671, -0.3988, 1.7499, -0.4614, -0.5423, 0.3007, 0.5894,
  12037. 0.3469],
  12038. [ 0.6174, -0.4225, 1.4975, -1.1709, -0.3174, -1.1806, 0.6539,
  12039. 0.1879],
  12040. [ 0.6161, -0.3075, 1.1678, -0.6500, 0.0813, -1.4006, 0.2545,
  12041. 0.5624]]], device='cuda:0')
  12042. loss_train_step before backward: tensor(0.0187, device='cuda:0', grad_fn=<MseLossBackward>)
  12043. loss_train_step after backward: tensor(0.0187, device='cuda:0', grad_fn=<MseLossBackward>)
  12044. loss_train: 0.22919072955846786
  12045. step: 7
  12046. running loss: 0.03274153279406684
  12047. Train Steps: 7/90 Loss: 0.0327 torch.Size([8, 600, 800])
  12048. torch.Size([8, 8])
  12049. tensor([[0.6250, 0.4103, 0.8950, 0.4400, 0.3912, 0.5650, 0.6050, 0.5133],
  12050. [0.6122, 0.4006, 0.8850, 0.4217, 0.4088, 0.5517, 0.6063, 0.5517],
  12051. [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100],
  12052. [0.6212, 0.4171, 0.7875, 0.3633, 0.3813, 0.2933, 0.5675, 0.5700],
  12053. [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760],
  12054. [ nan, nan, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650],
  12055. [0.6148, 0.3996, 0.8488, 0.3867, 0.3488, 0.4067, 0.5863, 0.5000],
  12056. [0.6147, 0.4081, 0.8538, 0.3400, 0.3663, 0.3150, 0.5142, 0.4875]],
  12057. device='cuda:0', dtype=torch.float64)
  12058. predictions are: tensor([[ 0.5962, -0.3807, 1.8008, -0.0904, -0.3684, 0.2387, 0.6538, 0.1876],
  12059. [ 0.5492, -0.4183, 1.7119, -0.1476, -0.3619, 0.1762, 0.4529, 0.2305],
  12060. [ 0.7731, -0.3102, 1.6519, 0.2326, -0.3488, -0.1404, 0.5892, 0.1673],
  12061. [ 0.7477, -0.2326, 1.3266, -0.5177, -0.3959, -0.8787, 0.3715, 0.3245],
  12062. [ 0.8838, -0.2180, 1.6062, 0.3130, -0.3376, -0.1118, 0.7455, 0.2682],
  12063. [-1.4788, -1.7508, 1.1179, -1.3049, -0.2588, -1.3062, 0.2117, 0.2637],
  12064. [ 0.8055, -0.2542, 1.6588, -0.3911, -0.6106, -0.3597, 0.4233, 0.1606],
  12065. [ 0.9708, -0.1498, 1.6112, -0.6890, -0.6115, -0.9635, 0.1644, 0.1881]],
  12066. device='cuda:0', grad_fn=<AddmmBackward>)
  12067. landmarks are: tensor([[[ 0.6010, -0.3909, 1.8480, -0.2536, -0.4788, 0.3238, 0.5085,
  12068. 0.0851],
  12069. [ 0.5417, -0.4355, 1.8018, -0.3383, -0.3979, 0.2622, 0.5143,
  12070. 0.2622],
  12071. [ 0.6125, -0.4273, 1.6402, 0.2006, -0.4788, -0.0303, 0.6182,
  12072. 0.0697],
  12073. [ 0.5832, -0.3593, 1.3515, -0.6077, -0.5249, -0.9310, 0.3353,
  12074. 0.3469],
  12075. [ 0.6421, -0.4008, 1.6055, 0.2160, -0.5076, -0.0534, 1.1020,
  12076. 0.3745],
  12077. [-2.2859, -2.2859, 0.9175, -1.3947, -0.3691, -1.2467, 0.2314,
  12078. 0.3238],
  12079. [ 0.5538, -0.4400, 1.6344, -0.4999, -0.6751, -0.4075, 0.4219,
  12080. 0.0236],
  12081. [ 0.5532, -0.4008, 1.6575, -0.7155, -0.5942, -0.8309, 0.0890,
  12082. -0.0340]]], device='cuda:0')
  12083. loss_train_step before backward: tensor(0.0315, device='cuda:0', grad_fn=<MseLossBackward>)
  12084. loss_train_step after backward: tensor(0.0315, device='cuda:0', grad_fn=<MseLossBackward>)
  12085. loss_train: 0.2607172168791294
  12086. step: 8
  12087. running loss: 0.032589652109891176
  12088. Train Steps: 8/90 Loss: 0.0326 torch.Size([8, 600, 800])
  12089. torch.Size([8, 8])
  12090. tensor([[0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528],
  12091. [0.6346, 0.4165, 0.9138, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471],
  12092. [ nan, nan, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609],
  12093. [0.6127, 0.4115, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495],
  12094. [0.6083, 0.3957, 0.8638, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980],
  12095. [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350],
  12096. [0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633],
  12097. [ nan, nan, 0.7335, 0.2569, 0.3788, 0.2667, 0.5066, 0.5578]],
  12098. device='cuda:0', dtype=torch.float64)
  12099. predictions are: tensor([[ 0.9991, -0.1294, 1.8289, -0.3092, -0.4135, -0.4943, 0.7495, 0.1788],
  12100. [ 0.6798, -0.3583, 1.8032, -0.0833, -0.4193, -0.0604, 0.7310, 0.2062],
  12101. [-1.1939, -1.5435, 1.6261, -0.7515, -0.1293, -1.0238, 0.6422, 0.2504],
  12102. [ 1.0001, -0.1213, 1.0754, -0.6870, -0.5014, -0.7288, 0.2028, 0.2405],
  12103. [ 1.0468, -0.0862, 1.6703, 0.3435, -0.3764, 0.3130, 0.1611, 0.1331],
  12104. [ 0.9204, -0.1412, 1.6262, 0.0114, -0.5259, -0.2258, 0.2021, 0.2598],
  12105. [ 0.7783, -0.2752, 1.5301, -0.8552, -0.2124, -1.0131, 0.8405, 0.2358],
  12106. [-0.8339, -1.2837, 1.0198, -0.9123, -0.5547, -0.8795, 0.0556, 0.2826]],
  12107. device='cuda:0', grad_fn=<AddmmBackward>)
  12108. landmarks are: tensor([[[ 6.3883e-01, -3.6231e-01, 1.9173e+00, -7.3857e-01, -3.5173e-01,
  12109. -5.8460e-01, 1.1495e+00, 2.6764e-01],
  12110. [ 6.4542e-01, -3.6231e-01, 1.9346e+00, -4.4604e-01, -4.9607e-01,
  12111. -2.9207e-01, 1.1642e+00, 2.4092e-01],
  12112. [-2.2859e+00, -2.2859e+00, 1.8249e+00, -8.0015e-01, 4.0878e-02,
  12113. -1.2543e+00, 8.0590e-01, 3.0505e-01],
  12114. [ 5.4417e-01, -3.8545e-01, 1.0224e+00, -9.5412e-01, -6.1155e-01,
  12115. -9.2333e-01, 1.7452e-01, 2.5215e-01],
  12116. [ 5.2379e-01, -4.5797e-01, 1.7037e+00, 4.6189e-04, -2.7090e-01,
  12117. 6.2048e-02, 1.8356e-01, 1.4106e-02],
  12118. [ 6.0722e-01, -3.2394e-01, 1.8423e+00, -3.5366e-01, -4.9607e-01,
  12119. -3.9215e-01, 2.0831e-01, 1.8522e-01],
  12120. [ 6.0425e-01, -4.1045e-01, 1.5478e+00, -1.2082e+00, -1.2079e-01,
  12121. -1.0927e+00, 9.7040e-01, 3.1574e-01],
  12122. [-2.2859e+00, -2.2859e+00, 1.1020e+00, -1.0994e+00, -5.3649e-01,
  12123. -1.0542e+00, 5.4227e-02, 2.9047e-01]]], device='cuda:0')
  12124. loss_train_step before backward: tensor(0.1190, device='cuda:0', grad_fn=<MseLossBackward>)
  12125. loss_train_step after backward: tensor(0.1190, device='cuda:0', grad_fn=<MseLossBackward>)
  12126. loss_train: 0.3797225020825863
  12127. step: 9
  12128. running loss: 0.042191389120287366
  12129. Train Steps: 9/90 Loss: 0.0422 torch.Size([8, 600, 800])
  12130. torch.Size([8, 8])
  12131. tensor([[0.6275, 0.4157, 0.8337, 0.5800, 0.3763, 0.4200, 0.5547, 0.6125],
  12132. [ nan, nan, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609],
  12133. [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869],
  12134. [0.6250, 0.4103, 0.8950, 0.4400, 0.3912, 0.5650, 0.6050, 0.5133],
  12135. [0.6255, 0.4017, 0.8688, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901],
  12136. [0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719],
  12137. [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583],
  12138. [ nan, nan, 0.8938, 0.2850, 0.4662, 0.3117, 0.7406, 0.5528]],
  12139. device='cuda:0', dtype=torch.float64)
  12140. predictions are: tensor([[ 1.2359, 0.0152, 1.4352, 0.4591, -0.4579, -0.1886, 0.3458, 0.3827],
  12141. [-1.4108, -1.7165, 1.6278, -0.9198, -0.0999, -1.1243, 0.6107, 0.2600],
  12142. [ 1.0880, -0.0631, 1.4730, -0.7942, -0.2523, -1.2433, 0.1780, 0.2025],
  12143. [ 0.9040, -0.1922, 1.7536, -0.0645, -0.4241, 0.4290, 0.6224, 0.1873],
  12144. [ 0.9999, -0.1657, 1.5762, -0.5998, -0.5918, -0.5274, 0.6032, 0.1100],
  12145. [ 0.6061, -0.3682, 1.4800, -0.3083, -0.6131, -0.1629, 0.2296, 0.2176],
  12146. [ 0.9515, -0.1362, 1.5629, -0.2965, -0.5961, -0.2386, 0.4166, 0.2649],
  12147. [-0.7224, -1.2803, 1.7950, -0.7486, -0.2483, -0.8924, 0.8542, 0.2345]],
  12148. device='cuda:0', grad_fn=<AddmmBackward>)
  12149. landmarks are: tensor([[[ 0.6124, -0.3658, 1.5651, 0.3931, -0.5480, -0.3460, 0.2761,
  12150. 0.5431],
  12151. [-2.2859, -2.2859, 1.8249, -0.8002, 0.0409, -1.2543, 0.8059,
  12152. 0.3050],
  12153. [ 0.6119, -0.3927, 1.6979, -0.7925, -0.2536, -1.3698, 0.4282,
  12154. -0.0368],
  12155. [ 0.6010, -0.3909, 1.8480, -0.2536, -0.4788, 0.3238, 0.5085,
  12156. 0.0851],
  12157. [ 0.6031, -0.4307, 1.7268, -0.8002, -0.6058, -0.6462, 0.6442,
  12158. -0.0220],
  12159. [ 0.5367, -0.4294, 1.5709, -0.4999, -0.6693, -0.3075, 0.2455,
  12160. 0.3559],
  12161. [ 0.5761, -0.3966, 1.6171, -0.4845, -0.6346, -0.4691, 0.4739,
  12162. 0.2930],
  12163. [-2.2859, -2.2859, 1.8423, -0.9695, -0.1323, -0.8463, 1.1349,
  12164. 0.2676]]], device='cuda:0')
  12165. loss_train_step before backward: tensor(0.1079, device='cuda:0', grad_fn=<MseLossBackward>)
  12166. loss_train_step after backward: tensor(0.1079, device='cuda:0', grad_fn=<MseLossBackward>)
  12167. loss_train: 0.48758096620440483
  12168. step: 10
  12169. running loss: 0.048758096620440484
  12170.  
  12171. Train Steps: 10/90 Loss: 0.0488 torch.Size([8, 600, 800])
  12172. torch.Size([8, 8])
  12173. tensor([[0.6120, 0.4014, 0.6863, 0.2817, 0.3700, 0.2783, 0.5513, 0.5667],
  12174. [ nan, nan, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729],
  12175. [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333],
  12176. [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395],
  12177. [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343],
  12178. [0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019],
  12179. [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133],
  12180. [0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901]],
  12181. device='cuda:0', dtype=torch.float64)
  12182. predictions are: tensor([[ 0.7067, -0.3148, 1.0492, -1.0354, -0.4981, -0.9338, 0.3020, 0.3286],
  12183. [-0.7523, -1.2924, 0.9029, -1.2760, -0.3634, -1.3894, 0.3071, 0.3262],
  12184. [ 0.6640, -0.3744, 1.8604, -0.2007, -0.5467, 0.1417, 0.5132, 0.2404],
  12185. [ 0.5437, -0.4182, 1.7214, 0.1975, -0.6007, -0.5034, 0.4236, 0.2390],
  12186. [ 0.3511, -0.5864, 1.7093, -0.9693, 0.0304, -1.2827, 1.0219, 0.1822],
  12187. [ 1.1674, -0.0858, 1.8699, -0.0512, -0.5583, -0.3091, 0.7732, 0.1525],
  12188. [ 0.3252, -0.5765, 1.7152, 0.1841, -0.2517, 0.2405, 0.2954, 0.2331],
  12189. [ 0.3638, -0.5548, 1.8978, -0.1514, -0.4938, -0.0809, 0.1914, 0.1650]],
  12190. device='cuda:0', grad_fn=<AddmmBackward>)
  12191. landmarks are: tensor([[[ 0.5410, -0.4321, 0.8838, -0.9849, -0.5769, -1.0003, 0.2603,
  12192. 0.3315],
  12193. [-2.2859, -2.2859, 0.8824, -1.2727, -0.3691, -1.4237, 0.0943,
  12194. 0.3604],
  12195. [ 0.6003, -0.3747, 1.8885, -0.1997, -0.5018, -0.0149, 0.5490,
  12196. 0.1775],
  12197. [ 0.5697, -0.4514, 1.7420, 0.2672, -0.6055, -0.6312, 0.3449,
  12198. 0.2058],
  12199. [ 0.6125, -0.4321, 1.8018, -0.9772, 0.0120, -1.2543, 0.9741,
  12200. 0.1821],
  12201. [ 0.6372, -0.4129, 1.8942, -0.0765, -0.6173, -0.4768, 0.6999,
  12202. 0.0325],
  12203. [ 0.5786, -0.4141, 1.7037, 0.1544, -0.1862, 0.0736, 0.4393,
  12204. 0.0851],
  12205. [ 0.5663, -0.3955, 1.7788, -0.2382, -0.4037, -0.2690, 0.0828,
  12206. -0.0220]]], device='cuda:0')
  12207. loss_train_step before backward: tensor(0.0692, device='cuda:0', grad_fn=<MseLossBackward>)
  12208. loss_train_step after backward: tensor(0.0692, device='cuda:0', grad_fn=<MseLossBackward>)
  12209. loss_train: 0.5567609034478664
  12210. step: 11
  12211. running loss: 0.05061462758616968
  12212. Train Steps: 11/90 Loss: 0.0506 torch.Size([8, 600, 800])
  12213. torch.Size([8, 8])
  12214. tensor([[0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170],
  12215. [0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400],
  12216. [0.6111, 0.3995, 0.8788, 0.4567, 0.3813, 0.4833, 0.5450, 0.5700],
  12217. [0.6200, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391],
  12218. [0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266],
  12219. [0.6058, 0.3986, 0.8324, 0.4626, 0.3838, 0.4983, 0.5147, 0.5466],
  12220. [0.6182, 0.4058, 0.8738, 0.4350, 0.3563, 0.3400, 0.5290, 0.5822],
  12221. [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402]],
  12222. device='cuda:0', dtype=torch.float64)
  12223. predictions are: tensor([[ 0.3899, -0.5611, 1.6083, -1.0398, -0.0411, -1.4380, 0.9611, 0.1850],
  12224. [ 0.7063, -0.3622, 1.7582, -0.4214, -0.6344, -0.0607, 0.7025, 0.2027],
  12225. [ 0.0398, -0.7886, 1.7258, -0.0711, -0.5843, 0.0780, 0.3114, 0.1939],
  12226. [ 0.3534, -0.5398, 1.6284, -0.9779, -0.2579, -1.2172, 0.6238, 0.2286],
  12227. [ 0.5881, -0.3905, 1.2062, -1.0493, -0.2944, -1.2115, 0.6174, 0.4106],
  12228. [ 0.1641, -0.6547, 1.5900, -0.0401, -0.4739, 0.0842, 0.2363, 0.2308],
  12229. [ 0.3576, -0.5340, 1.6793, -0.2877, -0.6649, -0.5644, 0.2157, 0.2440],
  12230. [ 0.3927, -0.5486, 1.7002, 0.2642, -0.4248, -0.0980, 0.4854, 0.1575]],
  12231. device='cuda:0', grad_fn=<AddmmBackward>)
  12232. landmarks are: tensor([[[ 0.6471, -0.3840, 1.5767, -1.0311, -0.0457, -1.5007, 0.6889,
  12233. 0.1020],
  12234. [ 0.5778, -0.4393, 1.8018, -0.4614, -0.6693, -0.1381, 0.5490,
  12235. 0.2083],
  12236. [ 0.5368, -0.4406, 1.7730, -0.1766, -0.5249, -0.0534, 0.2314,
  12237. 0.3469],
  12238. [ 0.5776, -0.3987, 1.6005, -0.9121, -0.3844, -1.2358, 0.4247,
  12239. 0.2043],
  12240. [ 0.5746, -0.4021, 1.0801, -1.1312, -0.3229, -1.1081, 0.4803,
  12241. 0.6084],
  12242. [ 0.5124, -0.4446, 1.5587, -0.1493, -0.5134, 0.0159, 0.0912,
  12243. 0.2386],
  12244. [ 0.5697, -0.4113, 1.7499, -0.2767, -0.6404, -0.7155, 0.1576,
  12245. 0.4032],
  12246. [ 0.5780, -0.4565, 1.6221, 0.2532, -0.3728, -0.1718, 0.4357,
  12247. 0.2091]]], device='cuda:0')
  12248. loss_train_step before backward: tensor(0.0200, device='cuda:0', grad_fn=<MseLossBackward>)
  12249. loss_train_step after backward: tensor(0.0200, device='cuda:0', grad_fn=<MseLossBackward>)
  12250. loss_train: 0.5767896324396133
  12251. step: 12
  12252. running loss: 0.04806580270330111
  12253. Train Steps: 12/90 Loss: 0.0481 torch.Size([8, 600, 800])
  12254. torch.Size([8, 8])
  12255. tensor([[0.6200, 0.3999, 0.8653, 0.5207, 0.4100, 0.5125, 0.5975, 0.5103],
  12256. [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317],
  12257. [0.6225, 0.4116, 0.8662, 0.3517, 0.3663, 0.3233, 0.5837, 0.5317],
  12258. [0.6286, 0.4060, 0.9188, 0.4333, 0.3675, 0.4167, 0.7034, 0.5528],
  12259. [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343],
  12260. [0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835],
  12261. [ nan, nan, 0.7648, 0.2722, 0.3962, 0.2183, 0.5060, 0.5422],
  12262. [0.6267, 0.4080, 0.8438, 0.2633, 0.4763, 0.1800, 0.6259, 0.5240]],
  12263. device='cuda:0', dtype=torch.float64)
  12264. predictions are: tensor([[ 0.7066, -0.3359, 1.6617, 0.1113, -0.4845, 0.2730, 0.3180, 0.1751],
  12265. [ 0.6655, -0.3258, 1.6267, 0.1583, -0.5994, -0.3720, 0.3459, 0.2149],
  12266. [ 0.7973, -0.2172, 1.5151, -0.6836, -0.6012, -0.6649, 0.2786, 0.2913],
  12267. [ 0.7946, -0.2888, 1.9036, -0.3246, -0.5414, -0.2549, 0.8325, 0.2533],
  12268. [ 0.5353, -0.4337, 1.5933, -0.9831, -0.0078, -1.1072, 0.9309, 0.2004],
  12269. [ 0.3392, -0.5292, 1.6934, -0.5276, -0.3595, -1.0432, 0.5401, 0.1994],
  12270. [-2.1027, -2.1584, 1.0679, -0.8955, -0.4812, -0.9877, 0.2749, 0.2430],
  12271. [ 0.5163, -0.4084, 1.4821, -1.0341, -0.1654, -1.1220, 0.6254, 0.2533]],
  12272. device='cuda:0', grad_fn=<AddmmBackward>)
  12273. landmarks are: tensor([[[ 0.5778, -0.4389, 1.7107, 0.1192, -0.3921, 0.0815, 0.4741,
  12274. 0.0711],
  12275. [ 0.5901, -0.4157, 1.7557, 0.1929, -0.5423, -0.5923, 0.3584,
  12276. 0.1698],
  12277. [ 0.5893, -0.3847, 1.7152, -0.6616, -0.5942, -0.7925, 0.4104,
  12278. 0.1698],
  12279. [ 0.6174, -0.4105, 1.9577, -0.2844, -0.5885, -0.3614, 0.9631,
  12280. 0.2676],
  12281. [ 0.6125, -0.4321, 1.8018, -0.9772, 0.0120, -1.2543, 0.9741,
  12282. 0.1821],
  12283. [ 0.6117, -0.3877, 1.8249, -0.5230, -0.3517, -1.2851, 0.4200,
  12284. -0.0529],
  12285. [-2.2859, -2.2859, 1.2469, -1.0288, -0.4557, -1.2774, 0.0511,
  12286. 0.2183],
  12287. [ 0.6088, -0.4015, 1.6113, -1.0696, -0.0861, -1.4545, 0.6051,
  12288. 0.1343]]], device='cuda:0')
  12289. loss_train_step before backward: tensor(0.0180, device='cuda:0', grad_fn=<MseLossBackward>)
  12290. loss_train_step after backward: tensor(0.0180, device='cuda:0', grad_fn=<MseLossBackward>)
  12291. loss_train: 0.5947947613894939
  12292. step: 13
  12293. running loss: 0.045753443183807224
  12294. Train Steps: 13/90 Loss: 0.0458 torch.Size([8, 600, 800])
  12295. torch.Size([8, 8])
  12296. tensor([[0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431],
  12297. [0.6076, 0.3953, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954],
  12298. [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285],
  12299. [0.6255, 0.4017, 0.8688, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901],
  12300. [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395],
  12301. [ nan, nan, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633],
  12302. [ nan, nan, 0.8888, 0.3100, 0.5262, 0.2817, 0.7145, 0.6003],
  12303. [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5463, 0.5800]],
  12304. device='cuda:0', dtype=torch.float64)
  12305. predictions are: tensor([[ 0.8568, -0.1986, 1.7606, -0.0271, -0.2534, -0.0415, 0.1967, 0.1760],
  12306. [ 1.1717, -0.0319, 1.6136, -0.6131, -0.6550, -0.5300, 0.3238, 0.1442],
  12307. [ 1.0539, -0.1441, 1.7893, -0.6783, -0.5686, -0.2087, 0.8835, 0.2074],
  12308. [ 1.0515, -0.1461, 1.7092, -0.7597, -0.5884, -0.7283, 0.6856, 0.1205],
  12309. [ 0.7875, -0.2442, 1.7666, 0.2316, -0.5501, -0.6627, 0.4504, 0.2163],
  12310. [-2.0296, -2.1249, 1.2080, -0.9242, -0.4599, -1.0183, 0.2166, 0.2925],
  12311. [-1.3036, -1.6394, 1.8677, -0.7513, 0.0723, -1.0658, 1.0924, 0.3734],
  12312. [ 0.5397, -0.3751, 1.1715, -0.8596, -0.4805, -1.1048, 0.2473, 0.3443]],
  12313. device='cuda:0', grad_fn=<AddmmBackward>)
  12314. landmarks are: tensor([[[ 5.3508e-01, -4.1527e-01, 1.7326e+00, -4.5727e-02, -2.2139e-01,
  12315. -4.6642e-02, 4.3431e-02, 2.2284e-01],
  12316. [ 5.2061e-01, -4.6028e-01, 1.4670e+00, -4.7683e-01, -6.5774e-01,
  12317. -3.9985e-01, 1.8356e-01, 2.0831e-03],
  12318. [ 6.1083e-01, -4.2731e-01, 1.8711e+00, -6.6159e-01, -5.7691e-01,
  12319. -1.9969e-01, 9.1557e-01, 1.5543e-01],
  12320. [ 6.0306e-01, -4.3072e-01, 1.7268e+00, -8.0015e-01, -6.0577e-01,
  12321. -6.4619e-01, 6.4417e-01, -2.1963e-02],
  12322. [ 5.6966e-01, -4.5138e-01, 1.7420e+00, 2.6720e-01, -6.0553e-01,
  12323. -6.3118e-01, 3.4489e-01, 2.0578e-01],
  12324. [-2.2859e+00, -2.2859e+00, 1.2030e+00, -1.0288e+00, -4.9607e-01,
  12325. -1.1081e+00, 8.1293e-02, 3.1609e-01],
  12326. [-2.2859e+00, -2.2859e+00, 1.8192e+00, -8.5404e-01, 1.4480e-01,
  12327. -9.8491e-01, 1.0143e+00, 4.8673e-01],
  12328. [ 5.5484e-01, -3.9360e-01, 1.1634e+00, -8.1049e-01, -5.1917e-01,
  12329. -1.0696e+00, 2.3718e-01, 3.9307e-01]]], device='cuda:0')
  12330. loss_train_step before backward: tensor(0.0493, device='cuda:0', grad_fn=<MseLossBackward>)
  12331. loss_train_step after backward: tensor(0.0493, device='cuda:0', grad_fn=<MseLossBackward>)
  12332. loss_train: 0.644109707325697
  12333. step: 14
  12334. running loss: 0.04600783623754978
  12335.  
  12336. Train Steps: 14/90 Loss: 0.0460 torch.Size([8, 600, 800])
  12337. torch.Size([8, 8])
  12338. tensor([[0.6142, 0.4127, 0.7575, 0.3067, 0.3438, 0.4383, 0.5778, 0.5207],
  12339. [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167],
  12340. [0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689],
  12341. [0.6193, 0.3930, 0.8949, 0.4437, 0.3852, 0.5435, 0.6263, 0.5263],
  12342. [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395],
  12343. [0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862],
  12344. [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5687, 0.5567],
  12345. [0.6109, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567]],
  12346. device='cuda:0', dtype=torch.float64)
  12347. predictions are: tensor([[ 0.3045, -0.5674, 1.3769, -0.9429, -0.6118, -0.4503, 0.2990, 0.2750],
  12348. [ 0.0977, -0.6868, 1.3143, -0.7793, 0.0222, -1.2691, 0.4035, 0.3906],
  12349. [ 0.4517, -0.4892, 1.5790, -0.6596, -0.6443, -0.5839, 0.3201, 0.2341],
  12350. [ 0.0941, -0.8263, 1.9781, -0.0808, -0.4988, 0.1383, 0.6655, 0.1129],
  12351. [ 0.4164, -0.5374, 1.9136, 0.3085, -0.5462, -0.6897, 0.5370, 0.1564],
  12352. [ 0.2482, -0.6505, 2.1673, -0.2775, -0.6107, -0.3451, 0.8151, 0.0446],
  12353. [-0.0211, -0.7762, 1.3772, -0.9917, -0.2255, -1.2161, 0.5604, 0.3626],
  12354. [ 0.3111, -0.6232, 1.2350, -1.1596, -0.4139, -1.2741, 0.4848, 0.2642]],
  12355. device='cuda:0', grad_fn=<AddmmBackward>)
  12356. landmarks are: tensor([[[ 0.5509, -0.3798, 1.2129, -0.8694, -0.6982, -0.2613, 0.3830,
  12357. 0.1193],
  12358. [ 0.6161, -0.3075, 1.1678, -0.6500, 0.0813, -1.4006, 0.2545,
  12359. 0.5624],
  12360. [ 0.5327, -0.4381, 1.2880, -0.6308, -0.6866, -0.5307, 0.2658,
  12361. 0.3417],
  12362. [ 0.5746, -0.4706, 1.8476, -0.2365, -0.5068, 0.2245, 0.6069,
  12363. 0.1449],
  12364. [ 0.5697, -0.4514, 1.7420, 0.2672, -0.6055, -0.6312, 0.3449,
  12365. 0.2058],
  12366. [ 0.6072, -0.4075, 1.8942, -0.3537, -0.5423, -0.1612, 0.6277,
  12367. -0.0400],
  12368. [ 0.5824, -0.3625, 1.0405, -0.9325, -0.4268, -1.1389, 0.3411,
  12369. 0.2853],
  12370. [ 0.5356, -0.4217, 1.0339, -1.4776, -0.5076, -1.1081, 0.4219,
  12371. 0.2853]]], device='cuda:0')
  12372. loss_train_step before backward: tensor(0.0427, device='cuda:0', grad_fn=<MseLossBackward>)
  12373. loss_train_step after backward: tensor(0.0427, device='cuda:0', grad_fn=<MseLossBackward>)
  12374. loss_train: 0.6868270449340343
  12375. step: 15
  12376. running loss: 0.04578846966226895
  12377. Train Steps: 15/90 Loss: 0.0458 torch.Size([8, 600, 800])
  12378. torch.Size([8, 8])
  12379. tensor([[0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835],
  12380. [0.6204, 0.4049, 0.7975, 0.2700, 0.3937, 0.2567, 0.5700, 0.5183],
  12381. [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329],
  12382. [0.6263, 0.4039, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804],
  12383. [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482],
  12384. [ nan, nan, 0.7648, 0.2722, 0.3962, 0.2183, 0.5060, 0.5422],
  12385. [0.6219, 0.3934, 0.8688, 0.5267, 0.4313, 0.4967, 0.5988, 0.4983],
  12386. [0.6257, 0.4167, 0.8775, 0.3433, 0.3563, 0.4133, 0.6200, 0.5667]],
  12387. device='cuda:0', dtype=torch.float64)
  12388. predictions are: tensor([[ 0.4056, -0.5116, 1.7553, -0.5207, -0.1930, -1.2530, 0.6542, 0.2057],
  12389. [ 0.0376, -0.7179, 1.3283, -0.9410, -0.3670, -1.1820, 0.3219, 0.2579],
  12390. [ 0.7586, -0.2953, 1.6136, -0.7672, -0.5980, -0.7300, 0.5055, 0.2464],
  12391. [ 1.1241, -0.0934, 1.7848, -0.3410, -0.5859, -0.2160, 0.5642, 0.0793],
  12392. [ 0.6242, -0.3637, 1.8110, -0.4475, -0.5276, -0.6335, 0.5430, 0.2081],
  12393. [-2.5332, -2.4693, 1.1160, -0.8603, -0.3118, -1.1134, 0.3019, 0.2916],
  12394. [ 0.6375, -0.3858, 1.7093, 0.0144, -0.3216, -0.1430, 0.4885, 0.1177],
  12395. [ 0.4408, -0.4880, 1.7590, -0.6851, -0.6015, -0.5339, 0.6534, 0.3028]],
  12396. device='cuda:0', grad_fn=<AddmmBackward>)
  12397. landmarks are: tensor([[[ 0.6117, -0.3877, 1.8249, -0.5230, -0.3517, -1.2851, 0.4200,
  12398. -0.0529],
  12399. [ 0.5798, -0.4156, 1.3977, -1.0388, -0.4672, -1.1004, 0.3469,
  12400. 0.1082],
  12401. [ 0.5680, -0.4393, 1.5920, -0.6672, -0.6453, -0.5457, 0.5149,
  12402. 0.1753],
  12403. [ 0.6072, -0.4206, 1.8711, -0.2536, -0.6115, -0.1304, 0.6812,
  12404. -0.0670],
  12405. [ 0.5791, -0.4127, 1.8442, -0.3985, -0.6031, -0.6154, 0.4473,
  12406. 0.2464],
  12407. [-2.2859, -2.2859, 1.2469, -1.0288, -0.4557, -1.2774, 0.0511,
  12408. 0.2183],
  12409. [ 0.5864, -0.4690, 1.7268, 0.1467, -0.2940, 0.0082, 0.4797,
  12410. 0.0159],
  12411. [ 0.6040, -0.3614, 1.7672, -0.7001, -0.6404, -0.3768, 0.5778,
  12412. 0.3315]]], device='cuda:0')
  12413. loss_train_step before backward: tensor(0.0251, device='cuda:0', grad_fn=<MseLossBackward>)
  12414. loss_train_step after backward: tensor(0.0251, device='cuda:0', grad_fn=<MseLossBackward>)
  12415. loss_train: 0.7119743004441261
  12416. step: 16
  12417. running loss: 0.04449839377775788
  12418. Train Steps: 16/90 Loss: 0.0445 torch.Size([8, 600, 800])
  12419. torch.Size([8, 8])
  12420. tensor([[0.6300, 0.4133, 0.8538, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413],
  12421. [0.6252, 0.4158, 0.8988, 0.4083, 0.3788, 0.4783, 0.6225, 0.5633],
  12422. [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413],
  12423. [0.6222, 0.3937, 0.8350, 0.5617, 0.4138, 0.4600, 0.5800, 0.5233],
  12424. [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183],
  12425. [0.6224, 0.4097, 0.7438, 0.2267, 0.3850, 0.2850, 0.5988, 0.5250],
  12426. [0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533],
  12427. [0.6289, 0.4019, 0.8113, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332]],
  12428. device='cuda:0', dtype=torch.float64)
  12429. predictions are: tensor([[-0.5698, -1.1630, 1.7988, -1.0738, 0.0985, -1.2631, 0.9678, 0.2631],
  12430. [ 0.6247, -0.3964, 1.8911, -0.4913, -0.7389, -0.3246, 0.3989, 0.1941],
  12431. [-1.1609, -1.5773, 1.9380, -0.7570, -0.0841, -1.2592, 0.9371, 0.2391],
  12432. [ 0.4757, -0.5144, 1.6757, 0.2291, -0.5092, -0.3942, 0.3250, 0.1567],
  12433. [ 0.1884, -0.6917, 1.8490, -0.1794, -0.5994, 0.2266, 0.3432, 0.1332],
  12434. [ 0.6206, -0.3816, 1.1967, -1.1353, -0.5213, -1.1738, 0.2903, 0.2247],
  12435. [ 0.6793, -0.3524, 1.1073, -1.2023, -0.6062, -1.0704, 0.3309, 0.2422],
  12436. [ 0.5047, -0.4610, 1.5853, 0.0454, -0.5763, -0.2399, 0.3652, 0.1877]],
  12437. device='cuda:0', grad_fn=<AddmmBackward>)
  12438. landmarks are: tensor([[[ 0.6240, -0.3768, 1.6575, -1.2851, 0.2949, -1.2467, 1.1276,
  12439. 0.2142],
  12440. [ 0.6017, -0.3654, 1.8654, -0.3998, -0.5365, -0.0765, 0.5894,
  12441. 0.3161],
  12442. [ 0.6108, -0.4201, 1.8711, -0.7848, -0.0053, -1.2236, 1.0362,
  12443. 0.2142],
  12444. [ 0.5880, -0.4676, 1.5709, 0.3084, -0.3748, -0.1612, 0.3931,
  12445. 0.1313],
  12446. [ 0.5934, -0.4276, 1.8192, -0.1458, -0.4499, 0.3777, 0.6702,
  12447. 0.1082],
  12448. [ 0.5891, -0.3937, 1.1494, -1.2390, -0.5076, -0.9695, 0.4797,
  12449. 0.1390],
  12450. [ 0.5878, -0.4052, 1.0229, -1.2855, -0.5596, -0.8232, 0.5316,
  12451. 0.2699],
  12452. [ 0.6191, -0.4297, 1.4612, 0.2391, -0.4961, 0.0313, 1.1166,
  12453. 0.1768]]], device='cuda:0')
  12454. loss_train_step before backward: tensor(0.1294, device='cuda:0', grad_fn=<MseLossBackward>)
  12455. loss_train_step after backward: tensor(0.1294, device='cuda:0', grad_fn=<MseLossBackward>)
  12456. loss_train: 0.8413371220231056
  12457. step: 17
  12458. running loss: 0.04949041894253563
  12459. Train Steps: 17/90 Loss: 0.0495 torch.Size([8, 600, 800])
  12460. torch.Size([8, 8])
  12461. tensor([[0.6223, 0.4028, 0.8988, 0.4200, 0.3763, 0.5733, 0.6375, 0.5167],
  12462. [0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567],
  12463. [0.6189, 0.4029, 0.8375, 0.5767, 0.4745, 0.4829, 0.5551, 0.5598],
  12464. [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817],
  12465. [0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528],
  12466. [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960],
  12467. [0.6176, 0.4030, 0.8850, 0.4850, 0.3688, 0.4050, 0.5312, 0.5783],
  12468. [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500]],
  12469. device='cuda:0', dtype=torch.float64)
  12470. predictions are: tensor([[ 0.3496, -0.6161, 1.8231, -0.7200, -0.6343, -0.1043, 0.8363, 0.1154],
  12471. [ 0.2936, -0.5610, 1.7201, -0.4867, -0.2092, -0.3321, 0.2514, 0.2062],
  12472. [ 0.5884, -0.4009, 1.6049, 0.0735, -0.1333, -0.5680, 0.4170, 0.2590],
  12473. [-0.0426, -0.8502, 1.6135, 0.1152, -0.2904, -0.5447, 0.2721, 0.1971],
  12474. [-0.1033, -0.8852, 2.0310, -0.9600, -0.3148, -1.1868, 1.0985, 0.2178],
  12475. [ 0.2483, -0.6758, 1.7849, -0.3754, -0.3378, -0.4320, 0.2930, 0.1095],
  12476. [-0.0405, -0.8126, 1.6816, -0.3414, -0.7438, -0.7737, 0.2725, 0.1434],
  12477. [ 0.3748, -0.5497, 1.3174, -1.2418, -0.7503, -0.8426, 0.5530, 0.2293]],
  12478. device='cuda:0', grad_fn=<AddmmBackward>)
  12479. landmarks are: tensor([[[ 0.5886, -0.4253, 1.8654, -0.3460, -0.5480, 0.3623, 0.6587,
  12480. 0.1005],
  12481. [ 0.5538, -0.4139, 1.7557, -0.1843, -0.0459, 0.1242, 0.4219,
  12482. 0.2853],
  12483. [ 0.5726, -0.4249, 1.5824, 0.3777, -0.0942, -0.0556, 0.2782,
  12484. 0.2997],
  12485. [ 0.5193, -0.4699, 1.5767, 0.4008, -0.2420, 0.0774, 0.1178,
  12486. -0.0610],
  12487. [ 0.6388, -0.3623, 1.9173, -0.7386, -0.3517, -0.5846, 1.1495,
  12488. 0.2676],
  12489. [ 0.5205, -0.4577, 1.7326, -0.1304, -0.1862, 0.0447, 0.1301,
  12490. 0.0051],
  12491. [ 0.5667, -0.4246, 1.8018, -0.0457, -0.5827, -0.4152, 0.1679,
  12492. 0.3854],
  12493. [ 0.5763, -0.4147, 1.3226, -1.0619, -0.6635, -0.4152, 0.5374,
  12494. 0.2545]]], device='cuda:0')
  12495. loss_train_step before backward: tensor(0.0813, device='cuda:0', grad_fn=<MseLossBackward>)
  12496. loss_train_step after backward: tensor(0.0813, device='cuda:0', grad_fn=<MseLossBackward>)
  12497. loss_train: 0.9225974231958389
  12498. step: 18
  12499. running loss: 0.05125541239976883
  12500.  
  12501. Train Steps: 18/90 Loss: 0.0513 torch.Size([8, 600, 800])
  12502. torch.Size([8, 8])
  12503. tensor([[0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200],
  12504. [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333],
  12505. [0.6198, 0.4101, 0.8838, 0.5283, 0.3763, 0.5267, 0.5913, 0.5567],
  12506. [ nan, nan, 0.7425, 0.2117, 0.3937, 0.2433, 0.5438, 0.5567],
  12507. [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524],
  12508. [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938],
  12509. [0.6268, 0.4102, 0.8938, 0.3667, 0.4025, 0.2833, 0.6275, 0.5183],
  12510. [0.6048, 0.3928, 0.8538, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719]],
  12511. device='cuda:0', dtype=torch.float64)
  12512. predictions are: tensor([[ 0.4430, -0.4844, 1.7350, -0.4449, -0.5268, -0.4250, 0.4806, 0.1754],
  12513. [ 0.7304, -0.3097, 1.7664, -0.6020, -0.5014, -0.2737, 0.5494, 0.1550],
  12514. [ 0.8088, -0.2873, 1.7017, -0.1383, -0.4139, -0.1332, 0.5881, 0.2076],
  12515. [-2.3069, -2.3295, 1.1612, -1.1341, -0.2896, -1.2715, 0.1326, 0.3356],
  12516. [ 0.6132, -0.3375, 1.7022, -0.2673, -0.5030, -0.4217, 0.3474, 0.1730],
  12517. [ 0.4882, -0.4704, 1.8124, -0.1721, -0.5237, -0.8142, 0.6059, 0.0388],
  12518. [ 0.4792, -0.4727, 1.7930, -0.9632, -0.2783, -1.1860, 0.7636, 0.1998],
  12519. [ 0.5194, -0.4366, 1.5938, -0.0664, -0.4184, -0.2661, 0.3575, 0.1813]],
  12520. device='cuda:0', grad_fn=<AddmmBackward>)
  12521. landmarks are: tensor([[[ 0.5865, -0.3973, 1.8423, -0.0688, -0.5192, -0.2305, 0.4162,
  12522. 0.1159],
  12523. [ 0.6003, -0.3747, 1.8885, -0.1997, -0.5018, -0.0149, 0.5490,
  12524. 0.1775],
  12525. [ 0.5770, -0.3918, 1.7961, 0.1544, -0.5480, 0.1467, 0.4450,
  12526. 0.2853],
  12527. [-2.2859, -2.2859, 1.1436, -1.3082, -0.4672, -1.1620, 0.2256,
  12528. 0.2853],
  12529. [ 0.5529, -0.3811, 1.7788, -0.0380, -0.5307, -0.2074, 0.0727,
  12530. 0.2657],
  12531. [ 0.6273, -0.4105, 1.8480, 0.1082, -0.5538, -0.5076, 0.6414,
  12532. -0.0049],
  12533. [ 0.6092, -0.3913, 1.8423, -0.5923, -0.4268, -0.9772, 0.6125,
  12534. 0.1082],
  12535. [ 0.5079, -0.4714, 1.6575, 0.2237, -0.4961, 0.0774, 0.1465,
  12536. -0.1061]]], device='cuda:0')
  12537. loss_train_step before backward: tensor(0.0290, device='cuda:0', grad_fn=<MseLossBackward>)
  12538. loss_train_step after backward: tensor(0.0290, device='cuda:0', grad_fn=<MseLossBackward>)
  12539. loss_train: 0.9516376163810492
  12540. step: 19
  12541. running loss: 0.050086190335844695
  12542. Train Steps: 19/90 Loss: 0.0501 torch.Size([8, 600, 800])
  12543. torch.Size([8, 8])
  12544. tensor([[0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250],
  12545. [0.6135, 0.3994, 0.7913, 0.3050, 0.3625, 0.3050, 0.5837, 0.5050],
  12546. [0.6206, 0.4001, 0.8900, 0.3933, 0.3588, 0.3567, 0.5837, 0.5083],
  12547. [0.6167, 0.4048, 0.6831, 0.3639, 0.3763, 0.3017, 0.5700, 0.5883],
  12548. [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869],
  12549. [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466],
  12550. [0.6199, 0.4093, 0.7913, 0.2533, 0.4288, 0.2467, 0.5975, 0.5700],
  12551. [0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933]],
  12552. device='cuda:0', dtype=torch.float64)
  12553. predictions are: tensor([[ 0.1724, -0.7311, 1.9600, -0.4847, -0.5312, -0.7648, 0.7926, 0.1386],
  12554. [ 0.4509, -0.5342, 1.4227, -1.0582, -0.5859, -0.7999, 0.5457, 0.1146],
  12555. [ 0.0037, -0.8193, 1.9162, -0.4894, -0.6796, -0.5751, 0.5336, 0.0400],
  12556. [-0.0795, -0.8154, 1.1245, -0.8385, -0.5245, -0.8974, 0.2899, 0.3562],
  12557. [ 0.7510, -0.2744, 1.6003, -0.7833, -0.2182, -1.1405, 0.3211, 0.1225],
  12558. [ 0.1794, -0.6694, 1.7174, 0.0391, -0.2750, 0.1796, 0.3027, 0.1166],
  12559. [-0.3175, -1.0336, 1.3863, -1.1355, -0.3639, -1.0212, 0.5602, 0.2073],
  12560. [ 0.6083, -0.3555, 1.6484, 0.3436, -0.0741, -0.2059, 0.3531, 0.2449]],
  12561. device='cuda:0', grad_fn=<AddmmBackward>)
  12562. landmarks are: tensor([[[ 0.6111, -0.3828, 1.8885, -0.3844, -0.5654, -0.8079, 0.5663,
  12563. 0.1390],
  12564. [ 0.5477, -0.4413, 1.3688, -0.8771, -0.6115, -0.8771, 0.4104,
  12565. 0.0467],
  12566. [ 0.5807, -0.4378, 1.8249, -0.4691, -0.6289, -0.6385, 0.4104,
  12567. 0.0620],
  12568. [ 0.5626, -0.4162, 0.8692, -0.6051, -0.5480, -0.8925, 0.3469,
  12569. 0.4316],
  12570. [ 0.6119, -0.3927, 1.6979, -0.7925, -0.2536, -1.3698, 0.4282,
  12571. -0.0368],
  12572. [ 0.5178, -0.4480, 1.6748, 0.0620, -0.2767, 0.2083, 0.1067,
  12573. 0.2386],
  12574. [ 0.5771, -0.3955, 1.3688, -1.1158, -0.3055, -1.1466, 0.4739,
  12575. 0.3469],
  12576. [ 0.5959, -0.3579, 1.6055, 0.3623, -0.0573, -0.2074, 0.3122,
  12577. 0.4547]]], device='cuda:0')
  12578. loss_train_step before backward: tensor(0.0521, device='cuda:0', grad_fn=<MseLossBackward>)
  12579. loss_train_step after backward: tensor(0.0521, device='cuda:0', grad_fn=<MseLossBackward>)
  12580. loss_train: 1.0037182997912169
  12581. step: 20
  12582. running loss: 0.05018591498956084
  12583. Train Steps: 20/90 Loss: 0.0502 torch.Size([8, 600, 800])
  12584. torch.Size([8, 8])
  12585. tensor([[0.6163, 0.4114, 0.7650, 0.2017, 0.3763, 0.2867, 0.5631, 0.5071],
  12586. [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600],
  12587. [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600],
  12588. [0.6128, 0.4022, 0.8738, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064],
  12589. [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683],
  12590. [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309],
  12591. [0.6275, 0.4050, 0.9038, 0.3767, 0.3838, 0.3533, 0.7074, 0.5575],
  12592. [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5413, 0.5717]],
  12593. device='cuda:0', dtype=torch.float64)
  12594. predictions are: tensor([[ 0.1581, -0.7393, 1.1670, -1.3299, -0.6123, -1.1580, 0.3761, 0.1508],
  12595. [ 0.7792, -0.3083, 1.7652, 0.1693, -0.4503, -0.1500, 0.5473, 0.1806],
  12596. [-0.0364, -0.8090, 1.1568, -1.1909, -0.6619, -0.9205, 0.3632, 0.2406],
  12597. [ 0.1614, -0.6613, 1.7122, -0.0781, -0.0744, -0.1382, 0.0251, 0.1446],
  12598. [ 0.1000, -0.7275, 1.7813, -0.1652, -0.1890, -0.0821, 0.1867, 0.1571],
  12599. [ 0.4845, -0.4482, 1.8444, -0.0939, -0.4233, -0.6957, 0.6399, 0.1452],
  12600. [ 0.5154, -0.4710, 1.9175, -0.6788, -0.4608, -0.8603, 0.9891, 0.1327],
  12601. [ 0.1317, -0.7220, 1.7761, -0.2175, -0.3860, -0.1320, 0.3566, 0.1352]],
  12602. device='cuda:0', grad_fn=<AddmmBackward>)
  12603. landmarks are: tensor([[[ 0.5606, -0.3856, 1.2476, -1.3544, -0.5480, -0.9618, 0.3152,
  12604. 0.0562],
  12605. [ 0.5883, -0.3594, 1.7557, 0.2545, -0.4152, -0.0611, 0.3353,
  12606. 0.3007],
  12607. [ 0.5466, -0.4080, 1.0668, -1.1764, -0.6289, -0.7617, 0.4855,
  12608. 0.3007],
  12609. [ 0.5446, -0.4280, 1.7499, 0.0543, 0.0156, 0.1301, 0.1918,
  12610. 0.0532],
  12611. [ 0.5359, -0.4370, 1.7095, -0.0303, -0.0804, -0.0380, 0.1044,
  12612. 0.3392],
  12613. [ 0.6520, -0.3623, 1.8885, 0.0313, -0.5538, -0.5384, 0.6926,
  12614. 0.1661],
  12615. [ 0.6125, -0.4153, 1.8885, -0.5461, -0.5134, -0.6539, 0.9814,
  12616. 0.2890],
  12617. [ 0.5432, -0.4462, 1.7557, -0.0380, -0.3113, -0.0765, 0.2141,
  12618. 0.3546]]], device='cuda:0')
  12619. loss_train_step before backward: tensor(0.0353, device='cuda:0', grad_fn=<MseLossBackward>)
  12620. loss_train_step after backward: tensor(0.0353, device='cuda:0', grad_fn=<MseLossBackward>)
  12621. loss_train: 1.0390332210808992
  12622. step: 21
  12623. running loss: 0.049477772432423774
  12624. Train Steps: 21/90 Loss: 0.0495 torch.Size([8, 600, 800])
  12625. torch.Size([8, 8])
  12626. tensor([[0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309],
  12627. [0.6339, 0.4118, 0.7988, 0.5800, 0.3912, 0.4583, 0.7343, 0.5760],
  12628. [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760],
  12629. [0.6321, 0.4048, 0.8738, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927],
  12630. [0.6139, 0.4019, 0.7137, 0.2150, 0.4375, 0.1533, 0.5293, 0.5006],
  12631. [0.6058, 0.3986, 0.8324, 0.4626, 0.3838, 0.4983, 0.5147, 0.5466],
  12632. [0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528],
  12633. [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517]],
  12634. device='cuda:0', dtype=torch.float64)
  12635. predictions are: tensor([[ 0.5319, -0.4074, 1.9047, -0.0513, -0.4148, -0.4593, 0.4412, 0.1348],
  12636. [ 0.2248, -0.6662, 1.6434, -0.0576, -0.4394, -0.2607, 0.5983, 0.2211],
  12637. [ 0.5160, -0.4834, 1.7566, 0.1108, -0.4688, 0.0316, 0.6468, 0.1698],
  12638. [ 0.5514, -0.4455, 1.8779, 0.2109, -0.5295, -0.2038, 0.3353, -0.0306],
  12639. [ 0.0569, -0.7394, 0.9787, -1.4228, -0.3317, -1.4684, 0.0992, 0.2083],
  12640. [ 0.1960, -0.6464, 1.6755, -0.2134, -0.4800, 0.0778, 0.1021, 0.1230],
  12641. [ 0.3866, -0.5533, 2.0356, -0.8076, -0.3010, -0.5609, 0.8130, 0.1686],
  12642. [ 0.2819, -0.5713, 1.0122, -1.2399, -0.3901, -1.1646, 0.2829, 0.3324]],
  12643. device='cuda:0', grad_fn=<AddmmBackward>)
  12644. landmarks are: tensor([[[ 0.6520, -0.3623, 1.8885, 0.0313, -0.5538, -0.5384, 0.6926,
  12645. 0.1661],
  12646. [ 0.6421, -0.3840, 1.4035, 0.3931, -0.4788, -0.1689, 1.1057,
  12647. 0.3745],
  12648. [ 0.6421, -0.4008, 1.6055, 0.2160, -0.5076, -0.0534, 1.1020,
  12649. 0.3745],
  12650. [ 0.6335, -0.4162, 1.7499, 0.3084, -0.4961, -0.2459, 0.6524,
  12651. -0.0102],
  12652. [ 0.5496, -0.4295, 1.0109, -1.2928, -0.2651, -1.5777, 0.1589,
  12653. 0.0261],
  12654. [ 0.5124, -0.4446, 1.5587, -0.1493, -0.5134, 0.0159, 0.0912,
  12655. 0.2386],
  12656. [ 0.6388, -0.3623, 1.9173, -0.7386, -0.3517, -0.5846, 1.1495,
  12657. 0.2676],
  12658. [ 0.5713, -0.3671, 0.8665, -1.0696, -0.3691, -1.2236, 0.3527,
  12659. 0.2622]]], device='cuda:0')
  12660. loss_train_step before backward: tensor(0.0374, device='cuda:0', grad_fn=<MseLossBackward>)
  12661. loss_train_step after backward: tensor(0.0374, device='cuda:0', grad_fn=<MseLossBackward>)
  12662. loss_train: 1.0764061901718378
  12663. step: 22
  12664. running loss: 0.0489275540987199
  12665.  
  12666. Train Steps: 22/90 Loss: 0.0489 torch.Size([8, 600, 800])
  12667. torch.Size([8, 8])
  12668. tensor([[0.6108, 0.4011, 0.8037, 0.3400, 0.3700, 0.2933, 0.5658, 0.5617],
  12669. [0.6127, 0.4066, 0.8550, 0.5567, 0.4662, 0.5141, 0.5070, 0.5412],
  12670. [0.6271, 0.4040, 0.9138, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413],
  12671. [0.6176, 0.3911, 0.8738, 0.4217, 0.3488, 0.4033, 0.6025, 0.4817],
  12672. [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133],
  12673. [0.6293, 0.3982, 0.8700, 0.5300, 0.3763, 0.4717, 0.7050, 0.5297],
  12674. [0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719],
  12675. [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879]],
  12676. device='cuda:0', dtype=torch.float64)
  12677. predictions are: tensor([[ 0.4221, -0.4916, 1.3856, -0.8910, -0.5794, -0.8885, 0.3294, 0.2227],
  12678. [ 0.3831, -0.5018, 1.5959, 0.2377, -0.1456, -0.1022, 0.0904, 0.2351],
  12679. [ 0.6868, -0.3415, 1.8932, -0.6819, -0.1232, -1.1595, 0.8876, 0.1838],
  12680. [ 0.5910, -0.4053, 1.6661, -0.5671, -0.6732, -0.4688, 0.3958, 0.1051],
  12681. [ 0.2762, -0.5875, 1.6663, 0.0127, -0.2104, -0.0351, 0.2161, 0.1858],
  12682. [ 0.2888, -0.6276, 1.7013, 0.0176, -0.4367, -0.1943, 0.6267, 0.1648],
  12683. [ 0.4352, -0.4806, 1.5642, -0.6004, -0.6763, -0.3811, 0.2263, 0.1577],
  12684. [ 0.2547, -0.6514, 1.8138, -0.4008, -0.4382, 0.1134, 0.4997, 0.0756]],
  12685. device='cuda:0', grad_fn=<AddmmBackward>)
  12686. landmarks are: tensor([[[ 0.5355, -0.4335, 1.4266, -0.7155, -0.5769, -0.9310, 0.3275,
  12687. 0.3087],
  12688. [ 0.5443, -0.4079, 1.6633, 0.2853, -0.1325, 0.0888, 0.0558,
  12689. 0.2138],
  12690. [ 0.6108, -0.4201, 1.9346, -0.5538, -0.1497, -1.0773, 1.0545,
  12691. 0.2142],
  12692. [ 0.5669, -0.4794, 1.7499, -0.3383, -0.6751, -0.4229, 0.4970,
  12693. -0.0611],
  12694. [ 0.5786, -0.4141, 1.7037, 0.1544, -0.1862, 0.0736, 0.4393,
  12695. 0.0851],
  12696. [ 0.6207, -0.4466, 1.7326, 0.1621, -0.5480, -0.1073, 0.9704,
  12697. 0.1608],
  12698. [ 0.5367, -0.4294, 1.5709, -0.4999, -0.6693, -0.3075, 0.2455,
  12699. 0.3559],
  12700. [ 0.5676, -0.4327, 1.8252, -0.2095, -0.4886, 0.3084, 0.4727,
  12701. -0.0322]]], device='cuda:0')
  12702. loss_train_step before backward: tensor(0.0177, device='cuda:0', grad_fn=<MseLossBackward>)
  12703. loss_train_step after backward: tensor(0.0177, device='cuda:0', grad_fn=<MseLossBackward>)
  12704. loss_train: 1.0940845366567373
  12705. step: 23
  12706. running loss: 0.04756889289811901
  12707. Train Steps: 23/90 Loss: 0.0476 torch.Size([8, 600, 800])
  12708. torch.Size([8, 8])
  12709. tensor([[0.6117, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103],
  12710. [0.6136, 0.4085, 0.6688, 0.2317, 0.3862, 0.2367, 0.5517, 0.5783],
  12711. [ nan, nan, 0.8300, 0.3150, 0.3588, 0.3383, 0.5208, 0.5194],
  12712. [0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496],
  12713. [0.6236, 0.4084, 0.7738, 0.2133, 0.3663, 0.3233, 0.5813, 0.5567],
  12714. [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167],
  12715. [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200],
  12716. [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5413, 0.5717]],
  12717. device='cuda:0', dtype=torch.float64)
  12718. predictions are: tensor([[ 0.4754, -0.4291, 0.9876, -0.9167, -0.4275, -0.9420, 0.2557, 0.2084],
  12719. [ 0.5949, -0.3673, 0.9835, -0.8483, -0.4139, -1.0030, 0.3359, 0.2510],
  12720. [-1.5239, -1.7801, 1.4724, -0.5498, -0.4969, -0.6206, 0.1061, 0.2190],
  12721. [ 1.0447, -0.0863, 1.6853, -0.6476, -0.2820, -0.6336, 0.6414, 0.1602],
  12722. [ 0.7545, -0.2793, 1.3460, -0.7836, -0.4708, -0.6146, 0.4870, 0.2678],
  12723. [ 0.8902, -0.1837, 1.6729, -0.4024, -0.5099, -0.6107, 0.3860, 0.1088],
  12724. [ 0.6083, -0.3687, 1.9450, 0.3126, -0.4421, 0.1083, 0.5850, 0.1294],
  12725. [ 0.5531, -0.4151, 1.8876, 0.2921, -0.2572, 0.2739, 0.4322, 0.0999]],
  12726. device='cuda:0', grad_fn=<AddmmBackward>)
  12727. landmarks are: tensor([[[ 0.5393, -0.4299, 0.7453, -1.3775, -0.5596, -1.1081, 0.1527,
  12728. 0.0712],
  12729. [ 0.5483, -0.3991, 0.8030, -1.2159, -0.5018, -1.1928, 0.2624,
  12730. 0.3852],
  12731. [-2.2859, -2.2859, 1.5478, -0.8309, -0.6289, -0.7232, 0.1198,
  12732. 0.1133],
  12733. [ 0.5790, -0.4031, 1.6915, -0.9564, -0.4152, -1.1063, 0.4425,
  12734. 0.2528],
  12735. [ 0.5946, -0.3995, 1.2880, -1.3005, -0.5942, -0.7925, 0.3988,
  12736. 0.2853],
  12737. [ 0.5483, -0.4249, 1.5305, -0.7386, -0.6115, -0.8694, 0.3353,
  12738. 0.1005],
  12739. [ 0.5865, -0.3973, 1.8423, -0.0688, -0.5192, -0.2305, 0.4162,
  12740. 0.1159],
  12741. [ 0.5432, -0.4462, 1.7557, -0.0380, -0.3113, -0.0765, 0.2141,
  12742. 0.3546]]], device='cuda:0')
  12743. loss_train_step before backward: tensor(0.0574, device='cuda:0', grad_fn=<MseLossBackward>)
  12744. loss_train_step after backward: tensor(0.0574, device='cuda:0', grad_fn=<MseLossBackward>)
  12745. loss_train: 1.1515279468148947
  12746. step: 24
  12747. running loss: 0.04798033111728728
  12748. Train Steps: 24/90 Loss: 0.0480 torch.Size([8, 600, 800])
  12749. torch.Size([8, 8])
  12750. tensor([[0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758],
  12751. [0.6289, 0.4081, 0.8720, 0.3487, 0.3900, 0.3183, 0.6703, 0.5376],
  12752. [ nan, nan, 0.7725, 0.2611, 0.3675, 0.2733, 0.5413, 0.5167],
  12753. [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672],
  12754. [0.6075, 0.4000, 0.8513, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280],
  12755. [0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204],
  12756. [0.6179, 0.4008, 0.8600, 0.4015, 0.3932, 0.2515, 0.5711, 0.5438],
  12757. [0.6200, 0.4098, 0.8237, 0.2917, 0.4012, 0.2967, 0.6000, 0.5683]],
  12758. device='cuda:0', dtype=torch.float64)
  12759. predictions are: tensor([[ 0.5418, -0.4124, 1.8235, -0.1119, -0.3812, 0.1074, 0.7229, 0.1254],
  12760. [ 0.6203, -0.3487, 1.6785, -0.5720, -0.4225, -0.8041, 0.6948, 0.1912],
  12761. [-1.7598, -1.9331, 1.1522, -1.0600, -0.4270, -0.9675, -0.0287, 0.3160],
  12762. [ 1.1561, -0.0339, 1.7192, 0.2305, -0.5978, -0.1364, 0.6343, 0.0543],
  12763. [ 0.7348, -0.2362, 1.6043, 0.1218, -0.2008, 0.1591, 0.0928, 0.1539],
  12764. [ 0.6645, -0.3278, 1.6605, 0.1640, -0.3562, 0.2098, 0.6699, 0.1587],
  12765. [ 0.8523, -0.1380, 1.3972, -0.5103, -0.4060, -0.9285, 0.1484, 0.2288],
  12766. [ 0.9301, -0.1348, 1.4233, -0.9516, -0.4670, -0.7360, 0.4224, 0.2452]],
  12767. device='cuda:0', grad_fn=<AddmmBackward>)
  12768. landmarks are: tensor([[[ 0.6072, -0.4250, 1.8249, -0.0072, -0.4037, 0.1082, 0.6730,
  12769. -0.0881],
  12770. [ 0.6191, -0.4008, 1.7420, -0.6753, -0.4845, -0.8156, 0.8101,
  12771. 0.1974],
  12772. [-2.2859, -2.2859, 1.2820, -1.0801, -0.5885, -1.0234, 0.2141,
  12773. 0.1005],
  12774. [ 0.6128, -0.4375, 1.7788, 0.2699, -0.6346, -0.2536, 0.5463,
  12775. -0.1278],
  12776. [ 0.5202, -0.4382, 1.6460, 0.1082, -0.2029, 0.1754, 0.1067,
  12777. 0.1530],
  12778. [ 0.6042, -0.4273, 1.7198, 0.2184, -0.3478, 0.1149, 0.8062,
  12779. 0.1176],
  12780. [ 0.5680, -0.4345, 1.6864, -0.4315, -0.4698, -1.1241, 0.3518,
  12781. 0.2261],
  12782. [ 0.5778, -0.3930, 1.5189, -0.9387, -0.4326, -0.9156, 0.4855,
  12783. 0.3392]]], device='cuda:0')
  12784. loss_train_step before backward: tensor(0.0297, device='cuda:0', grad_fn=<MseLossBackward>)
  12785. loss_train_step after backward: tensor(0.0297, device='cuda:0', grad_fn=<MseLossBackward>)
  12786. loss_train: 1.1811874378472567
  12787. step: 25
  12788. running loss: 0.04724749751389026
  12789. Train Steps: 25/90 Loss: 0.0472 torch.Size([8, 600, 800])
  12790. torch.Size([8, 8])
  12791. tensor([[0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461],
  12792. [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5363, 0.5550],
  12793. [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5988, 0.5283],
  12794. [0.6357, 0.4097, 0.9038, 0.3883, 0.4213, 0.2950, 0.6686, 0.5390],
  12795. [0.6260, 0.4214, 0.8538, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983],
  12796. [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123],
  12797. [ nan, nan, 0.6900, 0.1917, 0.3937, 0.2367, 0.5240, 0.5246],
  12798. [0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5787, 0.5600]],
  12799. device='cuda:0', dtype=torch.float64)
  12800. predictions are: tensor([[ 1.1092, -0.0598, 1.7175, -0.5010, -0.7258, -0.0769, 0.5840, 0.1207],
  12801. [ 0.7990, -0.2687, 1.0544, -0.9522, -0.4777, -1.0258, 0.1616, 0.2518],
  12802. [ 0.6795, -0.3569, 1.8876, 0.3645, -0.4024, 0.2626, 0.5803, 0.1490],
  12803. [ 0.8305, -0.2397, 1.7730, -0.4483, -0.3952, -0.6633, 0.7930, 0.1416],
  12804. [ 1.0673, -0.0639, 1.7255, 0.5401, -0.5086, -0.2005, 0.5255, 0.2547],
  12805. [ 0.7767, -0.2754, 1.1634, -1.0008, -0.5500, -0.8639, 0.2313, 0.1580],
  12806. [-1.7676, -1.9638, 1.1525, -1.1372, -0.3661, -1.0775, 0.0339, 0.3052],
  12807. [ 0.5077, -0.4275, 1.8660, 0.0588, -0.0673, 0.2542, 0.3526, 0.2022]],
  12808. device='cuda:0', grad_fn=<AddmmBackward>)
  12809. landmarks are: tensor([[[ 0.5124, -0.4483, 1.5420, -0.6231, -0.7155, -0.4152, 0.2679,
  12810. 0.2365],
  12811. [ 0.5448, -0.3859, 0.9242, -1.1466, -0.4152, -1.3005, 0.1910,
  12812. 0.2776],
  12813. [ 0.5603, -0.4319, 1.7788, 0.1775, -0.5538, -0.0611, 0.4797,
  12814. 0.1544],
  12815. [ 0.6504, -0.3936, 1.8885, -0.4922, -0.3402, -0.9233, 0.8022,
  12816. 0.2035],
  12817. [ 0.6055, -0.3393, 1.6575, 0.2545, -0.5942, -0.5461, 0.2949,
  12818. 0.4778],
  12819. [ 0.5342, -0.4280, 1.0051, -1.0619, -0.5711, -1.0388, 0.2226,
  12820. 0.0802],
  12821. [-2.2859, -2.2859, 0.9012, -1.4006, -0.4672, -1.1928, 0.1342,
  12822. 0.1373],
  12823. [ 0.5505, -0.4207, 1.7095, -0.0534, -0.0509, 0.1050, 0.3873,
  12824. 0.3007]]], device='cuda:0')
  12825. loss_train_step before backward: tensor(0.0428, device='cuda:0', grad_fn=<MseLossBackward>)
  12826.  
  12827. loss_train_step after backward: tensor(0.0428, device='cuda:0', grad_fn=<MseLossBackward>)
  12828. loss_train: 1.2239569071680307
  12829. step: 26
  12830. running loss: 0.047075265660308875
  12831. Train Steps: 26/90 Loss: 0.0471 torch.Size([8, 600, 800])
  12832. torch.Size([8, 8])
  12833. tensor([[0.6138, 0.4101, 0.8800, 0.5083, 0.4637, 0.5950, 0.5587, 0.5077],
  12834. [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320],
  12835. [0.6125, 0.4076, 0.8488, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505],
  12836. [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879],
  12837. [0.6148, 0.4076, 0.8666, 0.4820, 0.4138, 0.5067, 0.5250, 0.5767],
  12838. [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167],
  12839. [0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742],
  12840. [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892]],
  12841. device='cuda:0', dtype=torch.float64)
  12842. predictions are: tensor([[ 0.5252, -0.4151, 1.6701, 0.0175, -0.1096, 0.1072, 0.4594, 0.2154],
  12843. [ 0.4531, -0.4956, 1.5928, 0.0686, -0.4088, 0.0361, 0.6365, 0.2085],
  12844. [ 0.2903, -0.5712, 1.5617, -0.4637, -0.6604, -0.6051, 0.1524, 0.2101],
  12845. [ 0.3504, -0.5716, 1.7172, -0.3043, -0.4345, 0.1715, 0.5451, 0.1113],
  12846. [ 0.3246, -0.5665, 1.6040, -0.0296, -0.3346, 0.0444, 0.3115, 0.2818],
  12847. [ 0.7752, -0.2857, 1.4962, -0.7700, -0.5900, -0.9204, 0.3944, 0.1760],
  12848. [ 1.0919, -0.0991, 1.5440, -0.8091, -0.4919, -1.0663, 0.7723, 0.1144],
  12849. [ 0.3260, -0.5237, 1.5657, -0.0825, -0.6330, -0.5299, 0.1789, 0.2688]],
  12850. device='cuda:0', grad_fn=<AddmmBackward>)
  12851. landmarks are: tensor([[[ 0.5491, -0.3918, 1.7788, 0.0620, -0.1439, 0.4624, 0.2946,
  12852. 0.0592],
  12853. [ 0.6273, -0.4393, 1.6402, 0.1313, -0.5076, 0.0467, 1.1532,
  12854. 0.1715],
  12855. [ 0.5433, -0.4032, 1.6344, -0.4922, -0.5769, -0.5846, 0.0357,
  12856. 0.2567],
  12857. [ 0.5676, -0.4327, 1.8252, -0.2095, -0.4886, 0.3084, 0.4727,
  12858. -0.0322],
  12859. [ 0.5539, -0.4031, 1.7168, -0.0594, -0.3748, 0.0543, 0.1390,
  12860. 0.3777],
  12861. [ 0.5483, -0.4249, 1.5305, -0.7386, -0.6115, -0.8694, 0.3353,
  12862. 0.1005],
  12863. [ 0.6134, -0.4218, 1.7268, -0.6154, -0.4730, -1.0850, 0.5463,
  12864. -0.0957],
  12865. [ 0.5454, -0.4053, 1.6633, -0.1766, -0.6058, -0.5923, 0.1577,
  12866. 0.4357]]], device='cuda:0')
  12867. loss_train_step before backward: tensor(0.0250, device='cuda:0', grad_fn=<MseLossBackward>)
  12868. loss_train_step after backward: tensor(0.0250, device='cuda:0', grad_fn=<MseLossBackward>)
  12869. loss_train: 1.2489608246833086
  12870. step: 27
  12871. running loss: 0.046257808321604026
  12872. Train Steps: 27/90 Loss: 0.0463 torch.Size([8, 600, 800])
  12873. torch.Size([8, 8])
  12874. tensor([[0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5787, 0.5600],
  12875. [0.6207, 0.4081, 0.7662, 0.2067, 0.3962, 0.3200, 0.6312, 0.5300],
  12876. [0.6127, 0.4084, 0.8700, 0.4467, 0.3987, 0.4317, 0.5013, 0.5471],
  12877. [0.6200, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183],
  12878. [0.6262, 0.4052, 0.8888, 0.4700, 0.3675, 0.5117, 0.6350, 0.5233],
  12879. [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450],
  12880. [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5963, 0.6217],
  12881. [0.6206, 0.4001, 0.8900, 0.3933, 0.3588, 0.3567, 0.5837, 0.5083]],
  12882. device='cuda:0', dtype=torch.float64)
  12883. predictions are: tensor([[ 0.4644, -0.4701, 1.7021, 0.0982, -0.1020, 0.1515, 0.2594, 0.2304],
  12884. [ 0.6701, -0.3498, 1.1743, -1.0372, -0.5572, -0.7721, 0.4599, 0.2256],
  12885. [ 0.5408, -0.4026, 1.7313, 0.0126, -0.4380, -0.1173, 0.1420, 0.1929],
  12886. [ 0.6182, -0.3839, 1.3790, -0.8283, -0.5608, -0.8546, 0.4777, 0.1537],
  12887. [ 0.7149, -0.3733, 1.9399, 0.0853, -0.4579, 0.1658, 0.8671, 0.1258],
  12888. [ 0.4842, -0.4774, 1.7424, 0.0258, -0.5271, -0.1801, 0.4708, 0.1866],
  12889. [ 0.0581, -0.7235, 1.0430, -0.9465, -0.4204, -1.1660, 0.2980, 0.3445],
  12890. [ 0.5249, -0.4811, 1.8443, -0.2301, -0.6742, -0.5042, 0.4780, 0.0843]],
  12891. device='cuda:0', grad_fn=<AddmmBackward>)
  12892. landmarks are: tensor([[[ 0.5505, -0.4207, 1.7095, -0.0534, -0.0509, 0.1050, 0.3873,
  12893. 0.3007],
  12894. [ 0.5809, -0.4011, 1.2533, -1.3313, -0.4557, -0.8079, 0.6298,
  12895. 0.1621],
  12896. [ 0.5441, -0.3997, 1.7326, -0.2228, -0.4441, -0.2921, 0.0296,
  12897. 0.2409],
  12898. [ 0.5781, -0.4129, 1.4142, -0.9057, -0.5115, -0.9937, 0.4621,
  12899. 0.1080],
  12900. [ 0.6063, -0.4142, 1.8192, -0.1150, -0.5885, 0.0774, 0.6471,
  12901. 0.1313],
  12902. [ 0.5785, -0.4252, 1.7676, -0.1602, -0.5845, -0.3446, 0.4566,
  12903. 0.2314],
  12904. [ 0.5687, -0.4088, 1.0397, -1.1466, -0.3113, -1.1928, 0.4681,
  12905. 0.5855],
  12906. [ 0.5807, -0.4378, 1.8249, -0.4691, -0.6289, -0.6385, 0.4104,
  12907. 0.0620]]], device='cuda:0')
  12908. loss_train_step before backward: tensor(0.0180, device='cuda:0', grad_fn=<MseLossBackward>)
  12909. loss_train_step after backward: tensor(0.0180, device='cuda:0', grad_fn=<MseLossBackward>)
  12910. loss_train: 1.2669219318777323
  12911. step: 28
  12912. running loss: 0.04524721185277615
  12913. Train Steps: 28/90 Loss: 0.0452 torch.Size([8, 600, 800])
  12914. torch.Size([8, 8])
  12915. tensor([[0.6200, 0.4049, 0.8638, 0.5617, 0.4125, 0.5100, 0.6013, 0.5317],
  12916. [0.6126, 0.4073, 0.8750, 0.5133, 0.3800, 0.4333, 0.4986, 0.5378],
  12917. [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611],
  12918. [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170],
  12919. [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374],
  12920. [0.6278, 0.4253, 0.8875, 0.5017, 0.4113, 0.2750, 0.5413, 0.6196],
  12921. [ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600],
  12922. [0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719]],
  12923. device='cuda:0', dtype=torch.float64)
  12924. predictions are: tensor([[ 0.3292, -0.6127, 1.6667, 0.2913, -0.3799, 0.2480, 0.4851, 0.1940],
  12925. [ 0.5045, -0.4484, 1.7734, 0.2048, -0.5201, -0.0983, 0.3444, 0.1669],
  12926. [ 0.9764, -0.1649, 1.7343, -0.0197, -0.3502, 0.1187, 0.3336, 0.2008],
  12927. [ 0.6445, -0.4002, 1.4565, -1.2108, -0.2055, -1.2694, 0.7895, 0.1524],
  12928. [ 0.4371, -0.5265, 1.5985, -0.6078, -0.6801, -0.6091, 0.4320, 0.1052],
  12929. [ 0.9756, -0.1525, 1.5874, 0.0874, -0.4362, -0.7796, 0.5513, 0.2914],
  12930. [-0.1907, -0.9122, 1.2446, -1.0707, -0.4485, -0.9089, 0.2790, 0.2335],
  12931. [ 0.3977, -0.5413, 1.5528, -0.5059, -0.7190, -0.1582, 0.2814, 0.1526]],
  12932. device='cuda:0', grad_fn=<AddmmBackward>)
  12933. landmarks are: tensor([[[ 0.5776, -0.4159, 1.7037, 0.3084, -0.3806, 0.0697, 0.4912,
  12934. 0.1698],
  12935. [ 0.5435, -0.4045, 1.7557, 0.0851, -0.5307, -0.2844, 0.0172,
  12936. 0.1980],
  12937. [ 0.5491, -0.4290, 1.7788, -0.1073, -0.2651, -0.1073, 0.2555,
  12938. 0.3057],
  12939. [ 0.6471, -0.3840, 1.5767, -1.0311, -0.0457, -1.5007, 0.6889,
  12940. 0.1020],
  12941. [ 0.5777, -0.4416, 1.7044, -0.5827, -0.5962, -0.8361, 0.4862,
  12942. 0.1963],
  12943. [ 0.6139, -0.3216, 1.8134, 0.0313, -0.3864, -1.0157, 0.2144,
  12944. 0.5762],
  12945. [-2.2859, -2.2859, 1.4266, -1.1389, -0.4499, -1.1235, 0.2891,
  12946. 0.3007],
  12947. [ 0.5367, -0.4294, 1.5709, -0.4999, -0.6693, -0.3075, 0.2455,
  12948. 0.3559]]], device='cuda:0')
  12949. loss_train_step before backward: tensor(0.1221, device='cuda:0', grad_fn=<MseLossBackward>)
  12950. loss_train_step after backward: tensor(0.1221, device='cuda:0', grad_fn=<MseLossBackward>)
  12951. loss_train: 1.3889758083969355
  12952. step: 29
  12953. running loss: 0.04789571753092881
  12954. Train Steps: 29/90 Loss: 0.0479 torch.Size([8, 600, 800])
  12955. torch.Size([8, 8])
  12956. tensor([[0.6270, 0.4267, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183],
  12957. [0.6111, 0.3995, 0.8788, 0.4567, 0.3813, 0.4833, 0.5450, 0.5700],
  12958. [0.6200, 0.3978, 0.8900, 0.4550, 0.3775, 0.5200, 0.6150, 0.5367],
  12959. [0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038],
  12960. [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478],
  12961. [ nan, nan, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819],
  12962. [0.6076, 0.3953, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954],
  12963. [0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373]],
  12964. device='cuda:0', dtype=torch.float64)
  12965. predictions are: tensor([[ 0.6447, -0.3391, 1.0203, -1.0295, -0.2438, -1.3132, 0.3005, 0.3568],
  12966. [ 0.7181, -0.3352, 1.7661, -0.0676, -0.5545, 0.0418, 0.3674, 0.1823],
  12967. [ 0.3253, -0.6027, 1.8249, -0.0198, -0.5174, 0.2183, 0.4973, 0.1816],
  12968. [ 1.0088, -0.1655, 1.8082, -0.0407, -0.3024, 0.0897, 0.3959, 0.1048],
  12969. [ 0.6614, -0.4317, 1.8519, -0.5815, -0.5667, -0.8422, 0.6848, 0.1938],
  12970. [-0.8125, -1.3321, 1.0198, -1.1514, -0.5100, -1.2320, 0.2646, 0.2864],
  12971. [ 0.6633, -0.3855, 1.5774, -0.5996, -0.7283, -0.3927, 0.4222, 0.1267],
  12972. [ 0.8054, -0.3088, 1.6503, 0.4464, -0.5051, -0.2037, 0.5452, 0.1913]],
  12973. device='cuda:0', grad_fn=<AddmmBackward>)
  12974. landmarks are: tensor([[[ 6.1010e-01, -3.1524e-01, 1.0166e+00, -7.5396e-01, -2.2633e-02,
  12975. -1.4468e+00, 2.8226e-01, 5.7018e-01],
  12976. [ 5.3684e-01, -4.4057e-01, 1.7730e+00, -1.7660e-01, -5.2494e-01,
  12977. -5.3426e-02, 2.3141e-01, 3.4688e-01],
  12978. [ 5.7760e-01, -4.4842e-01, 1.8249e+00, -1.8430e-01, -5.4226e-01,
  12979. 1.1594e-01, 5.5473e-01, 1.9292e-01],
  12980. [ 5.2494e-01, -4.4734e-01, 1.7326e+00, -9.1917e-02, -2.0162e-01,
  12981. 1.5443e-01, 1.7328e-01, 4.1158e-02],
  12982. [ 5.7921e-01, -4.0523e-01, 1.8214e+00, -6.5874e-01, -5.3842e-01,
  12983. -8.9239e-01, 4.3812e-01, 2.4425e-01],
  12984. [-2.2859e+00, -2.2859e+00, 9.4385e-01, -9.9666e-01, -4.6143e-01,
  12985. -1.1851e+00, 2.4679e-01, 4.0188e-01],
  12986. [ 5.2061e-01, -4.6028e-01, 1.4670e+00, -4.7683e-01, -6.5774e-01,
  12987. -3.9985e-01, 1.8356e-01, 2.0831e-03],
  12988. [ 5.6801e-01, -4.5619e-01, 1.5697e+00, 4.9469e-01, -4.9038e-01,
  12989. -1.5026e-01, 3.5357e-01, 1.9563e-01]]], device='cuda:0')
  12990. loss_train_step before backward: tensor(0.0670, device='cuda:0', grad_fn=<MseLossBackward>)
  12991. loss_train_step after backward: tensor(0.0670, device='cuda:0', grad_fn=<MseLossBackward>)
  12992. loss_train: 1.4559749905019999
  12993. step: 30
  12994. running loss: 0.048532499683399996
  12995.  
  12996. Train Steps: 30/90 Loss: 0.0485 torch.Size([8, 600, 800])
  12997. torch.Size([8, 8])
  12998. tensor([[0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748],
  12999. [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083],
  13000. [0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117],
  13001. [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5463, 0.5800],
  13002. [0.6125, 0.3999, 0.8750, 0.4883, 0.4750, 0.4700, 0.5533, 0.5617],
  13003. [0.6179, 0.4040, 0.7412, 0.1850, 0.3825, 0.2783, 0.5837, 0.5600],
  13004. [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882],
  13005. [0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938]],
  13006. device='cuda:0', dtype=torch.float64)
  13007. predictions are: tensor([[ 0.3969, -0.5640, 1.6768, 0.0966, -0.4565, -0.4329, 0.6561, 0.2496],
  13008. [ 0.2968, -0.6093, 1.6604, -0.1043, -0.5758, -0.1632, 0.3509, 0.3169],
  13009. [ 0.7539, -0.3650, 1.8065, -0.1882, -0.5726, -0.0869, 0.6342, 0.1433],
  13010. [ 0.2691, -0.5931, 1.1348, -1.0432, -0.6014, -1.0747, 0.2123, 0.2751],
  13011. [ 0.8217, -0.2588, 1.6736, -0.1274, -0.1709, -0.0524, 0.2523, 0.2500],
  13012. [ 0.5199, -0.5038, 1.1993, -1.2389, -0.5359, -1.1787, 0.5091, 0.2795],
  13013. [ 0.3096, -0.5846, 1.7719, -0.4621, -0.6384, -0.5422, 0.2241, 0.1382],
  13014. [ 0.4792, -0.4978, 1.7307, 0.2366, -0.4688, -0.1269, 0.5286, 0.1288]],
  13015. device='cuda:0', grad_fn=<AddmmBackward>)
  13016. landmarks are: tensor([[[ 0.6421, -0.3647, 1.5940, 0.3084, -0.5192, -0.3691, 1.1057,
  13017. 0.3692],
  13018. [ 0.5713, -0.4105, 1.7557, 0.0467, -0.6520, -0.2690, 0.3988,
  13019. 0.5239],
  13020. [ 0.6262, -0.4461, 1.8480, -0.0534, -0.5827, -0.1227, 0.6587,
  13021. 0.0774],
  13022. [ 0.5548, -0.3936, 1.1634, -0.8105, -0.5192, -1.0696, 0.2372,
  13023. 0.3931],
  13024. [ 0.5432, -0.4388, 1.7557, -0.0303, -0.0919, -0.1150, 0.2699,
  13025. 0.3087],
  13026. [ 0.5680, -0.4201, 1.1379, -1.4314, -0.5192, -1.0003, 0.4104,
  13027. 0.3007],
  13028. [ 0.5595, -0.3988, 1.7672, -0.4460, -0.5538, -0.5384, 0.0828,
  13029. -0.0310],
  13030. [ 0.6395, -0.4213, 1.7037, 0.3623, -0.4326, -0.1073, 0.6560,
  13031. -0.0049]]], device='cuda:0')
  13032. loss_train_step before backward: tensor(0.0203, device='cuda:0', grad_fn=<MseLossBackward>)
  13033. loss_train_step after backward: tensor(0.0203, device='cuda:0', grad_fn=<MseLossBackward>)
  13034. loss_train: 1.4763062540441751
  13035. step: 31
  13036. running loss: 0.04762278238852178
  13037. Train Steps: 31/90 Loss: 0.0476 torch.Size([8, 600, 800])
  13038. torch.Size([8, 8])
  13039. tensor([[0.6109, 0.4041, 0.6975, 0.3167, 0.3513, 0.3383, 0.5153, 0.5319],
  13040. [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750],
  13041. [0.6224, 0.4179, 0.8700, 0.5683, 0.4037, 0.4683, 0.5650, 0.5600],
  13042. [0.6139, 0.4019, 0.7137, 0.2150, 0.4375, 0.1533, 0.5293, 0.5006],
  13043. [0.6224, 0.4061, 0.8988, 0.4300, 0.3838, 0.4750, 0.6112, 0.5483],
  13044. [0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297],
  13045. [0.6250, 0.4103, 0.8950, 0.4400, 0.3912, 0.5650, 0.6050, 0.5133],
  13046. [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167]],
  13047. device='cuda:0', dtype=torch.float64)
  13048. predictions are: tensor([[ 0.2273, -0.6596, 1.1092, -0.9489, -0.6768, -0.8663, 0.0910, 0.2431],
  13049. [ 0.7486, -0.3217, 1.6883, 0.0425, -0.3411, -0.1704, 0.2981, 0.2276],
  13050. [ 0.5208, -0.4361, 1.6414, 0.4044, -0.4322, -0.2129, 0.3308, 0.3269],
  13051. [-0.0311, -0.8324, 1.0396, -1.3393, -0.3552, -1.6871, 0.3319, 0.1816],
  13052. [ 0.7902, -0.3247, 1.8514, -0.2893, -0.5331, -0.0267, 0.5042, 0.1592],
  13053. [ 0.3829, -0.5580, 1.7809, 0.2276, -0.5510, -0.3927, 0.5745, 0.2348],
  13054. [ 0.6664, -0.3745, 1.8520, -0.2423, -0.4796, 0.2054, 0.6264, 0.1599],
  13055. [ 0.4054, -0.5725, 1.5891, -0.7661, -0.6209, -0.6981, 0.5075, 0.3368]],
  13056. device='cuda:0', grad_fn=<AddmmBackward>)
  13057. landmarks are: tensor([[[ 0.5359, -0.4193, 0.9358, -0.8232, -0.6635, -0.7232, 0.0943,
  13058. 0.1710],
  13059. [ 0.5525, -0.4463, 1.7326, 0.0313, -0.2536, -0.0688, 0.1968,
  13060. 0.3700],
  13061. [ 0.5891, -0.3557, 1.7326, 0.3392, -0.4210, -0.1227, 0.3238,
  13062. 0.3007],
  13063. [ 0.5496, -0.4295, 1.0109, -1.2928, -0.2651, -1.5777, 0.1589,
  13064. 0.0261],
  13065. [ 0.5888, -0.4101, 1.8654, -0.2998, -0.5134, -0.0919, 0.5374,
  13066. 0.2468],
  13067. [ 0.6504, -0.3647, 1.7730, 0.2930, -0.6058, -0.2382, 0.7109,
  13068. 0.1608],
  13069. [ 0.6010, -0.3909, 1.8480, -0.2536, -0.4788, 0.3238, 0.5085,
  13070. 0.0851],
  13071. [ 0.5768, -0.3899, 1.3861, -0.7771, -0.5885, -0.5461, 0.5028,
  13072. 0.5624]]], device='cuda:0')
  13073. loss_train_step before backward: tensor(0.0216, device='cuda:0', grad_fn=<MseLossBackward>)
  13074. loss_train_step after backward: tensor(0.0216, device='cuda:0', grad_fn=<MseLossBackward>)
  13075. loss_train: 1.4978971872478724
  13076. step: 32
  13077. running loss: 0.04680928710149601
  13078. Train Steps: 32/90 Loss: 0.0468 torch.Size([8, 600, 800])
  13079. torch.Size([8, 8])
  13080. tensor([[0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266],
  13081. [0.6202, 0.4066, 0.8746, 0.3376, 0.3717, 0.3090, 0.5842, 0.5165],
  13082. [ nan, nan, 0.8300, 0.3150, 0.3588, 0.3383, 0.5208, 0.5194],
  13083. [0.6179, 0.4008, 0.8600, 0.4015, 0.3932, 0.2515, 0.5711, 0.5438],
  13084. [0.6243, 0.4128, 0.7762, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417],
  13085. [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378],
  13086. [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332],
  13087. [0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118]],
  13088. device='cuda:0', dtype=torch.float64)
  13089. predictions are: tensor([[ 0.6682, -0.3635, 1.1771, -0.8218, -0.3968, -0.9152, 0.4136, 0.3601],
  13090. [ 0.6833, -0.3600, 1.7315, -0.4808, -0.6057, -0.7120, 0.4325, 0.1391],
  13091. [-1.7206, -1.9316, 1.4490, -0.7641, -0.6125, -0.7112, 0.2333, 0.2175],
  13092. [ 0.4376, -0.4658, 1.5066, -0.4000, -0.4101, -0.9178, 0.2045, 0.2454],
  13093. [ 0.2897, -0.5951, 1.3823, -0.8484, -0.4510, -0.7861, 0.5288, 0.2703],
  13094. [ 1.0572, -0.1334, 1.7365, 0.1202, -0.3583, 0.3940, 0.6201, 0.1901],
  13095. [ 1.1275, -0.1140, 1.7753, 0.0943, -0.6044, -0.1970, 0.5039, 0.1713],
  13096. [ 0.6546, -0.3253, 1.1122, -0.6033, -0.3702, -0.9839, 0.1899, 0.3367]],
  13097. device='cuda:0', grad_fn=<AddmmBackward>)
  13098. landmarks are: tensor([[[ 0.5746, -0.4021, 1.0801, -1.1312, -0.3229, -1.1081, 0.4803,
  13099. 0.6084],
  13100. [ 0.5787, -0.4077, 1.7537, -0.7263, -0.5692, -0.8586, 0.4126,
  13101. 0.1000],
  13102. [-2.2859, -2.2859, 1.5478, -0.8309, -0.6289, -0.7232, 0.1198,
  13103. 0.1133],
  13104. [ 0.5680, -0.4345, 1.6864, -0.4315, -0.4698, -1.1241, 0.3518,
  13105. 0.2261],
  13106. [ 0.5977, -0.3792, 1.2995, -1.0311, -0.5192, -0.8386, 0.5836,
  13107. 0.2160],
  13108. [ 0.6141, -0.4345, 1.6864, -0.0303, -0.2882, 0.1544, 0.9521,
  13109. 0.1982],
  13110. [ 0.6520, -0.4056, 1.9173, -0.0765, -0.5596, -0.4537, 0.7949,
  13111. 0.1768],
  13112. [ 0.6077, -0.3226, 0.9993, -0.6462, -0.2651, -1.3082, 0.2946,
  13113. 0.5401]]], device='cuda:0')
  13114. loss_train_step before backward: tensor(0.0360, device='cuda:0', grad_fn=<MseLossBackward>)
  13115. loss_train_step after backward: tensor(0.0360, device='cuda:0', grad_fn=<MseLossBackward>)
  13116. loss_train: 1.533901834860444
  13117. step: 33
  13118. running loss: 0.04648187378364982
  13119. Train Steps: 33/90 Loss: 0.0465 torch.Size([8, 600, 800])
  13120. torch.Size([8, 8])
  13121. tensor([[0.6147, 0.4107, 0.8137, 0.3333, 0.3750, 0.2683, 0.5006, 0.5412],
  13122. [0.6205, 0.4062, 0.8337, 0.2683, 0.3675, 0.4283, 0.6338, 0.5250],
  13123. [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123],
  13124. [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150],
  13125. [0.6364, 0.4165, 0.9088, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297],
  13126. [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131],
  13127. [0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750],
  13128. [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483]],
  13129. device='cuda:0', dtype=torch.float64)
  13130. predictions are: tensor([[ 0.1480, -0.7081, 1.4007, -0.5825, -0.5700, -1.0408, 0.2384, 0.2237],
  13131. [ 0.5896, -0.4433, 1.4925, -0.8253, -0.5779, -0.3485, 0.6271, 0.2747],
  13132. [ 0.3425, -0.5901, 1.0766, -0.9527, -0.5053, -1.1356, 0.2093, 0.2348],
  13133. [ 0.8501, -0.2678, 1.8594, 0.1508, -0.2696, 0.3967, 0.5798, 0.2531],
  13134. [ 0.2669, -0.6123, 1.8149, -0.1087, -0.3577, -0.8990, 0.6010, 0.2930],
  13135. [ 0.7211, -0.3709, 1.6834, -0.3010, -0.6311, -0.5038, 0.3394, 0.2056],
  13136. [ 0.0754, -0.7388, 1.3125, -0.8194, -0.4243, -1.1414, 0.2379, 0.2833],
  13137. [ 0.3136, -0.6088, 1.6698, -0.2275, -0.5166, -0.0057, 0.3620, 0.2574]],
  13138. device='cuda:0', grad_fn=<AddmmBackward>)
  13139. landmarks are: tensor([[[ 0.5532, -0.3888, 1.4727, -0.7463, -0.5538, -1.0465, 0.0265,
  13140. 0.2138],
  13141. [ 0.5799, -0.4099, 1.5651, -1.0465, -0.5885, -0.3075, 0.6413,
  13142. 0.1390],
  13143. [ 0.5342, -0.4280, 1.0051, -1.0619, -0.5711, -1.0388, 0.2226,
  13144. 0.0802],
  13145. [ 0.6014, -0.3883, 1.8192, -0.1150, -0.2882, 0.4008, 0.5201,
  13146. 0.0928],
  13147. [ 0.6537, -0.3623, 1.9115, -0.2690, -0.4037, -0.8309, 0.6926,
  13148. 0.1608],
  13149. [ 0.5783, -0.4216, 1.6847, -0.5078, -0.6732, -0.5377, 0.4752,
  13150. 0.0839],
  13151. [ 0.5695, -0.3927, 1.3226, -0.9002, -0.4672, -1.1928, 0.1737,
  13152. 0.3700],
  13153. [ 0.5546, -0.4620, 1.7788, -0.4229, -0.5192, -0.0226, 0.4277,
  13154. 0.2468]]], device='cuda:0')
  13155. loss_train_step before backward: tensor(0.0253, device='cuda:0', grad_fn=<MseLossBackward>)
  13156. loss_train_step after backward: tensor(0.0253, device='cuda:0', grad_fn=<MseLossBackward>)
  13157. loss_train: 1.5591852515935898
  13158. step: 34
  13159. running loss: 0.04585838975275264
  13160.  
  13161. Train Steps: 34/90 Loss: 0.0459 torch.Size([8, 600, 800])
  13162. torch.Size([8, 8])
  13163. tensor([[0.6266, 0.4067, 0.8588, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267],
  13164. [0.6200, 0.4118, 0.8287, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799],
  13165. [0.6142, 0.3982, 0.8650, 0.4883, 0.3912, 0.4317, 0.5315, 0.5350],
  13166. [0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250],
  13167. [0.6171, 0.4127, 0.8900, 0.4800, 0.4325, 0.5783, 0.5769, 0.5090],
  13168. [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301],
  13169. [0.6198, 0.4164, 0.8700, 0.5067, 0.4625, 0.5650, 0.5464, 0.5197],
  13170. [0.6153, 0.4117, 0.8688, 0.5167, 0.4895, 0.5647, 0.5524, 0.5136]],
  13171. device='cuda:0', dtype=torch.float64)
  13172. predictions are: tensor([[ 0.3723, -0.5713, 1.5412, -1.1410, -0.4123, -1.2833, 0.7458, 0.2572],
  13173. [-0.0439, -0.7983, 1.3700, -0.6209, -0.6206, -1.1788, 0.1694, 0.3199],
  13174. [ 0.4568, -0.5009, 1.5959, -0.0868, -0.5835, -0.4344, 0.1355, 0.2190],
  13175. [ 0.3854, -0.5706, 1.7432, -0.5431, -0.6401, -0.9886, 0.6589, 0.2581],
  13176. [ 0.6753, -0.3385, 1.6879, -0.1744, -0.3901, 0.1000, 0.3988, 0.2561],
  13177. [ 0.6335, -0.3926, 1.5830, -0.1774, -0.4380, -0.2220, 0.5967, 0.2536],
  13178. [ 0.2948, -0.5686, 1.6185, -0.0778, -0.3593, -0.0700, 0.2592, 0.2595],
  13179. [ 0.4754, -0.4777, 1.6022, -0.0768, -0.2555, -0.0874, 0.2974, 0.2777]],
  13180. device='cuda:0', grad_fn=<AddmmBackward>)
  13181. landmarks are: tensor([[[ 0.6084, -0.4076, 1.6806, -0.9618, -0.2998, -0.9695, 0.6356,
  13182. 0.1467],
  13183. [ 0.5779, -0.3840, 1.5420, -0.4306, -0.5423, -0.9772, 0.2041,
  13184. 0.3928],
  13185. [ 0.5513, -0.4467, 1.7095, -0.0303, -0.4788, -0.2921, 0.1692,
  13186. 0.1854],
  13187. [ 0.6111, -0.3828, 1.8885, -0.3844, -0.5654, -0.8079, 0.5663,
  13188. 0.1390],
  13189. [ 0.5645, -0.3797, 1.8249, -0.0688, -0.2882, 0.3854, 0.3789,
  13190. 0.0652],
  13191. [ 0.6010, -0.4562, 1.7198, -0.0090, -0.3464, 0.0108, 0.6294,
  13192. 0.1627],
  13193. [ 0.5770, -0.3624, 1.7326, 0.0543, -0.1497, 0.3238, 0.2378,
  13194. 0.1146],
  13195. [ 0.5564, -0.3842, 1.7268, 0.1005, -0.0250, 0.3225, 0.2658,
  13196. 0.0862]]], device='cuda:0')
  13197. loss_train_step before backward: tensor(0.0343, device='cuda:0', grad_fn=<MseLossBackward>)
  13198. loss_train_step after backward: tensor(0.0343, device='cuda:0', grad_fn=<MseLossBackward>)
  13199. loss_train: 1.593443851917982
  13200. step: 35
  13201. running loss: 0.04552696719765663
  13202. Train Steps: 35/90 Loss: 0.0455 torch.Size([8, 600, 800])
  13203. torch.Size([8, 8])
  13204. tensor([[0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869],
  13205. [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791],
  13206. [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150],
  13207. [0.6135, 0.4115, 0.8838, 0.4667, 0.4288, 0.6050, 0.5778, 0.5097],
  13208. [0.6196, 0.4088, 0.8888, 0.4583, 0.4500, 0.5683, 0.6138, 0.5883],
  13209. [0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367],
  13210. [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098],
  13211. [0.6079, 0.3964, 0.7420, 0.2958, 0.3563, 0.2917, 0.5351, 0.4980]],
  13212. device='cuda:0', dtype=torch.float64)
  13213. predictions are: tensor([[ 0.1774, -0.6739, 1.3619, -1.0813, -0.2807, -1.6227, 0.5180, 0.2143],
  13214. [ 0.3338, -0.5876, 1.7390, -0.0251, -0.5067, -0.1704, 0.4971, 0.2085],
  13215. [ 0.4705, -0.4790, 1.7364, -0.0638, -0.3783, 0.0956, 0.4293, 0.2739],
  13216. [ 0.6925, -0.3200, 1.7505, -0.1045, -0.3952, 0.1059, 0.3062, 0.2407],
  13217. [ 0.6138, -0.3833, 1.7019, -0.2001, -0.2819, 0.0471, 0.4653, 0.3456],
  13218. [ 0.4023, -0.5554, 1.5965, -0.7829, -0.4790, -1.2861, 0.6305, 0.2734],
  13219. [ 0.3821, -0.5534, 1.6308, -0.1025, -0.5651, -0.0953, 0.3352, 0.2354],
  13220. [ 0.2864, -0.6015, 1.1239, -0.9904, -0.6295, -1.1959, 0.0964, 0.2094]],
  13221. device='cuda:0', grad_fn=<AddmmBackward>)
  13222. landmarks are: tensor([[[ 0.6058, -0.3892, 1.4208, -1.0927, -0.1843, -1.4237, 0.6154,
  13223. -0.0370],
  13224. [ 0.6074, -0.4223, 1.8192, -0.0303, -0.4152, 0.1236, 0.6524,
  13225. -0.0731],
  13226. [ 0.6014, -0.3883, 1.8192, -0.1150, -0.2882, 0.4008, 0.5201,
  13227. 0.0928],
  13228. [ 0.5477, -0.3851, 1.7961, -0.1304, -0.3055, 0.5085, 0.3830,
  13229. 0.0682],
  13230. [ 0.5759, -0.3978, 1.8192, -0.1689, -0.2074, 0.3392, 0.5490,
  13231. 0.4316],
  13232. [ 0.6174, -0.4201, 1.7309, -0.8784, -0.4735, -0.9524, 0.6242,
  13233. 0.1931],
  13234. [ 0.5466, -0.4706, 1.7198, -0.0903, -0.5712, 0.1261, 0.4733,
  13235. 0.0688],
  13236. [ 0.5219, -0.4550, 1.1415, -0.9196, -0.6404, -0.9387, 0.1856,
  13237. 0.0141]]], device='cuda:0')
  13238. loss_train_step before backward: tensor(0.0277, device='cuda:0', grad_fn=<MseLossBackward>)
  13239. loss_train_step after backward: tensor(0.0277, device='cuda:0', grad_fn=<MseLossBackward>)
  13240. loss_train: 1.6211165934801102
  13241. step: 36
  13242. running loss: 0.045031016485558614
  13243. Train Steps: 36/90 Loss: 0.0450 torch.Size([8, 600, 800])
  13244. torch.Size([8, 8])
  13245. tensor([[0.6151, 0.4085, 0.8750, 0.4367, 0.3887, 0.4367, 0.5066, 0.5846],
  13246. [0.6185, 0.4042, 0.7700, 0.2250, 0.4062, 0.2117, 0.5763, 0.5150],
  13247. [0.6161, 0.4076, 0.8900, 0.4667, 0.4125, 0.5917, 0.6262, 0.5367],
  13248. [0.6259, 0.4156, 0.8812, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960],
  13249. [0.6286, 0.4086, 0.8408, 0.2801, 0.4163, 0.2800, 0.6725, 0.5393],
  13250. [0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133],
  13251. [0.6241, 0.4143, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550],
  13252. [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320]],
  13253. device='cuda:0', dtype=torch.float64)
  13254. predictions are: tensor([[ 0.3947, -0.4847, 1.5870, -0.3033, -0.5490, -0.3614, 0.0847, 0.3221],
  13255. [ 0.1355, -0.6723, 1.1828, -1.0753, -0.4606, -1.4371, 0.2556, 0.1990],
  13256. [ 0.5287, -0.4141, 1.7355, -0.2123, -0.3593, 0.2886, 0.3807, 0.2076],
  13257. [ 0.3074, -0.5518, 1.6790, -0.8450, -0.1968, -1.2325, 0.5182, 0.1985],
  13258. [ 0.5691, -0.4298, 1.5528, -0.9136, -0.3392, -1.1796, 0.6582, 0.2453],
  13259. [ 0.5131, -0.4907, 1.6979, 0.1381, -0.5536, -0.2265, 0.5239, 0.1741],
  13260. [ 0.5242, -0.4441, 1.7725, -0.1746, -0.4777, 0.0534, 0.4184, 0.2524],
  13261. [ 0.4037, -0.5380, 1.6122, -0.0100, -0.4977, -0.0729, 0.4697, 0.2294]],
  13262. device='cuda:0', grad_fn=<AddmmBackward>)
  13263. landmarks are: tensor([[[ 0.5553, -0.3992, 1.7557, -0.2690, -0.4903, -0.2690, 0.0542,
  13264. 0.4145],
  13265. [ 0.5707, -0.4189, 1.2707, -1.2467, -0.4095, -1.3082, 0.3758,
  13266. 0.0928],
  13267. [ 0.5598, -0.4032, 1.8249, -0.1304, -0.3806, 0.4470, 0.6067,
  13268. 0.1929],
  13269. [ 0.6052, -0.3663, 1.7845, -0.8156, -0.0804, -1.4237, 0.5866,
  13270. 0.0051],
  13271. [ 0.6174, -0.3984, 1.5975, -0.9921, -0.3633, -0.9926, 0.8205,
  13272. 0.2050],
  13273. [ 0.6148, -0.4130, 1.6864, 0.1698, -0.5307, -0.1150, 0.6125,
  13274. 0.0851],
  13275. [ 0.5968, -0.3725, 1.8423, -0.1381, -0.4037, 0.1852, 0.6009,
  13276. 0.2776],
  13277. [ 0.6273, -0.4393, 1.6402, 0.1313, -0.5076, 0.0467, 1.1532,
  13278. 0.1715]]], device='cuda:0')
  13279. loss_train_step before backward: tensor(0.0234, device='cuda:0', grad_fn=<MseLossBackward>)
  13280. loss_train_step after backward: tensor(0.0234, device='cuda:0', grad_fn=<MseLossBackward>)
  13281. loss_train: 1.6445499509572983
  13282. step: 37
  13283. running loss: 0.04444729597181887
  13284. Train Steps: 37/90 Loss: 0.0444 torch.Size([8, 600, 800])
  13285. torch.Size([8, 8])
  13286. tensor([[0.6222, 0.3957, 0.8838, 0.5017, 0.3937, 0.4600, 0.5900, 0.5017],
  13287. [0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869],
  13288. [0.6100, 0.4071, 0.7601, 0.3444, 0.3400, 0.4117, 0.5625, 0.5617],
  13289. [0.6030, 0.3969, 0.7988, 0.3917, 0.3450, 0.3667, 0.5266, 0.4700],
  13290. [0.6321, 0.4048, 0.8738, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927],
  13291. [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374],
  13292. [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284],
  13293. [0.6100, 0.4016, 0.8600, 0.5067, 0.4612, 0.5233, 0.5086, 0.5519]],
  13294. device='cuda:0', dtype=torch.float64)
  13295. predictions are: tensor([[ 0.1872, -0.6722, 1.8496, -0.0055, -0.4651, -0.0681, 0.4674, 0.1633],
  13296. [ 0.4701, -0.4734, 1.5983, -1.1720, -0.0912, -1.0880, 0.6450, 0.1887],
  13297. [ 0.2988, -0.5591, 1.3431, -0.8010, -0.6583, -0.3453, 0.2976, 0.2991],
  13298. [ 0.4385, -0.4926, 1.5840, -0.5594, -0.6529, -0.5741, 0.2479, 0.1788],
  13299. [ 0.5675, -0.4378, 1.8293, 0.2816, -0.5209, -0.2343, 0.4806, 0.1298],
  13300. [ 0.4914, -0.4685, 1.6776, -0.5615, -0.5379, -0.7593, 0.4647, 0.1668],
  13301. [ 0.6671, -0.3534, 1.2841, -1.0860, -0.1890, -1.3536, 0.4186, 0.2550],
  13302. [ 0.5687, -0.3997, 1.7216, 0.1144, -0.1886, 0.1014, 0.2823, 0.2538]],
  13303. device='cuda:0', grad_fn=<AddmmBackward>)
  13304. landmarks are: tensor([[[ 0.5879, -0.4583, 1.7961, 0.0313, -0.4672, -0.1612, 0.4393,
  13305. 0.0313],
  13306. [ 0.6127, -0.3944, 1.5189, -1.2467, -0.1323, -1.4622, 0.5646,
  13307. -0.0369],
  13308. [ 0.5318, -0.4056, 1.2249, -0.6949, -0.7155, -0.3844, 0.3122,
  13309. 0.3084],
  13310. [ 0.4992, -0.4525, 1.4035, -0.4768, -0.6924, -0.5923, 0.1465,
  13311. -0.1151],
  13312. [ 0.6335, -0.4162, 1.7499, 0.3084, -0.4961, -0.2459, 0.6524,
  13313. -0.0102],
  13314. [ 0.5777, -0.4416, 1.7044, -0.5827, -0.5962, -0.8361, 0.4862,
  13315. 0.1963],
  13316. [ 0.5761, -0.4070, 1.2452, -1.2541, -0.1725, -1.4835, 0.4511,
  13317. 0.1545],
  13318. [ 0.5318, -0.4310, 1.6864, 0.0543, -0.1554, 0.1313, 0.0635,
  13319. 0.2634]]], device='cuda:0')
  13320. loss_train_step before backward: tensor(0.0146, device='cuda:0', grad_fn=<MseLossBackward>)
  13321.  
  13322. loss_train_step after backward: tensor(0.0146, device='cuda:0', grad_fn=<MseLossBackward>)
  13323. loss_train: 1.6591525804251432
  13324. step: 38
  13325. running loss: 0.04366191001118798
  13326. Train Steps: 38/90 Loss: 0.0437 torch.Size([8, 600, 800])
  13327. torch.Size([8, 8])
  13328. tensor([[0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124],
  13329. [0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433],
  13330. [0.6135, 0.4115, 0.8838, 0.4667, 0.4288, 0.6050, 0.5778, 0.5097],
  13331. [0.6112, 0.4029, 0.8638, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567],
  13332. [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6088, 0.5583],
  13333. [ nan, nan, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733],
  13334. [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466],
  13335. [0.6201, 0.4151, 0.8588, 0.5467, 0.3700, 0.3950, 0.5637, 0.5933]],
  13336. device='cuda:0', dtype=torch.float64)
  13337. predictions are: tensor([[ 6.9073e-01, -3.3128e-01, 1.5764e+00, -1.0210e+00, -4.4448e-01,
  13338. -1.1284e+00, 5.1971e-01, 9.0985e-02],
  13339. [ 7.7653e-01, -2.6876e-01, 1.7526e+00, 9.8295e-04, -1.8860e-01,
  13340. -1.1193e-01, 3.4547e-01, 1.6992e-01],
  13341. [ 8.5491e-01, -2.1427e-01, 1.9309e+00, -2.1968e-01, -3.7515e-01,
  13342. 3.5782e-01, 5.1594e-01, 1.0342e-01],
  13343. [ 5.4454e-01, -4.1243e-01, 1.8518e+00, -1.7703e-01, -1.7207e-01,
  13344. -6.9722e-02, 5.0051e-01, 1.7933e-01],
  13345. [ 8.4427e-01, -2.5028e-01, 1.1858e+00, -1.1193e+00, -4.6734e-01,
  13346. -1.2457e+00, 5.1291e-01, 1.8033e-01],
  13347. [-1.3002e+00, -1.5918e+00, 1.1560e+00, -1.2955e+00, -2.7973e-01,
  13348. -1.4764e+00, 3.8677e-01, 2.7004e-01],
  13349. [ 6.1068e-01, -3.7035e-01, 1.7309e+00, -1.0349e-02, -3.6074e-01,
  13350. 1.1573e-01, 3.6811e-01, 1.5947e-01],
  13351. [ 6.7004e-01, -2.9821e-01, 1.7256e+00, 2.0917e-01, -6.2269e-01,
  13352. -4.3491e-01, 3.2126e-01, 2.6025e-01]], device='cuda:0',
  13353. grad_fn=<AddmmBackward>)
  13354. landmarks are: tensor([[[ 0.5790, -0.4079, 1.5929, -1.0630, -0.4729, -1.0725, 0.4137,
  13355. 0.0807],
  13356. [ 0.5295, -0.4373, 1.6553, 0.1011, 0.0380, 0.0671, 0.0813,
  13357. 0.2237],
  13358. [ 0.5477, -0.3851, 1.7961, -0.1304, -0.3055, 0.5085, 0.3830,
  13359. 0.0682],
  13360. [ 0.5371, -0.4252, 1.7037, -0.0688, -0.0342, 0.0620, 0.3758,
  13361. 0.2853],
  13362. [ 0.5992, -0.3968, 0.9307, -1.3497, -0.4730, -1.0465, 0.5259,
  13363. 0.2930],
  13364. [-2.2859, -2.2859, 0.7106, -1.4468, -0.2882, -1.4237, 0.2430,
  13365. 0.3623],
  13366. [ 0.5178, -0.4480, 1.6748, 0.0620, -0.2767, 0.2083, 0.1067,
  13367. 0.2386],
  13368. [ 0.5785, -0.3687, 1.6806, 0.2391, -0.5769, -0.4614, 0.3180,
  13369. 0.4547]]], device='cuda:0')
  13370. loss_train_step before backward: tensor(0.0425, device='cuda:0', grad_fn=<MseLossBackward>)
  13371. loss_train_step after backward: tensor(0.0425, device='cuda:0', grad_fn=<MseLossBackward>)
  13372. loss_train: 1.7016998585313559
  13373. step: 39
  13374. running loss: 0.0436333297059322
  13375. Train Steps: 39/90 Loss: 0.0436 torch.Size([8, 600, 800])
  13376. torch.Size([8, 8])
  13377. tensor([[0.6263, 0.4030, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771],
  13378. [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390],
  13379. [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633],
  13380. [0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148],
  13381. [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103],
  13382. [0.6200, 0.3999, 0.8653, 0.5207, 0.4100, 0.5125, 0.5975, 0.5103],
  13383. [0.6261, 0.4045, 0.8865, 0.5369, 0.3895, 0.4859, 0.6683, 0.5249],
  13384. [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717]],
  13385. device='cuda:0', dtype=torch.float64)
  13386. predictions are: tensor([[ 0.4242, -0.5076, 1.9139, -0.3800, -0.3886, -0.0722, 0.5852, 0.0703],
  13387. [ 0.6892, -0.3367, 1.8315, -0.3075, -0.3468, -0.2121, 0.3945, 0.1449],
  13388. [ 0.5346, -0.4161, 1.7029, -0.1029, -0.4170, -0.3621, 0.4635, 0.3057],
  13389. [ 0.5330, -0.4363, 1.8144, -0.1320, -0.4021, -0.2648, 0.4028, 0.0935],
  13390. [ 0.4335, -0.4818, 1.7706, -0.5227, -0.3796, -0.3209, 0.2563, 0.1315],
  13391. [ 0.5602, -0.4312, 1.8075, -0.1298, -0.2862, -0.0801, 0.4577, 0.1479],
  13392. [ 0.6446, -0.3702, 1.8362, -0.1281, -0.4253, -0.1565, 0.6927, 0.0867],
  13393. [ 0.4386, -0.4931, 1.0271, -1.2705, -0.3864, -1.4639, 0.2277, 0.2080]],
  13394. device='cuda:0', grad_fn=<AddmmBackward>)
  13395. landmarks are: tensor([[[ 0.6072, -0.4247, 1.8711, -0.0842, -0.5307, 0.1005, 0.6771,
  13396. -0.0821],
  13397. [ 0.5250, -0.4661, 1.7383, -0.0765, -0.4268, -0.0226, 0.2535,
  13398. 0.2035],
  13399. [ 0.5879, -0.3591, 1.8018, 0.1236, -0.5596, -0.1612, 0.3469,
  13400. 0.3161],
  13401. [ 0.5773, -0.4316, 1.7399, 0.1287, -0.5153, -0.0817, 0.4313,
  13402. 0.0919],
  13403. [ 0.5253, -0.4392, 1.7730, -0.2305, -0.4268, -0.1381, 0.1651,
  13404. 0.0712],
  13405. [ 0.5778, -0.4389, 1.7107, 0.1192, -0.3921, 0.0815, 0.4741,
  13406. 0.0711],
  13407. [ 0.6059, -0.4177, 1.8087, 0.1941, -0.4868, -0.0414, 0.8010,
  13408. 0.1385],
  13409. [ 0.5633, -0.3883, 0.8648, -1.0288, -0.5192, -1.1081, 0.2776,
  13410. 0.3546]]], device='cuda:0')
  13411. loss_train_step before backward: tensor(0.0207, device='cuda:0', grad_fn=<MseLossBackward>)
  13412. loss_train_step after backward: tensor(0.0207, device='cuda:0', grad_fn=<MseLossBackward>)
  13413. loss_train: 1.7223987337201834
  13414. step: 40
  13415. running loss: 0.04305996834300459
  13416. Train Steps: 40/90 Loss: 0.0431 torch.Size([8, 600, 800])
  13417. torch.Size([8, 8])
  13418. tensor([[0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204],
  13419. [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377],
  13420. [0.6219, 0.3934, 0.8688, 0.5267, 0.4313, 0.4967, 0.5988, 0.4983],
  13421. [0.6200, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148],
  13422. [0.6048, 0.3928, 0.8538, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719],
  13423. [ nan, nan, 0.7725, 0.2611, 0.3675, 0.2733, 0.5413, 0.5167],
  13424. [0.6203, 0.4096, 0.8862, 0.4267, 0.3538, 0.4117, 0.6025, 0.5650],
  13425. [0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343]],
  13426. device='cuda:0', dtype=torch.float64)
  13427. predictions are: tensor([[ 9.7662e-01, -1.1745e-01, 1.6946e+00, -1.8736e-02, -2.7298e-01,
  13428. 1.1961e-01, 6.4769e-01, 1.2871e-01],
  13429. [ 8.3737e-01, -2.2624e-01, 1.7305e+00, -1.1073e-03, -3.4214e-01,
  13430. -1.4048e-01, 4.8221e-01, 1.4542e-01],
  13431. [ 6.4155e-01, -3.4631e-01, 1.7510e+00, -1.0764e-01, -2.6191e-01,
  13432. -6.1281e-02, 3.9667e-01, 7.9823e-02],
  13433. [ 7.0425e-01, -2.9291e-01, 1.8040e+00, -2.7007e-01, -4.8139e-01,
  13434. -3.5236e-01, 3.2093e-01, 1.1178e-01],
  13435. [ 8.2886e-01, -2.0781e-01, 1.6514e+00, 6.9484e-02, -3.7936e-01,
  13436. -2.4129e-02, 2.7950e-01, 1.7206e-01],
  13437. [-1.6621e+00, -1.8554e+00, 1.2965e+00, -1.3455e+00, -3.9351e-01,
  13438. -1.0603e+00, 1.3691e-01, 1.4077e-01],
  13439. [ 7.7488e-01, -2.3785e-01, 1.8051e+00, -5.3855e-01, -4.5368e-01,
  13440. -4.5824e-01, 4.3108e-01, 2.1553e-01],
  13441. [ 8.9303e-01, -1.7966e-01, 1.6875e+00, -8.4745e-01, -2.5337e-01,
  13442. -1.0233e+00, 7.5601e-01, 7.6916e-02]], device='cuda:0',
  13443. grad_fn=<AddmmBackward>)
  13444. landmarks are: tensor([[[ 0.6042, -0.4273, 1.7198, 0.2184, -0.3478, 0.1149, 0.8062,
  13445. 0.1176],
  13446. [ 0.5775, -0.4607, 1.6741, 0.1962, -0.4036, -0.1212, 0.4588,
  13447. 0.1979],
  13448. [ 0.5864, -0.4690, 1.7268, 0.1467, -0.2940, 0.0082, 0.4797,
  13449. 0.0159],
  13450. [ 0.5781, -0.4205, 1.8159, -0.0691, -0.6116, -0.3069, 0.4236,
  13451. 0.0919],
  13452. [ 0.5079, -0.4714, 1.6575, 0.2237, -0.4961, 0.0774, 0.1465,
  13453. -0.1061],
  13454. [-2.2859, -2.2859, 1.2820, -1.0801, -0.5885, -1.0234, 0.2141,
  13455. 0.1005],
  13456. [ 0.5790, -0.3940, 1.8076, -0.3152, -0.6520, -0.3844, 0.4970,
  13457. 0.3238],
  13458. [ 0.6125, -0.4369, 1.9173, -0.5384, -0.2594, -0.8386, 0.9741,
  13459. 0.1821]]], device='cuda:0')
  13460. loss_train_step before backward: tensor(0.0354, device='cuda:0', grad_fn=<MseLossBackward>)
  13461. loss_train_step after backward: tensor(0.0354, device='cuda:0', grad_fn=<MseLossBackward>)
  13462. loss_train: 1.757778873667121
  13463. step: 41
  13464. running loss: 0.04287265545529563
  13465.  
  13466. Train Steps: 41/90 Loss: 0.0429 torch.Size([8, 600, 800])
  13467. torch.Size([8, 8])
  13468. tensor([[0.6125, 0.4076, 0.8488, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505],
  13469. [0.6115, 0.3998, 0.7063, 0.2383, 0.4037, 0.1950, 0.5320, 0.4993],
  13470. [0.6200, 0.4059, 0.8700, 0.4900, 0.4163, 0.5000, 0.6162, 0.5467],
  13471. [0.6243, 0.4128, 0.7762, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417],
  13472. [0.6200, 0.4118, 0.8287, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799],
  13473. [0.6076, 0.3953, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954],
  13474. [0.6168, 0.4029, 0.8523, 0.3417, 0.3588, 0.5000, 0.6125, 0.5400],
  13475. [0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131]],
  13476. device='cuda:0', dtype=torch.float64)
  13477. predictions are: tensor([[ 0.5364, -0.4022, 1.7602, -0.2904, -0.5344, -0.4302, 0.1283, 0.1212],
  13478. [ 0.6020, -0.3860, 1.1146, -1.0672, -0.2285, -1.3674, 0.2108, 0.0569],
  13479. [ 0.5184, -0.4487, 1.8782, 0.1548, -0.1674, 0.3226, 0.6550, 0.1418],
  13480. [ 0.6975, -0.3259, 1.4950, -0.8409, -0.2680, -0.8396, 0.5728, 0.1666],
  13481. [ 0.5019, -0.4184, 1.5995, -0.2894, -0.3780, -0.9116, 0.2311, 0.1745],
  13482. [ 0.4203, -0.4804, 1.6434, -0.4923, -0.5403, -0.3176, 0.3912, 0.0179],
  13483. [ 0.0798, -0.6898, 1.6994, -0.5779, -0.3896, 0.1759, 0.8041, 0.1626],
  13484. [ 0.5689, -0.4070, 1.7062, -0.3296, -0.4752, -0.0741, 0.4950, 0.0697]],
  13485. device='cuda:0', grad_fn=<AddmmBackward>)
  13486. landmarks are: tensor([[[ 0.5433, -0.4032, 1.6344, -0.4922, -0.5769, -0.5846, 0.0357,
  13487. 0.2567],
  13488. [ 0.5384, -0.4393, 0.9762, -1.1851, -0.4210, -1.3852, 0.1712,
  13489. 0.0201],
  13490. [ 0.5776, -0.4109, 1.7326, -0.0226, -0.3633, 0.0236, 0.5605,
  13491. 0.2391],
  13492. [ 0.5977, -0.3792, 1.2995, -1.0311, -0.5192, -0.8386, 0.5836,
  13493. 0.2160],
  13494. [ 0.5779, -0.3840, 1.5420, -0.4306, -0.5423, -0.9772, 0.2041,
  13495. 0.3928],
  13496. [ 0.5206, -0.4603, 1.4670, -0.4768, -0.6577, -0.3998, 0.1836,
  13497. 0.0021],
  13498. [ 0.5631, -0.4249, 1.6509, -0.7078, -0.6289, 0.0236, 0.5432,
  13499. 0.2083],
  13500. [ 0.5581, -0.4586, 1.5586, -0.3747, -0.6792, -0.2391, 0.4455,
  13501. 0.0840]]], device='cuda:0')
  13502. loss_train_step before backward: tensor(0.0208, device='cuda:0', grad_fn=<MseLossBackward>)
  13503. loss_train_step after backward: tensor(0.0208, device='cuda:0', grad_fn=<MseLossBackward>)
  13504. loss_train: 1.778610149398446
  13505. step: 42
  13506. running loss: 0.042347860699963005
  13507. Train Steps: 42/90 Loss: 0.0423 torch.Size([8, 600, 800])
  13508. torch.Size([8, 8])
  13509. tensor([[0.6161, 0.4099, 0.8738, 0.4383, 0.3788, 0.5483, 0.5605, 0.5019],
  13510. [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285],
  13511. [0.6250, 0.4236, 0.8638, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799],
  13512. [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578],
  13513. [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208],
  13514. [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5787, 0.5117],
  13515. [0.6264, 0.3972, 0.8853, 0.4771, 0.3853, 0.4511, 0.6293, 0.5334],
  13516. [0.6179, 0.4008, 0.8600, 0.4015, 0.3932, 0.2515, 0.5711, 0.5438]],
  13517. device='cuda:0', dtype=torch.float64)
  13518. predictions are: tensor([[ 0.3502, -0.5350, 1.8028, -0.1968, -0.4146, 0.5732, 0.5760, 0.0674],
  13519. [ 0.7001, -0.2969, 1.4979, -0.3223, -0.5520, -0.0171, 0.1787, 0.0619],
  13520. [ 0.5623, -0.3918, 1.8224, -0.2495, -0.3759, -0.4400, 0.4877, 0.2254],
  13521. [ 0.4395, -0.4990, 1.0913, -1.0361, -0.3294, -1.1480, 0.3250, 0.1471],
  13522. [ 0.3808, -0.5393, 1.5022, -0.8524, -0.3207, -0.9561, 0.5060, 0.0491],
  13523. [ 0.5369, -0.4363, 1.2601, -0.9757, -0.3010, -1.1186, 0.2567, 0.0580],
  13524. [ 0.4902, -0.4967, 1.9258, 0.1589, -0.4215, 0.2360, 0.8296, 0.0872],
  13525. [ 0.4893, -0.4287, 1.6650, -0.4043, -0.3079, -0.8211, 0.2554, 0.1332]],
  13526. device='cuda:0', grad_fn=<AddmmBackward>)
  13527. landmarks are: tensor([[[ 0.5597, -0.3928, 1.7499, -0.2613, -0.5365, 0.2468, 0.3028,
  13528. 0.0321],
  13529. [ 0.5253, -0.4296, 1.3084, -0.3098, -0.6115, -0.2767, 0.0928,
  13530. 0.1552],
  13531. [ 0.6008, -0.3293, 1.7037, -0.5461, -0.4152, -0.8309, 0.3234,
  13532. 0.3928],
  13533. [ 0.5232, -0.4189, 0.9358, -1.4006, -0.3921, -1.3698, 0.2555,
  13534. 0.2906],
  13535. [ 0.5885, -0.4429, 1.4266, -0.9926, -0.4383, -1.2313, 0.4228,
  13536. 0.1195],
  13537. [ 0.5598, -0.4201, 1.1898, -1.3005, -0.3806, -1.3313, 0.3873,
  13538. 0.0774],
  13539. [ 0.6075, -0.4514, 1.8032, -0.0822, -0.5061, -0.2023, 0.6208,
  13540. 0.1779],
  13541. [ 0.5680, -0.4345, 1.6864, -0.4315, -0.4698, -1.1241, 0.3518,
  13542. 0.2261]]], device='cuda:0')
  13543. loss_train_step before backward: tensor(0.0289, device='cuda:0', grad_fn=<MseLossBackward>)
  13544. loss_train_step after backward: tensor(0.0289, device='cuda:0', grad_fn=<MseLossBackward>)
  13545. loss_train: 1.8074938002973795
  13546. step: 43
  13547. running loss: 0.04203473954179952
  13548. Train Steps: 43/90 Loss: 0.0420 torch.Size([8, 600, 800])
  13549. torch.Size([8, 8])
  13550. tensor([[0.6038, 0.3946, 0.8413, 0.4883, 0.3563, 0.4550, 0.5266, 0.4693],
  13551. [0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869],
  13552. [ nan, nan, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621],
  13553. [0.6200, 0.3999, 0.8653, 0.5207, 0.4100, 0.5125, 0.5975, 0.5103],
  13554. [0.6289, 0.4032, 0.8419, 0.5446, 0.4075, 0.5017, 0.6312, 0.5117],
  13555. [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6038, 0.4833],
  13556. [0.6034, 0.4011, 0.7350, 0.2533, 0.3438, 0.3367, 0.5516, 0.5084],
  13557. [0.6100, 0.4016, 0.8600, 0.5067, 0.4612, 0.5233, 0.5086, 0.5519]],
  13558. device='cuda:0', dtype=torch.float64)
  13559. predictions are: tensor([[ 0.7905, -0.2252, 1.6335, -0.0263, -0.6232, -0.1224, 0.1979, 0.1167],
  13560. [ 0.7494, -0.2658, 1.4417, -1.1422, -0.2417, -1.3671, 0.5485, 0.0467],
  13561. [-1.5169, -1.7478, 1.5878, -1.2802, -0.0359, -1.0986, 0.6142, 0.1493],
  13562. [ 0.7136, -0.3303, 1.8095, 0.1998, -0.4240, 0.2751, 0.4201, 0.1101],
  13563. [ 0.9599, -0.1700, 1.7080, 0.1720, -0.4273, 0.1289, 0.6565, 0.1152],
  13564. [ 0.6922, -0.3478, 1.7639, -0.0507, -0.5437, 0.0132, 0.4204, 0.0642],
  13565. [ 0.7462, -0.2894, 1.2072, -1.1087, -0.6579, -0.7699, 0.3299, 0.0590],
  13566. [ 0.7810, -0.2467, 1.7174, 0.1712, -0.2060, 0.1815, 0.2094, 0.1444]],
  13567. device='cuda:0', grad_fn=<AddmmBackward>)
  13568. landmarks are: tensor([[[ 0.5030, -0.4631, 1.5998, -0.0303, -0.6404, -0.1843, 0.1465,
  13569. -0.1181],
  13570. [ 0.6058, -0.3892, 1.4208, -1.0927, -0.1843, -1.4237, 0.6154,
  13571. -0.0370],
  13572. [-2.2859, -2.2859, 1.7210, -0.9772, 0.1852, -1.3698, 0.7986,
  13573. 0.3104],
  13574. [ 0.5778, -0.4389, 1.7107, 0.1192, -0.3921, 0.0815, 0.4741,
  13575. 0.0711],
  13576. [ 0.6189, -0.4238, 1.6026, 0.2295, -0.4037, 0.0313, 0.6298,
  13577. 0.0774],
  13578. [ 0.5726, -0.4795, 1.7788, -0.0149, -0.5711, -0.0688, 0.5028,
  13579. -0.0534],
  13580. [ 0.5009, -0.4333, 1.1090, -1.1158, -0.6982, -0.7309, 0.2617,
  13581. 0.0622],
  13582. [ 0.5318, -0.4310, 1.6864, 0.0543, -0.1554, 0.1313, 0.0635,
  13583. 0.2634]]], device='cuda:0')
  13584. loss_train_step before backward: tensor(0.0319, device='cuda:0', grad_fn=<MseLossBackward>)
  13585. loss_train_step after backward: tensor(0.0319, device='cuda:0', grad_fn=<MseLossBackward>)
  13586. loss_train: 1.8393607195466757
  13587. step: 44
  13588. running loss: 0.041803652716969904
  13589. Train Steps: 44/90 Loss: 0.0418 torch.Size([8, 600, 800])
  13590. torch.Size([8, 8])
  13591. tensor([[0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274],
  13592. [0.6038, 0.3946, 0.8413, 0.4883, 0.3563, 0.4550, 0.5266, 0.4693],
  13593. [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167],
  13594. [0.6250, 0.4131, 0.8688, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367],
  13595. [0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148],
  13596. [0.6246, 0.4008, 0.8757, 0.5088, 0.4101, 0.5392, 0.6644, 0.5133],
  13597. [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611],
  13598. [0.6102, 0.4005, 0.8688, 0.5100, 0.4813, 0.5400, 0.5404, 0.5064]],
  13599. device='cuda:0', dtype=torch.float64)
  13600. predictions are: tensor([[ 0.4716, -0.4719, 1.7441, -0.2225, -0.5760, -0.4219, 0.7062, 0.1232],
  13601. [ 0.5111, -0.4207, 1.4883, -0.2494, -0.6212, -0.3269, 0.2111, 0.1063],
  13602. [ 0.5906, -0.4036, 1.6814, -0.0946, -0.4266, -0.0143, 0.4353, 0.0676],
  13603. [ 0.7047, -0.3238, 1.6344, -1.1320, -0.3182, -1.2865, 0.5688, 0.0724],
  13604. [ 0.4475, -0.5090, 1.6610, -0.0321, -0.5360, -0.1445, 0.3971, 0.0563],
  13605. [ 0.3423, -0.5918, 1.6331, -0.1741, -0.3868, 0.2066, 0.6619, 0.1233],
  13606. [ 0.3358, -0.5321, 1.5918, -0.2288, -0.3002, -0.1307, 0.2582, 0.1861],
  13607. [ 0.6290, -0.3624, 1.6425, -0.1294, -0.1318, -0.0388, 0.1791, 0.0794]],
  13608. device='cuda:0', grad_fn=<AddmmBackward>)
  13609. landmarks are: tensor([[[ 0.6108, -0.4008, 1.9088, -0.0253, -0.5769, -0.3075, 0.7905,
  13610. 0.1499],
  13611. [ 0.5030, -0.4631, 1.5998, -0.0303, -0.6404, -0.1843, 0.1465,
  13612. -0.1181],
  13613. [ 0.5828, -0.4066, 1.7557, 0.0774, -0.3691, -0.0226, 0.4277,
  13614. 0.1005],
  13615. [ 0.6011, -0.3778, 1.7268, -0.9079, -0.3113, -1.1928, 0.5605,
  13616. 0.1929],
  13617. [ 0.5773, -0.4316, 1.7399, 0.1287, -0.5153, -0.0817, 0.4313,
  13618. 0.0919],
  13619. [ 0.5993, -0.4345, 1.7587, 0.0641, -0.3917, 0.2048, 0.7827,
  13620. 0.0852],
  13621. [ 0.5491, -0.4290, 1.7788, -0.1073, -0.2651, -0.1073, 0.2555,
  13622. 0.3057],
  13623. [ 0.5328, -0.4361, 1.7268, 0.0697, -0.0630, 0.2083, 0.2103,
  13624. 0.0532]]], device='cuda:0')
  13625. loss_train_step before backward: tensor(0.0137, device='cuda:0', grad_fn=<MseLossBackward>)
  13626.  
  13627. loss_train_step after backward: tensor(0.0137, device='cuda:0', grad_fn=<MseLossBackward>)
  13628. loss_train: 1.8530531292781234
  13629. step: 45
  13630. running loss: 0.04117895842840274
  13631. Train Steps: 45/90 Loss: 0.0412 torch.Size([8, 600, 800])
  13632. torch.Size([8, 8])
  13633. tensor([[ nan, nan, 0.7625, 0.2433, 0.3713, 0.2867, 0.5235, 0.5220],
  13634. [0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5787, 0.5600],
  13635. [0.6193, 0.3930, 0.8949, 0.4437, 0.3852, 0.5435, 0.6263, 0.5263],
  13636. [0.6248, 0.4032, 0.7738, 0.1900, 0.4813, 0.1400, 0.5941, 0.4904],
  13637. [0.6200, 0.4049, 0.8638, 0.5617, 0.4125, 0.5100, 0.6013, 0.5317],
  13638. [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726],
  13639. [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5413, 0.5433],
  13640. [0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433]],
  13641. device='cuda:0', dtype=torch.float64)
  13642. predictions are: tensor([[-2.0062, -2.1088, 1.2378, -1.1293, -0.6208, -0.9271, 0.1880, 0.1063],
  13643. [ 0.6630, -0.3410, 1.7287, 0.0418, -0.1805, 0.1816, 0.3918, 0.1792],
  13644. [ 0.6831, -0.3443, 1.8144, -0.0936, -0.5714, 0.2910, 0.6572, 0.0492],
  13645. [ 0.7003, -0.3176, 1.2593, -1.1093, -0.1448, -1.5150, 0.5180, 0.0846],
  13646. [ 0.9637, -0.1827, 1.7292, 0.3907, -0.4397, 0.2822, 0.5706, 0.0913],
  13647. [ 1.0779, -0.0850, 1.2507, -0.9365, -0.5781, -0.9499, 0.3741, 0.0290],
  13648. [ 0.6943, -0.2916, 1.5082, -0.5144, -0.6949, -0.6832, 0.2075, 0.1789],
  13649. [ 0.9264, -0.1761, 1.7016, 0.2059, -0.1610, 0.0985, 0.3311, 0.1391]],
  13650. device='cuda:0', grad_fn=<AddmmBackward>)
  13651. landmarks are: tensor([[[-2.2859, -2.2859, 1.2360, -1.1620, -0.5711, -0.9618, 0.1322,
  13652. 0.1253],
  13653. [ 0.5505, -0.4207, 1.7095, -0.0534, -0.0509, 0.1050, 0.3873,
  13654. 0.3007],
  13655. [ 0.5746, -0.4706, 1.8476, -0.2365, -0.5068, 0.2245, 0.6069,
  13656. 0.1449],
  13657. [ 0.5999, -0.4236, 1.2880, -1.4083, -0.0630, -1.6393, 0.4584,
  13658. -0.0208],
  13659. [ 0.5776, -0.4159, 1.7037, 0.3084, -0.3806, 0.0697, 0.4912,
  13660. 0.1698],
  13661. [ 0.4997, -0.4446, 1.1610, -0.9772, -0.6058, -1.0311, 0.1404,
  13662. -0.1031],
  13663. [ 0.5894, -0.3479, 1.7730, -0.6847, -0.5538, -0.7155, 0.2141,
  13664. 0.2237],
  13665. [ 0.5295, -0.4373, 1.6553, 0.1011, 0.0380, 0.0671, 0.0813,
  13666. 0.2237]]], device='cuda:0')
  13667. loss_train_step before backward: tensor(0.0267, device='cuda:0', grad_fn=<MseLossBackward>)
  13668. loss_train_step after backward: tensor(0.0267, device='cuda:0', grad_fn=<MseLossBackward>)
  13669. loss_train: 1.8797959005460143
  13670. step: 46
  13671. running loss: 0.04086512827273944
  13672. Train Steps: 46/90 Loss: 0.0409 torch.Size([8, 600, 800])
  13673. torch.Size([8, 8])
  13674. tensor([[0.6137, 0.4038, 0.8563, 0.4050, 0.3813, 0.2550, 0.5106, 0.4954],
  13675. [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083],
  13676. [0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689],
  13677. [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333],
  13678. [0.6131, 0.4064, 0.8638, 0.5200, 0.4788, 0.4783, 0.5258, 0.5867],
  13679. [0.6199, 0.4071, 0.7600, 0.2117, 0.4037, 0.2767, 0.6138, 0.5550],
  13680. [0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204],
  13681. [0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967]],
  13682. device='cuda:0', dtype=torch.float64)
  13683. predictions are: tensor([[ 0.4136, -0.5211, 1.6518, -0.5711, -0.4877, -1.1473, 0.1868, 0.0327],
  13684. [ 0.0622, -0.7807, 1.6220, 0.0317, -0.5489, -0.1672, 0.4488, 0.2454],
  13685. [ 0.1679, -0.6337, 1.3389, -0.7607, -0.6656, -0.5568, 0.2576, 0.1309],
  13686. [ 0.6829, -0.3497, 1.8155, -0.0984, -0.4948, 0.0420, 0.4061, 0.0627],
  13687. [ 0.4903, -0.4681, 1.6231, 0.1573, -0.1295, -0.0120, 0.1925, 0.1658],
  13688. [ 0.7352, -0.3283, 1.3336, -1.2855, -0.3736, -1.1986, 0.5552, 0.1144],
  13689. [ 0.6767, -0.3760, 1.6842, 0.2048, -0.3348, 0.2051, 0.7426, 0.0626],
  13690. [ 0.5668, -0.4108, 1.6528, -0.3251, -0.5824, 0.0828, 0.4796, 0.0241]],
  13691. device='cuda:0', grad_fn=<AddmmBackward>)
  13692. landmarks are: tensor([[[ 0.5485, -0.4209, 1.6691, -0.4152, -0.5249, -1.1081, 0.0725,
  13693. 0.0021],
  13694. [ 0.5713, -0.4105, 1.7557, 0.0467, -0.6520, -0.2690, 0.3988,
  13695. 0.5239],
  13696. [ 0.5327, -0.4381, 1.2880, -0.6308, -0.6866, -0.5307, 0.2658,
  13697. 0.3417],
  13698. [ 0.6003, -0.3747, 1.8885, -0.1997, -0.5018, -0.0149, 0.5490,
  13699. 0.1775],
  13700. [ 0.5462, -0.4090, 1.7037, 0.1159, -0.0746, -0.0765, 0.1427,
  13701. 0.4239],
  13702. [ 0.5775, -0.4054, 1.2245, -1.3082, -0.4210, -1.0080, 0.5490,
  13703. 0.2776],
  13704. [ 0.6042, -0.4273, 1.7198, 0.2184, -0.3478, 0.1149, 0.8062,
  13705. 0.1176],
  13706. [ 0.4974, -0.4482, 1.6633, -0.3306, -0.6173, 0.1313, 0.2925,
  13707. 0.0081]]], device='cuda:0')
  13708. loss_train_step before backward: tensor(0.0180, device='cuda:0', grad_fn=<MseLossBackward>)
  13709. loss_train_step after backward: tensor(0.0180, device='cuda:0', grad_fn=<MseLossBackward>)
  13710. loss_train: 1.8977786405012012
  13711. step: 47
  13712. running loss: 0.040378268946834064
  13713. Train Steps: 47/90 Loss: 0.0404 torch.Size([8, 600, 800])
  13714. torch.Size([8, 8])
  13715. tensor([[0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200],
  13716. [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083],
  13717. [0.6109, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567],
  13718. [0.6128, 0.4084, 0.8738, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397],
  13719. [0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742],
  13720. [0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783],
  13721. [0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283],
  13722. [0.6218, 0.4137, 0.7263, 0.2233, 0.4075, 0.2650, 0.6212, 0.5783]],
  13723. device='cuda:0', dtype=torch.float64)
  13724. predictions are: tensor([[ 0.6224, -0.3776, 1.7414, -0.8322, -0.2218, -0.8424, 0.5022, 0.0806],
  13725. [ 0.4926, -0.5226, 1.6715, 0.2126, -0.5110, 0.1479, 0.4177, 0.2301],
  13726. [ 0.5360, -0.4621, 1.0516, -1.1743, -0.4601, -0.9034, 0.2978, 0.1392],
  13727. [ 0.8273, -0.2386, 1.7271, 0.1335, -0.6301, -0.1857, 0.1177, 0.0842],
  13728. [ 1.0371, -0.1818, 1.8480, 0.4197, -0.5808, -0.2250, 0.4951, 0.0019],
  13729. [-1.5094, -1.7850, 1.0574, -1.0092, -0.3379, -0.9945, 0.2675, 0.1895],
  13730. [ 0.6967, -0.3592, 1.9175, -0.3306, -0.5247, -0.1906, 0.6084, 0.0473],
  13731. [ 0.7549, -0.3261, 1.1601, -0.8692, -0.4138, -0.8388, 0.4756, 0.2019]],
  13732. device='cuda:0', grad_fn=<AddmmBackward>)
  13733. landmarks are: tensor([[[ 0.6089, -0.3948, 1.7383, -0.8617, -0.2536, -1.2390, 0.6009,
  13734. 0.1159],
  13735. [ 0.5713, -0.4105, 1.7557, 0.0467, -0.6520, -0.2690, 0.3988,
  13736. 0.5239],
  13737. [ 0.5356, -0.4217, 1.0339, -1.4776, -0.5076, -1.1081, 0.4219,
  13738. 0.2853],
  13739. [ 0.5443, -0.3994, 1.7499, -0.1227, -0.6173, -0.5769, 0.0049,
  13740. 0.2071],
  13741. [ 0.6113, -0.4057, 1.7499, 0.3007, -0.5885, -0.5384, 0.4513,
  13742. -0.0957],
  13743. [ 0.5742, -0.4474, 0.9834, -1.0159, -0.3229, -1.3159, 0.2314,
  13744. 0.3854],
  13745. [ 0.6073, -0.4058, 1.8885, -0.4999, -0.5942, -0.4768, 0.6413,
  13746. 0.1544],
  13747. [ 0.5863, -0.3749, 1.0686, -1.2543, -0.4037, -1.0619, 0.5836,
  13748. 0.3854]]], device='cuda:0')
  13749. loss_train_step before backward: tensor(0.1265, device='cuda:0', grad_fn=<MseLossBackward>)
  13750. loss_train_step after backward: tensor(0.1265, device='cuda:0', grad_fn=<MseLossBackward>)
  13751. loss_train: 2.0243153674528003
  13752. step: 48
  13753. running loss: 0.04217323682193334
  13754. Train Steps: 48/90 Loss: 0.0422 torch.Size([8, 600, 800])
  13755. torch.Size([8, 8])
  13756. tensor([[0.6202, 0.4079, 0.8025, 0.2500, 0.3763, 0.3217, 0.6125, 0.5533],
  13757. [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364],
  13758. [0.6202, 0.4066, 0.8746, 0.3376, 0.3717, 0.3090, 0.5842, 0.5165],
  13759. [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5787, 0.5550],
  13760. [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378],
  13761. [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383],
  13762. [0.6078, 0.4033, 0.8019, 0.3055, 0.3450, 0.4200, 0.6025, 0.5550],
  13763. [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334]],
  13764. device='cuda:0', dtype=torch.float64)
  13765. predictions are: tensor([[ 0.5666, -0.4526, 1.4781, -0.7345, -0.4664, -0.7452, 0.4017, 0.1356],
  13766. [ 0.3569, -0.5902, 1.4038, -0.7881, -0.1524, -1.2387, 0.5315, 0.1132],
  13767. [ 0.8075, -0.3212, 1.7756, -0.3874, -0.5380, -0.8027, 0.3158, 0.0359],
  13768. [ 0.4889, -0.5159, 1.4719, -0.5841, -0.6540, -0.5052, 0.1248, 0.1664],
  13769. [ 0.3145, -0.6566, 1.7905, 0.4144, -0.3301, 0.5480, 0.6483, 0.1989],
  13770. [ 0.2731, -0.6563, 1.5606, -0.4530, -0.5923, -0.3571, 0.4636, 0.1154],
  13771. [ 0.4306, -0.5213, 1.4185, -0.5322, -0.6252, -0.2094, 0.3687, 0.1786],
  13772. [ 0.3540, -0.5800, 1.2383, -0.7893, -0.4970, -0.7637, 0.4088, 0.1234]],
  13773. device='cuda:0', grad_fn=<AddmmBackward>)
  13774. landmarks are: tensor([[[ 0.5788, -0.4020, 1.4208, -1.1312, -0.5480, -0.8002, 0.5432,
  13775. 0.2699],
  13776. [ 0.6158, -0.3960, 1.4092, -1.2774, -0.2074, -1.1851, 0.8491,
  13777. 0.1917],
  13778. [ 0.5787, -0.4077, 1.7537, -0.7263, -0.5692, -0.8586, 0.4126,
  13779. 0.1000],
  13780. [ 0.5647, -0.4129, 1.4901, -1.0619, -0.6462, -0.5846, 0.3873,
  13781. 0.2776],
  13782. [ 0.6421, -0.3864, 1.7961, 0.0543, -0.4383, 0.2237, 1.2007,
  13783. 0.1982],
  13784. [ 0.5779, -0.4275, 1.5894, -0.8362, -0.6577, -0.5153, 0.5605,
  13785. 0.2006],
  13786. [ 0.5215, -0.4232, 1.4182, -0.8747, -0.6924, -0.3460, 0.4970,
  13787. 0.2776],
  13788. [ 0.5730, -0.4562, 1.2195, -1.2440, -0.5497, -0.7711, 0.5704,
  13789. 0.1779]]], device='cuda:0')
  13790. loss_train_step before backward: tensor(0.0430, device='cuda:0', grad_fn=<MseLossBackward>)
  13791. loss_train_step after backward: tensor(0.0430, device='cuda:0', grad_fn=<MseLossBackward>)
  13792. loss_train: 2.067306305281818
  13793. step: 49
  13794. running loss: 0.04218992459758812
  13795.  
  13796. Train Steps: 49/90 Loss: 0.0422 torch.Size([8, 600, 800])
  13797. torch.Size([8, 8])
  13798. tensor([[0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5787, 0.5117],
  13799. [0.6188, 0.4099, 0.7400, 0.2433, 0.3962, 0.2750, 0.6162, 0.5467],
  13800. [0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222],
  13801. [0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804],
  13802. [0.6084, 0.3981, 0.8588, 0.5233, 0.4600, 0.5367, 0.5680, 0.5006],
  13803. [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5787, 0.5550],
  13804. [0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124],
  13805. [ nan, nan, 0.6469, 0.1943, 0.4025, 0.2000, 0.5125, 0.5533]],
  13806. device='cuda:0', dtype=torch.float64)
  13807. predictions are: tensor([[ 0.5385, -0.4836, 1.2245, -0.9226, -0.4233, -1.0781, 0.2606, 0.1285],
  13808. [ 0.6310, -0.4203, 1.2385, -0.8954, -0.5206, -0.7738, 0.6003, 0.1909],
  13809. [ 0.7104, -0.3214, 1.5186, -0.3326, -0.2686, -0.7922, 0.3363, 0.3189],
  13810. [ 0.8336, -0.3251, 1.8312, -0.2580, -0.6616, -0.3110, 0.6699, 0.0044],
  13811. [ 0.6117, -0.4478, 1.7861, 0.4758, -0.2638, 0.4468, 0.2286, 0.1295],
  13812. [ 0.6076, -0.4409, 1.5538, -0.6097, -0.7504, -0.3698, 0.2756, 0.2015],
  13813. [ 0.5570, -0.4618, 1.5436, -0.8213, -0.4644, -0.8386, 0.4805, 0.1029],
  13814. [-1.1730, -1.5919, 0.9783, -1.0332, -0.4394, -1.1426, 0.3340, 0.2623]],
  13815. device='cuda:0', grad_fn=<AddmmBackward>)
  13816. landmarks are: tensor([[[ 0.5598, -0.4201, 1.1898, -1.3005, -0.3806, -1.3313, 0.3873,
  13817. 0.0774],
  13818. [ 0.5725, -0.3926, 1.1321, -1.1620, -0.4557, -1.0157, 0.5605,
  13819. 0.2391],
  13820. [ 0.6237, -0.2983, 1.3919, -0.4691, -0.0457, -1.2313, 0.2453,
  13821. 0.5882],
  13822. [ 0.6071, -0.4119, 1.7788, -0.5153, -0.6000, -0.5692, 0.6586,
  13823. -0.0670],
  13824. [ 0.5241, -0.4470, 1.6806, 0.1313, -0.1612, 0.1929, 0.3378,
  13825. 0.0261],
  13826. [ 0.5647, -0.4129, 1.4901, -1.0619, -0.6462, -0.5846, 0.3873,
  13827. 0.2776],
  13828. [ 0.5790, -0.4079, 1.5929, -1.0630, -0.4729, -1.0725, 0.4137,
  13829. 0.0807],
  13830. [-2.2859, -2.2859, 0.7023, -1.3883, -0.4268, -1.3621, 0.0813,
  13831. 0.2699]]], device='cuda:0')
  13832. loss_train_step before backward: tensor(0.0568, device='cuda:0', grad_fn=<MseLossBackward>)
  13833. loss_train_step after backward: tensor(0.0568, device='cuda:0', grad_fn=<MseLossBackward>)
  13834. loss_train: 2.1241164905950427
  13835. step: 50
  13836. running loss: 0.04248232981190085
  13837. Train Steps: 50/90 Loss: 0.0425 torch.Size([8, 600, 800])
  13838. torch.Size([8, 8])
  13839. tensor([[0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733],
  13840. [0.6189, 0.4049, 0.8888, 0.4417, 0.4213, 0.5200, 0.5988, 0.5633],
  13841. [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679],
  13842. [ nan, nan, 0.7515, 0.2708, 0.3987, 0.2267, 0.5162, 0.5567],
  13843. [0.6206, 0.4001, 0.8900, 0.3933, 0.3588, 0.3567, 0.5837, 0.5083],
  13844. [0.6126, 0.4067, 0.8638, 0.5383, 0.4188, 0.4850, 0.5016, 0.5392],
  13845. [0.6198, 0.4115, 0.7762, 0.2717, 0.3713, 0.3200, 0.5837, 0.5683],
  13846. [0.6243, 0.4128, 0.7762, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417]],
  13847. device='cuda:0', dtype=torch.float64)
  13848. predictions are: tensor([[ 0.7524, -0.3048, 1.7285, 0.0559, -0.5789, -0.3614, 0.4482, 0.2964],
  13849. [ 0.6012, -0.4074, 1.7178, -0.2578, -0.3618, 0.1045, 0.5429, 0.2140],
  13850. [ 0.7497, -0.3409, 1.7492, 0.1629, -0.3651, 0.2797, 0.8453, 0.2304],
  13851. [-1.1396, -1.5549, 1.0833, -1.2028, -0.4997, -1.1879, 0.1421, 0.2040],
  13852. [ 0.7176, -0.3714, 1.7701, -0.4835, -0.6730, -0.7595, 0.3634, 0.0367],
  13853. [ 0.5878, -0.4171, 1.6346, 0.3001, -0.3926, -0.0520, 0.1840, 0.1566],
  13854. [ 0.6174, -0.4194, 1.3560, -1.0567, -0.6214, -1.0043, 0.3839, 0.2182],
  13855. [ 0.4703, -0.5073, 1.3387, -1.1426, -0.4838, -1.0649, 0.5185, 0.2052]],
  13856. device='cuda:0', grad_fn=<AddmmBackward>)
  13857. landmarks are: tensor([[[ 0.6121, -0.3138, 1.8423, 0.0082, -0.6462, -0.3075, 0.3469,
  13858. 0.3623],
  13859. [ 0.5726, -0.4159, 1.8192, -0.2459, -0.3402, 0.1159, 0.4797,
  13860. 0.3161],
  13861. [ 0.6257, -0.4249, 1.5998, 0.1236, -0.3806, 0.3084, 0.9887,
  13862. 0.3371],
  13863. [-2.2859, -2.2859, 1.1854, -1.0352, -0.4441, -1.2390, 0.0986,
  13864. 0.2853],
  13865. [ 0.5807, -0.4378, 1.8249, -0.4691, -0.6289, -0.6385, 0.4104,
  13866. 0.0620],
  13867. [ 0.5436, -0.4076, 1.7037, 0.2006, -0.3517, -0.0457, 0.0311,
  13868. 0.2048],
  13869. [ 0.5768, -0.3852, 1.2995, -1.0311, -0.5711, -0.8079, 0.4104,
  13870. 0.3392],
  13871. [ 0.5977, -0.3792, 1.2995, -1.0311, -0.5192, -0.8386, 0.5836,
  13872. 0.2160]]], device='cuda:0')
  13873. loss_train_step before backward: tensor(0.0363, device='cuda:0', grad_fn=<MseLossBackward>)
  13874. loss_train_step after backward: tensor(0.0363, device='cuda:0', grad_fn=<MseLossBackward>)
  13875. loss_train: 2.1603687768802047
  13876. step: 51
  13877. running loss: 0.04236017209569029
  13878. Train Steps: 51/90 Loss: 0.0424 torch.Size([8, 600, 800])
  13879. torch.Size([8, 8])
  13880. tensor([[0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153],
  13881. [0.6075, 0.4000, 0.8513, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280],
  13882. [0.6200, 0.4118, 0.8287, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799],
  13883. [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217],
  13884. [ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600],
  13885. [0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733],
  13886. [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646],
  13887. [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649]],
  13888. device='cuda:0', dtype=torch.float64)
  13889. predictions are: tensor([[ 0.7951, -0.3181, 1.3667, -0.9975, -0.6434, -0.7699, 0.4930, 0.1455],
  13890. [ 0.6817, -0.3643, 1.6849, -0.0330, -0.1792, -0.0081, 0.2371, 0.2157],
  13891. [ 0.5805, -0.4193, 1.4511, -0.7298, -0.5682, -1.1159, 0.2997, 0.2842],
  13892. [ 0.7572, -0.3580, 1.8490, -0.1689, -0.4425, 0.1329, 0.8386, 0.1823],
  13893. [-1.5188, -1.8176, 1.2768, -1.1645, -0.4267, -1.1418, 0.3316, 0.2596],
  13894. [ 0.8045, -0.2769, 1.7545, -0.1248, -0.5806, -0.3896, 0.5359, 0.3604],
  13895. [ 0.3195, -0.5944, 1.3143, -1.0067, -0.6936, -0.9331, 0.1250, 0.2017],
  13896. [ 0.8338, -0.3000, 1.6501, 0.2392, -0.5147, -0.1556, 0.5634, 0.1801]],
  13897. device='cuda:0', grad_fn=<AddmmBackward>)
  13898. landmarks are: tensor([[[ 0.5614, -0.4514, 1.3252, -0.7240, -0.6237, -0.7215, 0.4299,
  13899. 0.0942],
  13900. [ 0.5202, -0.4382, 1.6460, 0.1082, -0.2029, 0.1754, 0.1067,
  13901. 0.1530],
  13902. [ 0.5779, -0.3840, 1.5420, -0.4306, -0.5423, -0.9772, 0.2041,
  13903. 0.3928],
  13904. [ 0.6047, -0.4209, 1.7557, -0.0303, -0.4845, 0.2545, 0.6587,
  13905. 0.1236],
  13906. [-2.2859, -2.2859, 1.4266, -1.1389, -0.4499, -1.1235, 0.2891,
  13907. 0.3007],
  13908. [ 0.6121, -0.3138, 1.8423, 0.0082, -0.6462, -0.3075, 0.3469,
  13909. 0.3623],
  13910. [ 0.5483, -0.4105, 1.4208, -0.8002, -0.6000, -0.9002, 0.0511,
  13911. 0.3220],
  13912. [ 0.6105, -0.4360, 1.6171, 0.5162, -0.5076, 0.0159, 0.5171,
  13913. -0.1385]]], device='cuda:0')
  13914. loss_train_step before backward: tensor(0.0298, device='cuda:0', grad_fn=<MseLossBackward>)
  13915. loss_train_step after backward: tensor(0.0298, device='cuda:0', grad_fn=<MseLossBackward>)
  13916. loss_train: 2.1901937695220113
  13917. step: 52
  13918. running loss: 0.04211911095234637
  13919. Train Steps: 52/90 Loss: 0.0421 torch.Size([8, 600, 800])
  13920. torch.Size([8, 8])
  13921. tensor([[0.6136, 0.4085, 0.6688, 0.2317, 0.3862, 0.2367, 0.5517, 0.5783],
  13922. [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758],
  13923. [0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6262, 0.5167],
  13924. [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5963, 0.6217],
  13925. [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478],
  13926. [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947],
  13927. [0.6230, 0.4113, 0.7213, 0.1983, 0.4325, 0.2367, 0.6262, 0.5400],
  13928. [0.6339, 0.4123, 0.8638, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436]],
  13929. device='cuda:0', dtype=torch.float64)
  13930. predictions are: tensor([[ 0.2910, -0.6319, 0.8737, -1.2362, -0.5089, -1.1692, 0.2117, 0.3007],
  13931. [ 0.5419, -0.4641, 1.7910, -0.0319, -0.4816, 0.1344, 0.4855, 0.1718],
  13932. [ 0.4724, -0.5371, 1.8747, -0.3285, -0.5541, -0.7206, 0.4187, 0.2375],
  13933. [ 0.4847, -0.4866, 1.0764, -1.1323, -0.4219, -1.1624, 0.3798, 0.4126],
  13934. [ 0.3838, -0.5823, 1.7736, -0.6931, -0.6253, -0.8531, 0.3493, 0.2185],
  13935. [ 0.4651, -0.5428, 1.7617, 0.0150, -0.6195, 0.0082, 0.4829, 0.1659],
  13936. [-0.0310, -0.8469, 1.2318, -1.2809, -0.3572, -1.3031, 0.5550, 0.2932],
  13937. [ 0.6677, -0.3943, 1.7676, 0.1537, -0.4608, 0.2785, 0.6763, 0.2784]],
  13938. device='cuda:0', grad_fn=<AddmmBackward>)
  13939. landmarks are: tensor([[[ 5.4827e-01, -3.9908e-01, 8.0300e-01, -1.2159e+00, -5.0185e-01,
  13940. -1.1928e+00, 2.6243e-01, 3.8516e-01],
  13941. [ 6.0716e-01, -4.2502e-01, 1.8249e+00, -7.2363e-03, -4.0370e-01,
  13942. 1.0824e-01, 6.7296e-01, -8.8090e-02],
  13943. [ 6.0935e-01, -3.9469e-01, 1.8885e+00, -2.9977e-01, -5.7691e-01,
  13944. -6.7698e-01, 6.0670e-01, 1.0054e-01],
  13945. [ 5.6871e-01, -4.0878e-01, 1.0397e+00, -1.1466e+00, -3.1132e-01,
  13946. -1.1928e+00, 4.6813e-01, 5.8553e-01],
  13947. [ 5.7921e-01, -4.0523e-01, 1.8214e+00, -6.5874e-01, -5.3842e-01,
  13948. -8.9239e-01, 4.3812e-01, 2.4425e-01],
  13949. [ 5.8031e-01, -4.3672e-01, 1.8423e+00, -3.0331e-02, -5.9423e-01,
  13950. 2.3557e-02, 6.5034e-01, -9.2270e-04],
  13951. [ 5.9169e-01, -3.8607e-01, 1.0455e+00, -1.3698e+00, -2.8822e-01,
  13952. -1.1928e+00, 6.0670e-01, 2.0831e-01],
  13953. [ 6.4212e-01, -3.8157e-01, 1.7037e+00, 1.9292e-01, -4.0370e-01,
  13954. 2.3911e-01, 1.1861e+00, 2.2489e-01]]], device='cuda:0')
  13955. loss_train_step before backward: tensor(0.0241, device='cuda:0', grad_fn=<MseLossBackward>)
  13956. loss_train_step after backward: tensor(0.0241, device='cuda:0', grad_fn=<MseLossBackward>)
  13957. loss_train: 2.2142448211088777
  13958. step: 53
  13959. running loss: 0.04177820417186562
  13960.  
  13961. Train Steps: 53/90 Loss: 0.0418 torch.Size([8, 600, 800])
  13962. torch.Size([8, 8])
  13963. tensor([[0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517],
  13964. [0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6188, 0.5433],
  13965. [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667],
  13966. [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153],
  13967. [0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433],
  13968. [0.6265, 0.4071, 0.8875, 0.3367, 0.3975, 0.3350, 0.6312, 0.5250],
  13969. [0.6212, 0.4159, 0.8675, 0.5783, 0.4088, 0.4317, 0.5613, 0.5917],
  13970. [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749]],
  13971. device='cuda:0', dtype=torch.float64)
  13972. predictions are: tensor([[ 2.5260e-02, -7.6672e-01, 1.0277e+00, -1.2165e+00, -3.4002e-01,
  13973. -1.3102e+00, 4.8023e-01, 3.9226e-01],
  13974. [ 6.2153e-01, -4.3315e-01, 1.9337e+00, 1.7966e-03, -5.1253e-01,
  13975. 9.0811e-02, 5.9042e-01, 1.6422e-01],
  13976. [ 4.9724e-01, -5.0812e-01, 1.6399e+00, -7.6851e-01, -6.8466e-01,
  13977. -7.5338e-01, 3.2359e-01, 3.3033e-01],
  13978. [ 6.1532e-01, -4.3510e-01, 1.4187e+00, -8.1107e-01, -6.2200e-01,
  13979. -7.1508e-01, 4.0486e-01, 1.7140e-01],
  13980. [ 3.6046e-01, -5.7041e-01, 1.3250e+00, -1.0281e+00, -6.1943e-01,
  13981. -6.3158e-01, 4.3742e-01, 2.4319e-01],
  13982. [ 3.0706e-01, -6.2405e-01, 1.9129e+00, -7.9856e-01, -4.4438e-01,
  13983. -9.1659e-01, 7.7821e-01, 1.9522e-01],
  13984. [ 5.9681e-01, -4.0429e-01, 1.6356e+00, 4.3390e-01, -3.5430e-01,
  13985. -3.2069e-01, 4.2337e-01, 4.7371e-01],
  13986. [ 2.7878e-01, -6.1629e-01, 1.1768e+00, -9.2799e-01, -6.1763e-01,
  13987. -7.4626e-01, 2.4131e-01, 3.0193e-01]], device='cuda:0',
  13988. grad_fn=<AddmmBackward>)
  13989. landmarks are: tensor([[[ 0.5713, -0.3671, 0.8665, -1.0696, -0.3691, -1.2236, 0.3527,
  13990. 0.2622],
  13991. [ 0.5773, -0.4105, 1.8192, -0.1304, -0.5307, 0.0467, 0.5721,
  13992. 0.2237],
  13993. [ 0.5768, -0.3857, 1.5305, -0.7694, -0.6462, -0.6308, 0.3988,
  13994. 0.3315],
  13995. [ 0.5614, -0.4514, 1.3252, -0.7240, -0.6237, -0.7215, 0.4299,
  13996. 0.0942],
  13997. [ 0.5665, -0.3963, 1.3457, -0.9926, -0.6058, -0.6308, 0.5721,
  13998. 0.2237],
  13999. [ 0.6078, -0.4057, 1.8134, -0.7309, -0.4499, -0.7386, 0.6298,
  14000. 0.1390],
  14001. [ 0.5836, -0.3649, 1.7210, 0.3854, -0.3979, -0.2921, 0.3065,
  14002. 0.4470],
  14003. [ 0.5419, -0.4160, 1.1810, -0.8939, -0.6808, -0.7463, 0.2485,
  14004. 0.3694]]], device='cuda:0')
  14005. loss_train_step before backward: tensor(0.0175, device='cuda:0', grad_fn=<MseLossBackward>)
  14006. loss_train_step after backward: tensor(0.0175, device='cuda:0', grad_fn=<MseLossBackward>)
  14007. loss_train: 2.2317135045304894
  14008. step: 54
  14009. running loss: 0.04132802786167573
  14010. Train Steps: 54/90 Loss: 0.0413 torch.Size([8, 600, 800])
  14011. torch.Size([8, 8])
  14012. tensor([[0.6201, 0.3970, 0.8413, 0.4950, 0.4413, 0.5183, 0.6088, 0.5400],
  14013. [0.6307, 0.4029, 0.8988, 0.4817, 0.3937, 0.3500, 0.7311, 0.5378],
  14014. [0.6136, 0.4085, 0.6688, 0.2317, 0.3862, 0.2367, 0.5517, 0.5783],
  14015. [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103],
  14016. [0.6234, 0.4179, 0.7825, 0.3450, 0.3813, 0.2867, 0.5675, 0.5617],
  14017. [0.6192, 0.4128, 0.8513, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633],
  14018. [0.6261, 0.4045, 0.8865, 0.5369, 0.3895, 0.4859, 0.6683, 0.5249],
  14019. [0.6198, 0.4164, 0.8700, 0.5067, 0.4625, 0.5650, 0.5464, 0.5197]],
  14020. device='cuda:0', dtype=torch.float64)
  14021. predictions are: tensor([[ 0.3991, -0.5190, 1.6326, -0.1913, -0.3595, -0.1455, 0.4408, 0.2888],
  14022. [ 0.2619, -0.6540, 1.8745, -0.3648, -0.5207, -0.9587, 0.7523, 0.2107],
  14023. [ 0.3949, -0.5408, 0.8073, -1.4307, -0.5081, -1.3098, 0.3159, 0.3495],
  14024. [ 0.3794, -0.5318, 1.4255, -1.1229, -0.7340, -0.7321, 0.3473, 0.2941],
  14025. [ 0.3057, -0.5531, 1.3458, -1.0888, -0.5595, -1.1431, 0.3626, 0.4253],
  14026. [ 0.6391, -0.3797, 1.6804, 0.1259, -0.4331, -0.0430, 0.4921, 0.3337],
  14027. [ 0.5814, -0.4455, 1.7452, -0.0697, -0.6352, -0.1981, 0.7558, 0.1840],
  14028. [ 0.3373, -0.5463, 1.6790, -0.1625, -0.2885, -0.1123, 0.2628, 0.2724]],
  14029. device='cuda:0', grad_fn=<AddmmBackward>)
  14030. landmarks are: tensor([[[ 5.7812e-01, -4.5219e-01, 1.5998e+00, 4.6189e-04, -2.4781e-01,
  14031. 1.0824e-01, 5.2587e-01, 2.0831e-01],
  14032. [ 6.2730e-01, -4.2490e-01, 1.8654e+00, -6.1124e-02, -4.6721e-01,
  14033. -6.6928e-01, 1.0910e+00, 1.9818e-01],
  14034. [ 5.4827e-01, -3.9908e-01, 8.0300e-01, -1.2159e+00, -5.0185e-01,
  14035. -1.1928e+00, 2.6243e-01, 3.8516e-01],
  14036. [ 5.8863e-01, -3.7837e-01, 1.4554e+00, -9.0793e-01, -6.5774e-01,
  14037. -4.8453e-01, 3.4395e-01, 7.1216e-02],
  14038. [ 5.9348e-01, -3.5581e-01, 1.3284e+00, -6.9238e-01, -5.2494e-01,
  14039. -9.6182e-01, 3.3533e-01, 3.0839e-01],
  14040. [ 5.7419e-01, -3.7921e-01, 1.6460e+00, 3.0839e-01, -3.4596e-01,
  14041. 1.4673e-01, 4.1617e-01, 3.1609e-01],
  14042. [ 6.0589e-01, -4.1768e-01, 1.8087e+00, 1.9408e-01, -4.8680e-01,
  14043. -4.1391e-02, 8.0095e-01, 1.3848e-01],
  14044. [ 5.7696e-01, -3.6243e-01, 1.7326e+00, 5.4350e-02, -1.4965e-01,
  14045. 3.2379e-01, 2.3775e-01, 1.1464e-01]]], device='cuda:0')
  14046. loss_train_step before backward: tensor(0.0297, device='cuda:0', grad_fn=<MseLossBackward>)
  14047. loss_train_step after backward: tensor(0.0297, device='cuda:0', grad_fn=<MseLossBackward>)
  14048. loss_train: 2.261423231102526
  14049. step: 55
  14050. running loss: 0.04111678602004593
  14051. Train Steps: 55/90 Loss: 0.0411 torch.Size([8, 600, 800])
  14052. torch.Size([8, 8])
  14053. tensor([[0.6118, 0.4052, 0.8463, 0.3917, 0.3538, 0.3450, 0.5053, 0.5593],
  14054. [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050],
  14055. [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167],
  14056. [0.6274, 0.4003, 0.8638, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661],
  14057. [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167],
  14058. [0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5737, 0.5350],
  14059. [0.6252, 0.4158, 0.8988, 0.4083, 0.3788, 0.4783, 0.6225, 0.5633],
  14060. [0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804]],
  14061. device='cuda:0', dtype=torch.float64)
  14062. predictions are: tensor([[ 0.0752, -0.7276, 1.4581, -0.7417, -0.6950, -0.7660, 0.0415, 0.3244],
  14063. [ 0.3550, -0.5610, 1.6460, -0.5112, -0.6129, -0.5533, 0.4340, 0.2236],
  14064. [ 0.4733, -0.4840, 1.6904, -0.1513, -0.3921, -0.0886, 0.5491, 0.2488],
  14065. [ 0.5271, -0.4809, 1.5573, 0.0957, -0.5033, -0.2430, 0.6557, 0.2935],
  14066. [ 0.3911, -0.4719, 1.0264, -1.1537, -0.1195, -1.2667, 0.3460, 0.4831],
  14067. [ 0.4838, -0.4367, 1.6707, -0.3050, -0.5068, -0.9674, 0.4554, 0.3163],
  14068. [ 0.3380, -0.5800, 1.6796, -0.6661, -0.6049, -0.2165, 0.6289, 0.3577],
  14069. [ 0.7290, -0.3204, 1.6073, -0.8694, -0.6387, -0.6692, 0.6848, 0.1262]],
  14070. device='cuda:0', grad_fn=<AddmmBackward>)
  14071. landmarks are: tensor([[[ 0.5399, -0.4142, 1.6229, -0.4768, -0.6520, -0.6924, 0.0481,
  14072. 0.2972],
  14073. [ 0.5844, -0.4466, 1.8423, -0.1997, -0.5942, -0.3998, 0.4219,
  14074. 0.0467],
  14075. [ 0.5828, -0.4066, 1.7557, 0.0774, -0.3691, -0.0226, 0.4277,
  14076. 0.1005],
  14077. [ 0.6120, -0.4371, 1.7037, 0.4701, -0.5827, -0.0226, 0.5354,
  14078. -0.1331],
  14079. [ 0.6161, -0.3075, 1.1678, -0.6500, 0.0813, -1.4006, 0.2545,
  14080. 0.5624],
  14081. [ 0.5901, -0.4000, 1.8423, -0.0688, -0.5307, -0.9233, 0.3642,
  14082. 0.1852],
  14083. [ 0.6017, -0.3654, 1.8654, -0.3998, -0.5365, -0.0765, 0.5894,
  14084. 0.3161],
  14085. [ 0.6071, -0.4119, 1.7788, -0.5153, -0.6000, -0.5692, 0.6586,
  14086. -0.0670]]], device='cuda:0')
  14087. loss_train_step before backward: tensor(0.0350, device='cuda:0', grad_fn=<MseLossBackward>)
  14088. loss_train_step after backward: tensor(0.0350, device='cuda:0', grad_fn=<MseLossBackward>)
  14089. loss_train: 2.296431795693934
  14090. step: 56
  14091. running loss: 0.04100771063739168
  14092. Train Steps: 56/90 Loss: 0.0410 torch.Size([8, 600, 800])
  14093. torch.Size([8, 8])
  14094. tensor([[0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240],
  14095. [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6038, 0.4833],
  14096. [0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6262, 0.5167],
  14097. [0.6208, 0.4082, 0.8538, 0.3067, 0.3588, 0.3717, 0.6112, 0.5517],
  14098. [0.6219, 0.4097, 0.8738, 0.3400, 0.3563, 0.4117, 0.5975, 0.5683],
  14099. [ nan, nan, 0.7612, 0.3250, 0.4037, 0.2533, 0.5438, 0.5767],
  14100. [0.6179, 0.4118, 0.7278, 0.4237, 0.3588, 0.3400, 0.5675, 0.5917],
  14101. [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524]],
  14102. device='cuda:0', dtype=torch.float64)
  14103. predictions are: tensor([[-0.5338, -1.0908, 1.0218, -1.2956, -0.2762, -1.3676, 0.3314, 0.2925],
  14104. [ 0.7514, -0.3161, 1.7225, -0.0812, -0.4777, -0.1617, 0.6445, 0.1176],
  14105. [ 0.6451, -0.3701, 1.8502, -0.3280, -0.4483, -0.7106, 0.7212, 0.2513],
  14106. [ 0.7114, -0.2943, 1.5651, -0.8324, -0.6431, -0.5288, 0.6222, 0.2602],
  14107. [ 0.7963, -0.2482, 1.6338, -0.7183, -0.6423, -0.4676, 0.5556, 0.3234],
  14108. [-0.1757, -0.8331, 1.1903, -0.9809, -0.4011, -1.0549, 0.2312, 0.3846],
  14109. [ 0.4617, -0.4524, 1.3170, -0.5441, -0.5445, -0.7966, 0.2700, 0.4307],
  14110. [ 0.5571, -0.3778, 1.7054, 0.1239, -0.4832, -0.2008, 0.3696, 0.2623]],
  14111. device='cuda:0', grad_fn=<AddmmBackward>)
  14112. landmarks are: tensor([[[ 0.5676, -0.4112, 1.1898, -1.2467, -0.2940, -1.4622, 0.2103,
  14113. 0.1343],
  14114. [ 0.5726, -0.4795, 1.7788, -0.0149, -0.5711, -0.0688, 0.5028,
  14115. -0.0534],
  14116. [ 0.6094, -0.3947, 1.8885, -0.2998, -0.5769, -0.6770, 0.6067,
  14117. 0.1005],
  14118. [ 0.5814, -0.4003, 1.6575, -0.8694, -0.6289, -0.5692, 0.5374,
  14119. 0.2622],
  14120. [ 0.5867, -0.3937, 1.7499, -0.7155, -0.6404, -0.3844, 0.4739,
  14121. 0.3392],
  14122. [-2.2859, -2.2859, 1.2303, -0.7848, -0.4210, -1.1158, 0.2256,
  14123. 0.3777],
  14124. [ 0.5680, -0.3840, 1.0756, -0.3290, -0.6289, -0.7155, 0.3353,
  14125. 0.4470],
  14126. [ 0.5529, -0.3811, 1.7788, -0.0380, -0.5307, -0.2074, 0.0727,
  14127. 0.2657]]], device='cuda:0')
  14128. loss_train_step before backward: tensor(0.1399, device='cuda:0', grad_fn=<MseLossBackward>)
  14129.  
  14130. loss_train_step after backward: tensor(0.1399, device='cuda:0', grad_fn=<MseLossBackward>)
  14131. loss_train: 2.436326772905886
  14132. step: 57
  14133. running loss: 0.042742574963261164
  14134. Train Steps: 57/90 Loss: 0.0427 torch.Size([8, 600, 800])
  14135. torch.Size([8, 8])
  14136. tensor([[0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633],
  14137. [0.6332, 0.4118, 0.9238, 0.4267, 0.4012, 0.4733, 0.7525, 0.5436],
  14138. [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575],
  14139. [0.6199, 0.4102, 0.8950, 0.4417, 0.4012, 0.5367, 0.6112, 0.5967],
  14140. [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395],
  14141. [0.6163, 0.4114, 0.7650, 0.2017, 0.3763, 0.2867, 0.5631, 0.5071],
  14142. [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285],
  14143. [0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183]],
  14144. device='cuda:0', dtype=torch.float64)
  14145. predictions are: tensor([[ 3.8910e-01, -4.9305e-01, 1.6736e+00, -1.0089e+00, -2.0297e-01,
  14146. -1.2898e+00, 8.2985e-01, 3.0091e-01],
  14147. [ 4.2471e-01, -4.7600e-01, 1.7696e+00, -1.4306e-01, -4.9965e-01,
  14148. -9.5795e-02, 7.4937e-01, 2.1200e-01],
  14149. [ 5.2550e-01, -4.1549e-01, 1.6856e+00, -5.4736e-01, -4.8786e-01,
  14150. -3.7032e-01, 7.5352e-01, 2.1056e-01],
  14151. [ 4.8281e-01, -4.3335e-01, 1.8100e+00, 1.3539e-03, -3.4686e-01,
  14152. 2.0120e-01, 5.2785e-01, 3.4740e-01],
  14153. [ 4.8238e-01, -4.4421e-01, 1.1127e+00, -1.1223e+00, -4.4218e-01,
  14154. -1.2524e+00, 3.8058e-01, 2.7940e-01],
  14155. [ 2.9010e-01, -5.9548e-01, 1.1687e+00, -1.0469e+00, -6.2078e-01,
  14156. -1.0666e+00, 3.2697e-01, 1.9013e-01],
  14157. [ 3.8660e-01, -4.7169e-01, 1.3892e+00, -2.7819e-01, -6.6921e-01,
  14158. -3.4179e-01, -8.3418e-02, 1.6513e-01],
  14159. [ 4.1583e-01, -4.8280e-01, 1.3151e+00, -8.3128e-01, -5.0556e-01,
  14160. -9.7859e-01, 4.4635e-01, 5.0129e-01]], device='cuda:0',
  14161. grad_fn=<AddmmBackward>)
  14162. landmarks are: tensor([[[ 0.6042, -0.4105, 1.5478, -1.2082, -0.1208, -1.0927, 0.9704,
  14163. 0.3157],
  14164. [ 0.6388, -0.3840, 1.9808, -0.3152, -0.4326, -0.0996, 1.1897,
  14165. 0.2249],
  14166. [ 0.6158, -0.4249, 1.8654, -0.9002, -0.3229, -0.3537, 0.9667,
  14167. 0.2890],
  14168. [ 0.5772, -0.3913, 1.8480, -0.2459, -0.4326, 0.1929, 0.5374,
  14169. 0.4701],
  14170. [ 0.5713, -0.4538, 1.1028, -1.3659, -0.3831, -1.1273, 0.5340,
  14171. 0.2058],
  14172. [ 0.5606, -0.3856, 1.2476, -1.3544, -0.5480, -0.9618, 0.3152,
  14173. 0.0562],
  14174. [ 0.5253, -0.4296, 1.3084, -0.3098, -0.6115, -0.2767, 0.0928,
  14175. 0.1552],
  14176. [ 0.5759, -0.3948, 1.2072, -0.9849, -0.4672, -0.8156, 0.4912,
  14177. 0.5701]]], device='cuda:0')
  14178. loss_train_step before backward: tensor(0.0216, device='cuda:0', grad_fn=<MseLossBackward>)
  14179. loss_train_step after backward: tensor(0.0216, device='cuda:0', grad_fn=<MseLossBackward>)
  14180. loss_train: 2.4578881757333875
  14181. step: 58
  14182. running loss: 0.04237738234023082
  14183. Train Steps: 58/90 Loss: 0.0424 torch.Size([8, 600, 800])
  14184. torch.Size([8, 8])
  14185. tensor([[0.6245, 0.4100, 0.7762, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417],
  14186. [0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5737, 0.5350],
  14187. [0.6272, 0.4045, 0.8538, 0.5900, 0.3750, 0.4417, 0.5989, 0.4649],
  14188. [0.6275, 0.4050, 0.9038, 0.3767, 0.3838, 0.3533, 0.7074, 0.5575],
  14189. [ nan, nan, 0.8488, 0.2300, 0.5563, 0.2100, 0.7390, 0.5679],
  14190. [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533],
  14191. [0.6266, 0.4067, 0.8588, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267],
  14192. [0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433]],
  14193. device='cuda:0', dtype=torch.float64)
  14194. predictions are: tensor([[ 0.4276, -0.4216, 1.0481, -1.0719, -0.1652, -1.3034, 0.2338, 0.3338],
  14195. [ 0.6741, -0.2671, 1.6596, 0.0128, -0.5760, -0.7280, 0.2549, 0.2385],
  14196. [ 0.5957, -0.3851, 1.6734, 0.3216, -0.6498, -0.2558, 0.4546, 0.1138],
  14197. [ 0.6606, -0.3073, 1.7693, -0.5237, -0.6310, -0.6105, 0.7902, 0.2340],
  14198. [-0.8477, -1.3076, 1.6011, -1.1105, 0.0148, -1.0858, 0.9775, 0.3887],
  14199. [ 0.6939, -0.2546, 1.6675, -0.2473, -0.4186, 0.2012, 0.3346, 0.2056],
  14200. [ 0.5684, -0.3351, 1.5003, -1.0045, -0.4395, -0.9511, 0.5698, 0.2595],
  14201. [ 0.5738, -0.3424, 1.1729, -1.0006, -0.6760, -0.5028, 0.3477, 0.2576]],
  14202. device='cuda:0', grad_fn=<AddmmBackward>)
  14203. landmarks are: tensor([[[ 0.5985, -0.3921, 1.2995, -1.0927, 0.0062, -1.5854, 0.4277,
  14204. 0.2160],
  14205. [ 0.5901, -0.4000, 1.8423, -0.0688, -0.5307, -0.9233, 0.3642,
  14206. 0.1852],
  14207. [ 0.6109, -0.4177, 1.6575, 0.4393, -0.5538, -0.2459, 0.4805,
  14208. -0.1385],
  14209. [ 0.6125, -0.4153, 1.8885, -0.5461, -0.5134, -0.6539, 0.9814,
  14210. 0.2890],
  14211. [-2.2859, -2.2859, 1.6344, -1.2236, 0.2834, -1.3159, 1.1276,
  14212. 0.3371],
  14213. [ 0.5499, -0.4225, 1.8018, -0.2921, -0.3055, 0.0543, 0.4046,
  14214. 0.2699],
  14215. [ 0.6084, -0.4076, 1.6806, -0.9618, -0.2998, -0.9695, 0.6356,
  14216. 0.1467],
  14217. [ 0.5665, -0.3963, 1.3457, -0.9926, -0.6058, -0.6308, 0.5721,
  14218. 0.2237]]], device='cuda:0')
  14219. loss_train_step before backward: tensor(0.0630, device='cuda:0', grad_fn=<MseLossBackward>)
  14220. loss_train_step after backward: tensor(0.0630, device='cuda:0', grad_fn=<MseLossBackward>)
  14221. loss_train: 2.520900216884911
  14222. step: 59
  14223. running loss: 0.04272712232008324
  14224. Train Steps: 59/90 Loss: 0.0427 torch.Size([8, 600, 800])
  14225. torch.Size([8, 8])
  14226. tensor([[0.6275, 0.4048, 0.8488, 0.2883, 0.4463, 0.2033, 0.6321, 0.5155],
  14227. [0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117],
  14228. [0.6314, 0.4107, 0.8750, 0.5100, 0.3788, 0.4900, 0.7121, 0.5864],
  14229. [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261],
  14230. [0.6325, 0.4165, 0.9000, 0.4617, 0.3813, 0.4900, 0.7485, 0.5447],
  14231. [0.6265, 0.4071, 0.8875, 0.3367, 0.3975, 0.3350, 0.6312, 0.5250],
  14232. [0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5988, 0.5700],
  14233. [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927]],
  14234. device='cuda:0', dtype=torch.float64)
  14235. predictions are: tensor([[ 0.9893, -0.0856, 1.4816, -1.0489, -0.1731, -1.3726, 0.6884, 0.1718],
  14236. [ 0.5411, -0.3862, 1.4689, 0.1822, -0.3924, -0.1881, 0.3981, 0.4081],
  14237. [ 0.5717, -0.3800, 1.8185, 0.0153, -0.4474, -0.1558, 0.8028, 0.2387],
  14238. [ 0.0362, -0.6552, 1.6265, -0.2109, -0.3773, -0.2945, 0.1722, 0.2111],
  14239. [ 0.5384, -0.3899, 1.8012, -0.2197, -0.4780, -0.1419, 0.7585, 0.1865],
  14240. [ 0.3051, -0.5194, 1.7123, -0.9239, -0.4580, -0.9475, 0.5561, 0.2275],
  14241. [-0.0521, -0.7613, 1.2508, -1.1285, -0.5612, -0.7800, 0.2824, 0.3768],
  14242. [ 0.6793, -0.2890, 1.6110, -0.4560, -0.6494, -0.4286, 0.3983, 0.0995]],
  14243. device='cuda:0', grad_fn=<AddmmBackward>)
  14244. landmarks are: tensor([[[ 0.6126, -0.4161, 1.6344, -0.9541, -0.2247, -1.3467, 0.6339,
  14245. 0.0953],
  14246. [ 0.6118, -0.3983, 1.5824, 0.3469, -0.4268, -0.0688, 0.3469,
  14247. 0.5393],
  14248. [ 0.6306, -0.3888, 1.7557, 0.0697, -0.5365, -0.0226, 1.0033,
  14249. 0.4226],
  14250. [ 0.5746, -0.3623, 1.7961, -0.1150, -0.3691, -0.0380, 0.2208,
  14251. 0.1439],
  14252. [ 0.6355, -0.3623, 1.8711, -0.1535, -0.5249, -0.0226, 1.1715,
  14253. 0.2302],
  14254. [ 0.6078, -0.4057, 1.8134, -0.7309, -0.4499, -0.7386, 0.6298,
  14255. 0.1390],
  14256. [ 0.5960, -0.3888, 1.4840, -1.0095, -0.6115, -0.6231, 0.4797,
  14257. 0.3469],
  14258. [ 0.6010, -0.4345, 1.8480, -0.3537, -0.6462, -0.2613, 0.6524,
  14259. -0.0099]]], device='cuda:0')
  14260. loss_train_step before backward: tensor(0.0357, device='cuda:0', grad_fn=<MseLossBackward>)
  14261. loss_train_step after backward: tensor(0.0357, device='cuda:0', grad_fn=<MseLossBackward>)
  14262. loss_train: 2.5565678672865033
  14263. step: 60
  14264. running loss: 0.04260946445477506
  14265. Train Steps: 60/90 Loss: 0.0426 torch.Size([8, 600, 800])
  14266. torch.Size([8, 8])
  14267. tensor([[0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882],
  14268. [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459],
  14269. [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377],
  14270. [0.6053, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168],
  14271. [ nan, nan, 0.6688, 0.2513, 0.4113, 0.2117, 0.5193, 0.5933],
  14272. [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938],
  14273. [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355],
  14274. [0.6110, 0.4047, 0.8700, 0.4483, 0.3713, 0.3967, 0.5088, 0.5517]],
  14275. device='cuda:0', dtype=torch.float64)
  14276. predictions are: tensor([[ 0.7788, -0.2101, 1.7978, -0.4085, -0.6473, -0.3535, 0.1756, 0.1227],
  14277. [ 0.8946, -0.1416, 1.6702, -0.9497, 0.0064, -1.1562, 0.9449, 0.2182],
  14278. [ 0.9714, -0.1470, 1.8020, 0.1940, -0.4284, 0.0314, 0.6996, 0.1681],
  14279. [-0.5611, -1.1197, 0.9617, -1.1811, -0.3930, -1.2175, 0.3303, 0.3160],
  14280. [-1.0809, -1.4155, 0.9003, -1.1566, -0.3876, -1.1537, 0.2993, 0.4127],
  14281. [ 0.6877, -0.3139, 1.9421, 0.1481, -0.5822, -0.3625, 0.5538, 0.1134],
  14282. [ 0.7454, -0.2486, 1.7348, -0.7554, -0.2176, -0.6656, 0.9340, 0.1684],
  14283. [ 0.8294, -0.1889, 1.6893, -0.1422, -0.6164, -0.2008, 0.1228, 0.2367]],
  14284. device='cuda:0', grad_fn=<AddmmBackward>)
  14285. landmarks are: tensor([[[ 0.5595, -0.3988, 1.7672, -0.4460, -0.5538, -0.5384, 0.0828,
  14286. -0.0310],
  14287. [ 0.6520, -0.3912, 1.7095, -0.9079, -0.0284, -1.3621, 0.8096,
  14288. 0.2356],
  14289. [ 0.5775, -0.4607, 1.6741, 0.1962, -0.4036, -0.1212, 0.4588,
  14290. 0.1979],
  14291. [ 0.5102, -0.4223, 0.8999, -1.4011, -0.4383, -1.3082, 0.2267,
  14292. 0.1013],
  14293. [-2.2859, -2.2859, 0.8033, -1.1250, -0.3864, -1.3082, 0.1126,
  14294. 0.4543],
  14295. [ 0.6273, -0.4105, 1.8480, 0.1082, -0.5538, -0.5076, 0.6414,
  14296. -0.0049],
  14297. [ 0.6224, -0.4105, 1.9173, -0.7771, -0.1030, -0.7308, 1.1532,
  14298. 0.1875],
  14299. [ 0.5363, -0.4168, 1.7326, -0.2151, -0.5711, -0.4537, 0.0640,
  14300. 0.2622]]], device='cuda:0')
  14301. loss_train_step before backward: tensor(0.0806, device='cuda:0', grad_fn=<MseLossBackward>)
  14302. loss_train_step after backward: tensor(0.0806, device='cuda:0', grad_fn=<MseLossBackward>)
  14303. loss_train: 2.6371513130143285
  14304. step: 61
  14305. running loss: 0.04323198873793981
  14306.  
  14307. Train Steps: 61/90 Loss: 0.0432 torch.Size([8, 600, 800])
  14308. torch.Size([8, 8])
  14309. tensor([[0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500],
  14310. [0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400],
  14311. [0.6175, 0.3957, 0.8700, 0.4817, 0.4662, 0.5133, 0.5800, 0.5517],
  14312. [0.6193, 0.4108, 0.7438, 0.2700, 0.3650, 0.3683, 0.6238, 0.5717],
  14313. [ nan, nan, 0.6900, 0.1917, 0.3937, 0.2367, 0.5240, 0.5246],
  14314. [0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433],
  14315. [0.6250, 0.4054, 0.8770, 0.4723, 0.4662, 0.5367, 0.6162, 0.5433],
  14316. [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517]],
  14317. device='cuda:0', dtype=torch.float64)
  14318. predictions are: tensor([[ 0.5904, -0.3481, 1.3853, -0.9300, -0.7038, -0.5478, 0.6628, 0.1901],
  14319. [ 0.7472, -0.2219, 1.4910, -0.8652, -0.2222, -1.3944, 0.4443, 0.2165],
  14320. [ 0.5629, -0.3763, 1.8600, -0.0448, -0.2762, -0.1915, 0.5234, 0.1410],
  14321. [ 0.7760, -0.2245, 1.2375, -0.9950, -0.6542, -0.6808, 0.6083, 0.2506],
  14322. [-2.2670, -2.2596, 1.1679, -1.1750, -0.4814, -1.3681, 0.2612, 0.2891],
  14323. [ 0.9036, -0.1748, 1.7474, 0.1289, -0.1123, -0.3393, 0.3888, 0.2046],
  14324. [ 0.9780, -0.1009, 1.8675, -0.0857, -0.2561, -0.0475, 0.7317, 0.1477],
  14325. [ 0.9501, -0.1388, 1.8309, -0.0459, -0.2355, 0.0528, 0.6644, 0.2409]],
  14326. device='cuda:0', grad_fn=<AddmmBackward>)
  14327. landmarks are: tensor([[[ 0.5763, -0.4147, 1.3226, -1.0619, -0.6635, -0.4152, 0.5374,
  14328. 0.2545],
  14329. [ 0.5911, -0.3888, 1.4727, -0.9541, -0.0919, -1.4930, 0.3988,
  14330. 0.2083],
  14331. [ 0.5662, -0.4581, 1.7326, -0.0611, -0.1323, 0.0851, 0.3931,
  14332. 0.2622],
  14333. [ 0.5747, -0.3886, 1.1494, -1.0388, -0.6000, -0.5846, 0.5952,
  14334. 0.3546],
  14335. [-2.2859, -2.2859, 0.9012, -1.4006, -0.4672, -1.1928, 0.1342,
  14336. 0.1373],
  14337. [ 0.5295, -0.4373, 1.6553, 0.1011, 0.0380, 0.0671, 0.0813,
  14338. 0.2237],
  14339. [ 0.6009, -0.4135, 1.7651, -0.1043, -0.1323, 0.1929, 0.5605,
  14340. 0.2237],
  14341. [ 0.5897, -0.3804, 1.7788, -0.0226, -0.2016, 0.3007, 0.6125,
  14342. 0.2622]]], device='cuda:0')
  14343. loss_train_step before backward: tensor(0.0268, device='cuda:0', grad_fn=<MseLossBackward>)
  14344. loss_train_step after backward: tensor(0.0268, device='cuda:0', grad_fn=<MseLossBackward>)
  14345. loss_train: 2.6639770111069083
  14346. step: 62
  14347. running loss: 0.04296737114688562
  14348. Train Steps: 62/90 Loss: 0.0430 torch.Size([8, 600, 800])
  14349. torch.Size([8, 8])
  14350. tensor([[0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585],
  14351. [0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136],
  14352. [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5988, 0.5283],
  14353. [0.6353, 0.4128, 0.8488, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540],
  14354. [0.6336, 0.4086, 0.8900, 0.3950, 0.3900, 0.2950, 0.6504, 0.5066],
  14355. [0.6200, 0.4098, 0.8237, 0.2917, 0.4012, 0.2967, 0.6000, 0.5683],
  14356. [0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689],
  14357. [0.6336, 0.4191, 0.8938, 0.5167, 0.3937, 0.3517, 0.7343, 0.5748]],
  14358. device='cuda:0', dtype=torch.float64)
  14359. predictions are: tensor([[ 0.3569, -0.4949, 1.6938, -0.3559, -0.2054, 0.1943, 0.5201, 0.1686],
  14360. [ 0.4867, -0.4273, 1.7035, -0.3954, -0.6700, -0.4678, 0.2266, 0.1097],
  14361. [ 0.6877, -0.3373, 1.6928, 0.1891, -0.4250, -0.0477, 0.4994, 0.1601],
  14362. [-0.9558, -1.3822, 1.5410, -1.2279, 0.1464, -1.2743, 0.9212, 0.3460],
  14363. [ 0.9958, -0.1204, 1.7140, -0.5015, -0.3964, -0.9068, 0.7189, 0.0435],
  14364. [ 0.8597, -0.1777, 1.4339, -1.0562, -0.4009, -0.8427, 0.4997, 0.2891],
  14365. [ 0.3033, -0.5074, 1.3019, -0.7391, -0.6111, -0.5300, 0.2686, 0.2457],
  14366. [ 0.5942, -0.3914, 1.8203, 0.0827, -0.3343, -0.6801, 0.7941, 0.2058]],
  14367. device='cuda:0', grad_fn=<AddmmBackward>)
  14368. landmarks are: tensor([[[ 0.5115, -0.4332, 1.7557, -0.3152, -0.2536, 0.3931, 0.4139,
  14369. 0.2936],
  14370. [ 0.5235, -0.4273, 1.7499, -0.4306, -0.5827, -0.4614, 0.1651,
  14371. 0.0862],
  14372. [ 0.5603, -0.4319, 1.7788, 0.1775, -0.5538, -0.0611, 0.4797,
  14373. 0.1544],
  14374. [ 0.6487, -0.3792, 1.6344, -1.0850, 0.2659, -1.5397, 0.8059,
  14375. 0.2730],
  14376. [ 0.6405, -0.3984, 1.8249, -0.4614, -0.4845, -0.9233, 0.7182,
  14377. 0.0539],
  14378. [ 0.5778, -0.3930, 1.5189, -0.9387, -0.4326, -0.9156, 0.4855,
  14379. 0.3392],
  14380. [ 0.5327, -0.4381, 1.2880, -0.6308, -0.6866, -0.5307, 0.2658,
  14381. 0.3417],
  14382. [ 0.6405, -0.3503, 1.8423, 0.1005, -0.4672, -0.6616, 1.1057,
  14383. 0.3692]]], device='cuda:0')
  14384. loss_train_step before backward: tensor(0.0702, device='cuda:0', grad_fn=<MseLossBackward>)
  14385. loss_train_step after backward: tensor(0.0702, device='cuda:0', grad_fn=<MseLossBackward>)
  14386. loss_train: 2.7341925194486976
  14387. step: 63
  14388. running loss: 0.043399881261090435
  14389. Train Steps: 63/90 Loss: 0.0434 torch.Size([8, 600, 800])
  14390. torch.Size([8, 8])
  14391. tensor([[0.6203, 0.4096, 0.8862, 0.4267, 0.3538, 0.4117, 0.6025, 0.5650],
  14392. [0.6091, 0.3997, 0.8314, 0.4334, 0.3788, 0.4550, 0.5213, 0.5656],
  14393. [0.6223, 0.4028, 0.8988, 0.4200, 0.3763, 0.5733, 0.6375, 0.5167],
  14394. [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5637, 0.5633],
  14395. [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717],
  14396. [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133],
  14397. [0.6277, 0.4118, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938],
  14398. [0.6135, 0.4115, 0.8838, 0.4667, 0.4288, 0.6050, 0.5778, 0.5097]],
  14399. device='cuda:0', dtype=torch.float64)
  14400. predictions are: tensor([[ 0.8489, -0.2227, 1.7997, -0.5051, -0.4486, -0.5622, 0.6693, 0.2327],
  14401. [ 0.2771, -0.5652, 1.6160, -0.3772, -0.4482, -0.3172, 0.3491, 0.1680],
  14402. [ 0.1536, -0.6882, 1.8718, -0.3745, -0.4070, 0.1483, 0.8937, 0.1137],
  14403. [ 0.5407, -0.3957, 1.7259, -0.1100, -0.5389, -0.6844, 0.4568, 0.2921],
  14404. [ 0.3525, -0.5436, 0.8690, -1.2150, -0.4159, -1.3178, 0.2979, 0.2886],
  14405. [ 0.5503, -0.4333, 1.7132, 0.0185, -0.0879, -0.1173, 0.5199, 0.1360],
  14406. [ 0.4545, -0.4887, 1.8710, -0.6144, -0.2999, -1.2758, 0.7149, 0.0996],
  14407. [ 0.3100, -0.5452, 1.8769, -0.2250, -0.1633, 0.1682, 0.5864, 0.1355]],
  14408. device='cuda:0', grad_fn=<AddmmBackward>)
  14409. landmarks are: tensor([[[ 0.5790, -0.3940, 1.8076, -0.3152, -0.6520, -0.3844, 0.4970,
  14410. 0.3238],
  14411. [ 0.5275, -0.4396, 1.5543, -0.2841, -0.5365, -0.1843, 0.1221,
  14412. 0.3265],
  14413. [ 0.5886, -0.4253, 1.8654, -0.3460, -0.5480, 0.3623, 0.6587,
  14414. 0.1005],
  14415. [ 0.5902, -0.3493, 1.7961, -0.0072, -0.5942, -0.5615, 0.3180,
  14416. 0.3161],
  14417. [ 0.5633, -0.3883, 0.8648, -1.0288, -0.5192, -1.1081, 0.2776,
  14418. 0.3546],
  14419. [ 0.5786, -0.4141, 1.7037, 0.1544, -0.1862, 0.0736, 0.4393,
  14420. 0.0851],
  14421. [ 0.6135, -0.3841, 1.8654, -0.5153, -0.4614, -1.0619, 0.6195,
  14422. -0.0049],
  14423. [ 0.5477, -0.3851, 1.7961, -0.1304, -0.3055, 0.5085, 0.3830,
  14424. 0.0682]]], device='cuda:0')
  14425. loss_train_step before backward: tensor(0.0240, device='cuda:0', grad_fn=<MseLossBackward>)
  14426. loss_train_step after backward: tensor(0.0240, device='cuda:0', grad_fn=<MseLossBackward>)
  14427. loss_train: 2.7581892935559154
  14428. step: 64
  14429. running loss: 0.04309670771181118
  14430. Train Steps: 64/90 Loss: 0.0431 torch.Size([8, 600, 800])
  14431. torch.Size([8, 8])
  14432. tensor([[0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374],
  14433. [0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933],
  14434. [0.6148, 0.4076, 0.8666, 0.4820, 0.4138, 0.5067, 0.5250, 0.5767],
  14435. [0.6239, 0.4206, 0.8750, 0.5400, 0.3688, 0.4850, 0.5737, 0.5700],
  14436. [0.6256, 0.4199, 0.8638, 0.5800, 0.3987, 0.4383, 0.5600, 0.5950],
  14437. [0.6289, 0.4032, 0.8419, 0.5446, 0.4075, 0.5017, 0.6312, 0.5117],
  14438. [0.6293, 0.4024, 0.8750, 0.5000, 0.4012, 0.5733, 0.7121, 0.5633],
  14439. [0.6223, 0.4028, 0.8988, 0.4200, 0.3763, 0.5733, 0.6375, 0.5167]],
  14440. device='cuda:0', dtype=torch.float64)
  14441. predictions are: tensor([[ 0.5196, -0.4510, 1.7051, -0.8963, -0.5621, -1.1083, 0.6279, 0.0341],
  14442. [ 0.4282, -0.4812, 1.6406, -0.0781, 0.0335, -0.6546, 0.3637, 0.3195],
  14443. [ 0.3719, -0.5328, 1.7015, -0.3336, -0.2812, -0.1746, 0.2990, 0.2221],
  14444. [ 0.5406, -0.4317, 1.7712, -0.1215, -0.4843, -0.2928, 0.5235, 0.1871],
  14445. [ 0.6402, -0.3560, 1.6619, 0.0413, -0.3322, -0.5363, 0.4270, 0.3390],
  14446. [ 0.6243, -0.4074, 1.6774, -0.2420, -0.3619, -0.3324, 0.7050, 0.0932],
  14447. [ 0.3252, -0.5740, 1.8606, -0.2420, -0.3334, 0.1080, 0.7974, 0.2020],
  14448. [ 0.1856, -0.6645, 1.8283, -0.6080, -0.4700, 0.0561, 0.7195, 0.0907]],
  14449. device='cuda:0', grad_fn=<AddmmBackward>)
  14450. landmarks are: tensor([[[ 0.5777, -0.4416, 1.7044, -0.5827, -0.5962, -0.8361, 0.4862,
  14451. 0.1963],
  14452. [ 0.5959, -0.3579, 1.6055, 0.3623, -0.0573, -0.2074, 0.3122,
  14453. 0.4547],
  14454. [ 0.5539, -0.4031, 1.7168, -0.0594, -0.3748, 0.0543, 0.1390,
  14455. 0.3777],
  14456. [ 0.5960, -0.3430, 1.7557, 0.2083, -0.5827, -0.0457, 0.3642,
  14457. 0.3469],
  14458. [ 0.6038, -0.3464, 1.7037, 0.3931, -0.4441, -0.2613, 0.3007,
  14459. 0.4624],
  14460. [ 0.6189, -0.4238, 1.6026, 0.2295, -0.4037, 0.0313, 0.6298,
  14461. 0.0774],
  14462. [ 0.6207, -0.4273, 1.7557, 0.0236, -0.4326, 0.3623, 1.0033,
  14463. 0.3157],
  14464. [ 0.5886, -0.4253, 1.8654, -0.3460, -0.5480, 0.3623, 0.6587,
  14465. 0.1005]]], device='cuda:0')
  14466. loss_train_step before backward: tensor(0.0390, device='cuda:0', grad_fn=<MseLossBackward>)
  14467. loss_train_step after backward: tensor(0.0390, device='cuda:0', grad_fn=<MseLossBackward>)
  14468. loss_train: 2.797200477682054
  14469. step: 65
  14470. running loss: 0.04303385350280083
  14471.  
  14472. Train Steps: 65/90 Loss: 0.0430 torch.Size([8, 600, 800])
  14473. torch.Size([8, 8])
  14474. tensor([[0.6203, 0.4078, 0.8800, 0.5083, 0.3900, 0.5000, 0.6100, 0.5583],
  14475. [0.6125, 0.3999, 0.8750, 0.4883, 0.4750, 0.4700, 0.5533, 0.5617],
  14476. [0.6213, 0.4131, 0.8438, 0.3550, 0.3513, 0.4400, 0.5716, 0.5123],
  14477. [ nan, nan, 0.7512, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617],
  14478. [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683],
  14479. [0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433],
  14480. [0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742],
  14481. [0.6201, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044]],
  14482. device='cuda:0', dtype=torch.float64)
  14483. predictions are: tensor([[ 0.6687, -0.3593, 1.8971, 0.1707, -0.4330, 0.0961, 0.6451, 0.1299],
  14484. [ 0.5644, -0.4138, 1.7877, 0.0097, -0.0366, -0.0669, 0.4079, 0.2327],
  14485. [ 0.7685, -0.2672, 1.7709, -0.5620, -0.6552, -0.2216, 0.4247, 0.1788],
  14486. [-1.0692, -1.4496, 1.1144, -1.0959, -0.2064, -1.3422, 0.3366, 0.3403],
  14487. [ 0.3732, -0.5458, 1.8620, 0.0651, -0.0702, -0.0228, 0.4075, 0.1777],
  14488. [ 0.4668, -0.4828, 1.4026, -0.9949, -0.5788, -0.5876, 0.6636, 0.2020],
  14489. [ 0.5879, -0.4394, 1.9501, 0.2348, -0.5701, -0.4968, 0.7669, 0.0505],
  14490. [ 0.9447, -0.1949, 1.2377, -1.2274, -0.2411, -1.2837, 0.5303, 0.2160]],
  14491. device='cuda:0', grad_fn=<AddmmBackward>)
  14492. landmarks are: tensor([[[ 0.5794, -0.4023, 1.7788, 0.0620, -0.4845, 0.0236, 0.5316,
  14493. 0.2930],
  14494. [ 0.5432, -0.4388, 1.7557, -0.0303, -0.0919, -0.1150, 0.2699,
  14495. 0.3087],
  14496. [ 0.5841, -0.3778, 1.6113, -0.6462, -0.6635, -0.2536, 0.3542,
  14497. 0.0802],
  14498. [-2.2859, -2.2859, 1.1841, -1.3082, -0.3055, -1.3621, 0.3007,
  14499. 0.3084],
  14500. [ 0.5359, -0.4370, 1.7095, -0.0303, -0.0804, -0.0380, 0.1044,
  14501. 0.3392],
  14502. [ 0.5665, -0.3963, 1.3457, -0.9926, -0.6058, -0.6308, 0.5721,
  14503. 0.2237],
  14504. [ 0.6113, -0.4057, 1.7499, 0.3007, -0.5885, -0.5384, 0.4513,
  14505. -0.0957],
  14506. [ 0.5785, -0.4129, 1.2785, -1.3996, -0.3227, -1.3259, 0.4258,
  14507. 0.0438]]], device='cuda:0')
  14508. loss_train_step before backward: tensor(0.0485, device='cuda:0', grad_fn=<MseLossBackward>)
  14509. loss_train_step after backward: tensor(0.0485, device='cuda:0', grad_fn=<MseLossBackward>)
  14510. loss_train: 2.845730713568628
  14511. step: 66
  14512. running loss: 0.04311713202376709
  14513. Train Steps: 66/90 Loss: 0.0431 torch.Size([8, 600, 800])
  14514. torch.Size([8, 8])
  14515. tensor([[0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283],
  14516. [0.6284, 0.4127, 0.8538, 0.5867, 0.4363, 0.5083, 0.6038, 0.5433],
  14517. [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332],
  14518. [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150],
  14519. [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413],
  14520. [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461],
  14521. [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5787, 0.5117],
  14522. [0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528]],
  14523. device='cuda:0', dtype=torch.float64)
  14524. predictions are: tensor([[ 0.3285, -0.5892, 1.8296, -0.5351, -0.5713, -0.4515, 0.6056, 0.1702],
  14525. [ 0.1763, -0.6597, 1.6175, 0.2585, -0.3035, 0.0659, 0.3303, 0.1897],
  14526. [ 0.4622, -0.5033, 1.6857, -0.1942, -0.5921, -0.4273, 0.6031, 0.1828],
  14527. [ 0.2868, -0.5815, 1.7712, -0.0843, -0.2788, 0.4183, 0.5205, 0.2098],
  14528. [ 0.4892, -0.4744, 1.8808, -0.3870, -0.2168, -0.9448, 0.8276, 0.1716],
  14529. [ 0.4710, -0.4649, 1.6061, 0.1596, -0.1887, 0.1203, 0.1593, 0.2353],
  14530. [ 0.7560, -0.3035, 1.0503, -1.2743, -0.3153, -1.2866, 0.1925, 0.1969],
  14531. [ 0.5216, -0.4476, 1.9730, -0.6449, -0.3307, -0.5987, 0.9147, 0.1548]],
  14532. device='cuda:0', grad_fn=<AddmmBackward>)
  14533. landmarks are: tensor([[[ 0.6073, -0.4058, 1.8885, -0.4999, -0.5942, -0.4768, 0.6413,
  14534. 0.1544],
  14535. [ 0.6166, -0.3795, 1.6575, 0.4239, -0.2709, 0.0620, 0.5028,
  14536. 0.2237],
  14537. [ 0.6520, -0.4056, 1.9173, -0.0765, -0.5596, -0.4537, 0.7949,
  14538. 0.1768],
  14539. [ 0.6014, -0.3883, 1.8192, -0.1150, -0.2882, 0.4008, 0.5201,
  14540. 0.0928],
  14541. [ 0.6224, -0.4345, 1.9404, -0.2921, -0.3171, -0.8771, 1.0655,
  14542. 0.2142],
  14543. [ 0.5443, -0.3831, 1.7095, 0.1621, -0.2016, 0.1390, 0.1437,
  14544. 0.2364],
  14545. [ 0.5598, -0.4201, 1.1898, -1.3005, -0.3806, -1.3313, 0.3873,
  14546. 0.0774],
  14547. [ 0.6388, -0.3623, 1.9173, -0.7386, -0.3517, -0.5846, 1.1495,
  14548. 0.2676]]], device='cuda:0')
  14549. loss_train_step before backward: tensor(0.0175, device='cuda:0', grad_fn=<MseLossBackward>)
  14550. loss_train_step after backward: tensor(0.0175, device='cuda:0', grad_fn=<MseLossBackward>)
  14551. loss_train: 2.86319727730006
  14552. step: 67
  14553. running loss: 0.042734287720896415
  14554. Train Steps: 67/90 Loss: 0.0427 torch.Size([8, 600, 800])
  14555. torch.Size([8, 8])
  14556. tensor([[0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5787, 0.5600],
  14557. [0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6188, 0.5433],
  14558. [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901],
  14559. [0.6117, 0.4019, 0.8538, 0.4067, 0.3513, 0.3583, 0.5663, 0.5133],
  14560. [0.6268, 0.4029, 0.8500, 0.2683, 0.3937, 0.3500, 0.6860, 0.5297],
  14561. [0.6278, 0.4253, 0.8875, 0.5017, 0.4113, 0.2750, 0.5413, 0.6196],
  14562. [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726],
  14563. [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960]],
  14564. device='cuda:0', dtype=torch.float64)
  14565. predictions are: tensor([[ 0.4266, -0.5112, 1.7698, 0.0365, -0.0220, 0.1191, 0.4562, 0.2976],
  14566. [ 0.4087, -0.5572, 1.9169, 0.0073, -0.5026, 0.2506, 0.6863, 0.1113],
  14567. [ 0.1075, -0.7343, 1.3653, -0.8740, -0.4800, -0.9601, 0.2076, 0.2042],
  14568. [ 0.3999, -0.5319, 1.7445, -0.3835, -0.6190, -0.5022, 0.2338, 0.1659],
  14569. [ 0.5842, -0.4576, 1.6458, -0.9018, -0.4912, -0.6226, 1.0400, 0.1461],
  14570. [ 0.7194, -0.3208, 1.6277, 0.1032, -0.2578, -0.7793, 0.5120, 0.4292],
  14571. [ 0.5912, -0.4189, 1.2441, -1.0201, -0.4948, -0.9039, 0.3610, 0.1170],
  14572. [ 0.3899, -0.5879, 1.9004, 0.0208, -0.1882, 0.1316, 0.4283, 0.0949]],
  14573. device='cuda:0', grad_fn=<AddmmBackward>)
  14574. landmarks are: tensor([[[ 0.5505, -0.4207, 1.7095, -0.0534, -0.0509, 0.1050, 0.3873,
  14575. 0.3007],
  14576. [ 0.5773, -0.4105, 1.8192, -0.1304, -0.5307, 0.0467, 0.5721,
  14577. 0.2237],
  14578. [ 0.5515, -0.4129, 1.4785, -1.0080, -0.5192, -1.1004, 0.1034,
  14579. -0.0220],
  14580. [ 0.5393, -0.4294, 1.6575, -0.4075, -0.6635, -0.6308, 0.3296,
  14581. 0.0851],
  14582. [ 0.6092, -0.4249, 1.6402, -1.0465, -0.4672, -0.6693, 0.8827,
  14583. 0.1608],
  14584. [ 0.6139, -0.3216, 1.8134, 0.0313, -0.3864, -1.0157, 0.2144,
  14585. 0.5762],
  14586. [ 0.4997, -0.4446, 1.1610, -0.9772, -0.6058, -1.0311, 0.1404,
  14587. -0.1031],
  14588. [ 0.5205, -0.4577, 1.7326, -0.1304, -0.1862, 0.0447, 0.1301,
  14589. 0.0051]]], device='cuda:0')
  14590. loss_train_step before backward: tensor(0.0202, device='cuda:0', grad_fn=<MseLossBackward>)
  14591. loss_train_step after backward: tensor(0.0202, device='cuda:0', grad_fn=<MseLossBackward>)
  14592. loss_train: 2.883371683768928
  14593. step: 68
  14594. running loss: 0.042402524761307764
  14595. Train Steps: 68/90 Loss: 0.0424 torch.Size([8, 600, 800])
  14596. torch.Size([8, 8])
  14597. tensor([[0.6266, 0.4067, 0.8588, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267],
  14598. [ nan, nan, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650],
  14599. [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426],
  14600. [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667],
  14601. [ nan, nan, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533],
  14602. [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374],
  14603. [0.6198, 0.3997, 0.8582, 0.5361, 0.4117, 0.5016, 0.5942, 0.5134],
  14604. [0.6222, 0.4169, 0.8638, 0.5650, 0.4313, 0.4783, 0.5637, 0.5633]],
  14605. device='cuda:0', dtype=torch.float64)
  14606. predictions are: tensor([[ 1.0256, -0.1258, 1.8074, -0.7311, -0.2773, -0.7034, 0.7876, 0.1885],
  14607. [-1.4408, -1.7286, 1.1651, -1.0770, -0.1554, -0.9629, 0.2680, 0.3037],
  14608. [ 0.9199, -0.2048, 1.3228, -1.0172, -0.2033, -1.1164, 0.5515, 0.1754],
  14609. [ 1.0787, -0.1060, 1.6983, -0.5107, -0.6539, -0.3158, 0.4087, 0.2556],
  14610. [-1.2559, -1.5952, 1.3378, -0.7338, -0.5142, -0.7789, 0.1824, 0.2635],
  14611. [ 0.8443, -0.2714, 1.8309, -0.3378, -0.5479, -0.5533, 0.6412, 0.0626],
  14612. [ 0.9600, -0.2078, 1.8367, 0.3667, -0.4265, 0.3660, 0.4879, 0.1222],
  14613. [ 0.9529, -0.1759, 1.7623, 0.5614, -0.3391, 0.2584, 0.3591, 0.2860]],
  14614. device='cuda:0', grad_fn=<AddmmBackward>)
  14615. landmarks are: tensor([[[ 0.6084, -0.4076, 1.6806, -0.9618, -0.2998, -0.9695, 0.6356,
  14616. 0.1467],
  14617. [-2.2859, -2.2859, 0.9175, -1.3947, -0.3691, -1.2467, 0.2314,
  14618. 0.3238],
  14619. [ 0.5784, -0.4153, 1.2972, -1.2541, -0.2265, -1.4553, 0.4741,
  14620. 0.2203],
  14621. [ 0.5768, -0.3857, 1.5305, -0.7694, -0.6462, -0.6308, 0.3988,
  14622. 0.3315],
  14623. [-2.2859, -2.2859, 1.5074, -1.0388, -0.5423, -0.9849, 0.2199,
  14624. 0.2699],
  14625. [ 0.5777, -0.4416, 1.7044, -0.5827, -0.5962, -0.8361, 0.4862,
  14626. 0.1963],
  14627. [ 0.5767, -0.4396, 1.6782, 0.1905, -0.3844, 0.0308, 0.4588,
  14628. 0.0855],
  14629. [ 0.5880, -0.3605, 1.7037, 0.3238, -0.2940, -0.0765, 0.3180,
  14630. 0.3161]]], device='cuda:0')
  14631. loss_train_step before backward: tensor(0.0832, device='cuda:0', grad_fn=<MseLossBackward>)
  14632. loss_train_step after backward: tensor(0.0832, device='cuda:0', grad_fn=<MseLossBackward>)
  14633. loss_train: 2.9666087506338954
  14634. step: 69
  14635. running loss: 0.042994329719331814
  14636.  
  14637. Train Steps: 69/90 Loss: 0.0430 torch.Size([8, 600, 800])
  14638. torch.Size([8, 8])
  14639. tensor([[0.6193, 0.4050, 0.7313, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656],
  14640. [0.6339, 0.4102, 0.8588, 0.3133, 0.4425, 0.2117, 0.6417, 0.5089],
  14641. [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938],
  14642. [0.6275, 0.4024, 0.8600, 0.2283, 0.5350, 0.1800, 0.7074, 0.5413],
  14643. [0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283],
  14644. [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283],
  14645. [0.6115, 0.4005, 0.8838, 0.3867, 0.3763, 0.4700, 0.5800, 0.5550],
  14646. [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395]],
  14647. device='cuda:0', dtype=torch.float64)
  14648. predictions are: tensor([[ 0.0945, -0.7125, 1.0857, -0.9566, -0.3616, -0.9317, 0.1844, 0.2622],
  14649. [ 0.6179, -0.4146, 1.6800, -0.5893, -0.2319, -0.8998, 0.7288, 0.1170],
  14650. [ 0.4128, -0.5567, 1.8628, 0.3670, -0.6036, -0.1763, 0.3295, 0.0783],
  14651. [-0.0069, -0.8548, 1.7534, -0.8855, 0.1549, -1.0048, 0.9400, 0.2685],
  14652. [ 0.4008, -0.5571, 1.9227, -0.1961, -0.5610, -0.1070, 0.5655, 0.1825],
  14653. [ 0.7813, -0.2972, 1.7793, -0.0681, -0.6172, -0.1469, 0.3549, 0.2691],
  14654. [ 0.6439, -0.3781, 1.7482, -0.2119, -0.6423, 0.3156, 0.2430, 0.1745],
  14655. [ 0.5604, -0.4501, 1.1170, -1.0097, -0.3723, -0.8632, 0.3318, 0.2643]],
  14656. device='cuda:0', grad_fn=<AddmmBackward>)
  14657. landmarks are: tensor([[[ 0.5746, -0.4153, 1.0917, -1.1620, -0.4037, -1.3082, 0.3234,
  14658. 0.3267],
  14659. [ 0.6421, -0.3912, 1.6806, -0.8386, -0.2420, -1.3082, 0.6780,
  14660. 0.0646],
  14661. [ 0.6273, -0.4105, 1.8480, 0.1082, -0.5538, -0.5076, 0.6414,
  14662. -0.0049],
  14663. [ 0.6125, -0.4273, 1.6864, -1.2313, 0.1852, -1.4545, 0.9814,
  14664. 0.2142],
  14665. [ 0.6073, -0.4058, 1.8885, -0.4999, -0.5942, -0.4768, 0.6413,
  14666. 0.1544],
  14667. [ 0.5880, -0.3887, 1.8423, -0.3306, -0.6231, -0.5230, 0.4046,
  14668. 0.1544],
  14669. [ 0.5384, -0.4361, 1.7961, -0.4999, -0.5480, -0.1150, 0.3931,
  14670. 0.2776],
  14671. [ 0.5713, -0.4538, 1.1028, -1.3659, -0.3831, -1.1273, 0.5340,
  14672. 0.2058]]], device='cuda:0')
  14673. loss_train_step before backward: tensor(0.0494, device='cuda:0', grad_fn=<MseLossBackward>)
  14674. loss_train_step after backward: tensor(0.0494, device='cuda:0', grad_fn=<MseLossBackward>)
  14675. loss_train: 3.0159969767555594
  14676. step: 70
  14677. running loss: 0.043085671096507995
  14678. Train Steps: 70/90 Loss: 0.0431 torch.Size([8, 600, 800])
  14679. torch.Size([8, 8])
  14680. tensor([[0.6361, 0.4076, 0.8862, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297],
  14681. [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483],
  14682. [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679],
  14683. [0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461],
  14684. [0.6276, 0.4235, 0.8888, 0.5333, 0.3800, 0.3117, 0.5427, 0.6164],
  14685. [0.6270, 0.4267, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183],
  14686. [ nan, nan, 0.7525, 0.2291, 0.3838, 0.3017, 0.6050, 0.5667],
  14687. [0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117]],
  14688. device='cuda:0', dtype=torch.float64)
  14689. predictions are: tensor([[ 0.6908, -0.4021, 1.8993, 0.1808, -0.5602, -0.0866, 0.6491, 0.1600],
  14690. [ 0.3828, -0.5673, 1.8360, -0.3507, -0.4966, 0.0851, 0.3386, 0.1186],
  14691. [ 0.7606, -0.3255, 1.8966, 0.1732, -0.3291, 0.5183, 0.7372, 0.2568],
  14692. [ 0.5708, -0.4434, 1.7218, -0.5815, -0.7062, -0.3660, 0.2640, 0.0845],
  14693. [ 0.6328, -0.3785, 1.7117, 0.1811, -0.4657, -0.6712, 0.3862, 0.3782],
  14694. [ 0.5071, -0.4653, 1.1100, -1.0328, -0.0600, -1.3080, 0.1529, 0.3689],
  14695. [-0.3958, -1.1060, 1.3111, -1.2376, -0.3739, -0.9371, 0.3930, 0.2738],
  14696. [ 0.4604, -0.5685, 1.9232, -0.0375, -0.5582, -0.0750, 0.5697, 0.0881]],
  14697. device='cuda:0', grad_fn=<AddmmBackward>)
  14698. landmarks are: tensor([[[ 0.6520, -0.4032, 1.8076, 0.1852, -0.5711, -0.1381, 0.7876,
  14699. 0.1608],
  14700. [ 0.5546, -0.4620, 1.7788, -0.4229, -0.5192, -0.0226, 0.4277,
  14701. 0.2468],
  14702. [ 0.6257, -0.4249, 1.5998, 0.1236, -0.3806, 0.3084, 0.9887,
  14703. 0.3371],
  14704. [ 0.5124, -0.4483, 1.5420, -0.6231, -0.7155, -0.4152, 0.2679,
  14705. 0.2365],
  14706. [ 0.6131, -0.3299, 1.8192, 0.1775, -0.5307, -0.8463, 0.2206,
  14707. 0.5612],
  14708. [ 0.6101, -0.3152, 1.0166, -0.7540, -0.0226, -1.4468, 0.2823,
  14709. 0.5702],
  14710. [-2.2859, -2.2859, 1.1898, -1.2278, -0.5134, -0.8925, 0.5085,
  14711. 0.3315],
  14712. [ 0.6262, -0.4461, 1.8480, -0.0534, -0.5827, -0.1227, 0.6587,
  14713. 0.0774]]], device='cuda:0')
  14714. loss_train_step before backward: tensor(0.0906, device='cuda:0', grad_fn=<MseLossBackward>)
  14715. loss_train_step after backward: tensor(0.0906, device='cuda:0', grad_fn=<MseLossBackward>)
  14716. loss_train: 3.1066278582438827
  14717. step: 71
  14718. running loss: 0.043755321947096935
  14719. Train Steps: 71/90 Loss: 0.0438 torch.Size([8, 600, 800])
  14720. torch.Size([8, 8])
  14721. tensor([[0.6175, 0.3957, 0.8700, 0.4817, 0.4662, 0.5133, 0.5800, 0.5517],
  14722. [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986],
  14723. [0.6317, 0.4038, 0.8287, 0.5900, 0.3800, 0.4717, 0.6295, 0.4986],
  14724. [0.6176, 0.3911, 0.8738, 0.4217, 0.3488, 0.4033, 0.6025, 0.4817],
  14725. [ nan, nan, 0.8850, 0.3000, 0.5363, 0.2250, 0.7343, 0.5771],
  14726. [0.6125, 0.4035, 0.7825, 0.3100, 0.3463, 0.4900, 0.5832, 0.5637],
  14727. [0.6138, 0.4101, 0.8800, 0.5083, 0.4637, 0.5950, 0.5587, 0.5077],
  14728. [0.6224, 0.4097, 0.7438, 0.2267, 0.3850, 0.2850, 0.5988, 0.5250]],
  14729. device='cuda:0', dtype=torch.float64)
  14730. predictions are: tensor([[ 0.5412, -0.4559, 1.7843, 0.0259, -0.2841, 0.1660, 0.2377, 0.1969],
  14731. [ 0.6741, -0.3356, 1.8233, -0.4519, -0.2874, -1.1560, 0.4323, 0.2217],
  14732. [ 0.7896, -0.3125, 1.7426, 0.3888, -0.6745, -0.0785, 0.5314, 0.1593],
  14733. [ 0.6071, -0.4343, 1.7290, -0.3111, -0.7544, -0.4238, 0.3074, 0.0894],
  14734. [-1.1140, -1.5608, 1.7046, -0.9713, 0.0438, -1.1867, 0.8088, 0.3351],
  14735. [ 0.4090, -0.5324, 1.3724, -0.7669, -0.7136, -0.0075, 0.4359, 0.2871],
  14736. [ 0.6645, -0.3611, 1.7667, 0.1716, -0.2005, 0.3721, 0.2576, 0.2559],
  14737. [ 0.7018, -0.3551, 1.2018, -1.1815, -0.5421, -0.9914, 0.5265, 0.1972]],
  14738. device='cuda:0', grad_fn=<AddmmBackward>)
  14739. landmarks are: tensor([[[ 0.5662, -0.4581, 1.7326, -0.0611, -0.1323, 0.0851, 0.3931,
  14740. 0.2622],
  14741. [ 0.6054, -0.3641, 1.8711, -0.5153, -0.2305, -1.3005, 0.5948,
  14742. 0.0171],
  14743. [ 0.6320, -0.4206, 1.5420, 0.4393, -0.5307, -0.1073, 0.6216,
  14744. 0.0171],
  14745. [ 0.5669, -0.4794, 1.7499, -0.3383, -0.6751, -0.4229, 0.4970,
  14746. -0.0611],
  14747. [-2.2859, -2.2859, 1.8018, -0.9002, 0.1910, -1.2467, 1.1057,
  14748. 0.3799],
  14749. [ 0.5432, -0.4224, 1.3284, -0.8540, -0.6866, -0.0226, 0.4077,
  14750. 0.3177],
  14751. [ 0.5491, -0.3918, 1.7788, 0.0620, -0.1439, 0.4624, 0.2946,
  14752. 0.0592],
  14753. [ 0.5891, -0.3937, 1.1494, -1.2390, -0.5076, -0.9695, 0.4797,
  14754. 0.1390]]], device='cuda:0')
  14755. loss_train_step before backward: tensor(0.0398, device='cuda:0', grad_fn=<MseLossBackward>)
  14756. loss_train_step after backward: tensor(0.0398, device='cuda:0', grad_fn=<MseLossBackward>)
  14757. loss_train: 3.146448607556522
  14758. step: 72
  14759. running loss: 0.043700675104951694
  14760. Train Steps: 72/90 Loss: 0.0437 torch.Size([8, 600, 800])
  14761. torch.Size([8, 8])
  14762. tensor([[0.6250, 0.4236, 0.8638, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799],
  14763. [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209],
  14764. [0.6265, 0.4091, 0.8950, 0.3533, 0.3600, 0.3967, 0.6295, 0.4901],
  14765. [0.6115, 0.4005, 0.8838, 0.3867, 0.3763, 0.4700, 0.5800, 0.5550],
  14766. [0.6098, 0.3991, 0.8638, 0.4717, 0.4263, 0.4967, 0.5212, 0.5650],
  14767. [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767],
  14768. [ nan, nan, 0.6900, 0.1917, 0.3937, 0.2367, 0.5240, 0.5246],
  14769. [0.6273, 0.4100, 0.7137, 0.2133, 0.4000, 0.2650, 0.6075, 0.5633]],
  14770. device='cuda:0', dtype=torch.float64)
  14771. predictions are: tensor([[ 0.5118, -0.4524, 1.7195, -0.3490, -0.4138, -0.5709, 0.4172, 0.3501],
  14772. [ 1.0943, -0.0776, 1.6530, -0.0977, -0.2364, -0.7991, 0.5785, 0.3261],
  14773. [ 0.6776, -0.4029, 1.9237, -0.5634, -0.6072, -0.3619, 0.7648, 0.0746],
  14774. [ 1.0923, -0.1342, 1.7725, -0.4052, -0.6281, 0.1162, 0.4874, 0.1301],
  14775. [ 0.6887, -0.3659, 1.7554, 0.0926, -0.3180, 0.2316, 0.3344, 0.1946],
  14776. [ 0.6969, -0.3097, 1.6428, 0.0207, -0.5718, -0.5985, 0.2854, 0.2422],
  14777. [-2.1434, -2.2292, 1.0703, -1.2591, -0.3749, -1.2681, 0.2250, 0.1873],
  14778. [ 0.3197, -0.6014, 1.0105, -1.1921, -0.3945, -1.0479, 0.4393, 0.2994]],
  14779. device='cuda:0', grad_fn=<AddmmBackward>)
  14780. landmarks are: tensor([[[ 0.6008, -0.3293, 1.7037, -0.5461, -0.4152, -0.8309, 0.3234,
  14781. 0.3928],
  14782. [ 0.6174, -0.3118, 1.6402, -0.2074, -0.1958, -1.0927, 0.2267,
  14783. 0.5822],
  14784. [ 0.6077, -0.3965, 1.8480, -0.6539, -0.6231, -0.4537, 0.6216,
  14785. -0.0220],
  14786. [ 0.5384, -0.4361, 1.7961, -0.4999, -0.5480, -0.1150, 0.3931,
  14787. 0.2776],
  14788. [ 0.5308, -0.4425, 1.7037, -0.1073, -0.3171, 0.0082, 0.1217,
  14789. 0.3238],
  14790. [ 0.5781, -0.3848, 1.7441, -0.0996, -0.5769, -0.8002, 0.2021,
  14791. 0.3778],
  14792. [-2.2859, -2.2859, 0.9012, -1.4006, -0.4672, -1.1928, 0.1342,
  14793. 0.1373],
  14794. [ 0.6115, -0.3924, 1.0109, -1.3005, -0.4383, -1.0619, 0.5201,
  14795. 0.3161]]], device='cuda:0')
  14796. loss_train_step before backward: tensor(0.0283, device='cuda:0', grad_fn=<MseLossBackward>)
  14797. loss_train_step after backward: tensor(0.0283, device='cuda:0', grad_fn=<MseLossBackward>)
  14798. loss_train: 3.174727371893823
  14799. step: 73
  14800. running loss: 0.043489416053340044
  14801.  
  14802. Train Steps: 73/90 Loss: 0.0435 torch.Size([8, 600, 800])
  14803. torch.Size([8, 8])
  14804. tensor([[0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483],
  14805. [0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136],
  14806. [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209],
  14807. [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967],
  14808. [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869],
  14809. [0.6124, 0.4069, 0.8314, 0.5001, 0.3738, 0.4650, 0.5167, 0.5402],
  14810. [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274],
  14811. [0.6289, 0.4019, 0.8113, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332]],
  14812. device='cuda:0', dtype=torch.float64)
  14813. predictions are: tensor([[ 0.5926, -0.4584, 1.1286, -1.4791, -0.3460, -1.2189, 0.5719, 0.2300],
  14814. [ 0.2274, -0.6594, 1.7152, -0.5702, -0.6758, -0.4601, 0.1921, 0.1128],
  14815. [ 0.6375, -0.3545, 1.6312, -0.4032, -0.2160, -0.9396, 0.4620, 0.3717],
  14816. [ 0.2176, -0.6772, 1.8095, -0.1802, -0.4826, -0.6983, 0.5700, 0.1636],
  14817. [ 0.0604, -0.7763, 1.6917, 0.1112, -0.4883, -0.2882, 0.1321, 0.1681],
  14818. [ 0.4266, -0.5131, 1.5818, -0.1841, -0.4556, -0.1109, 0.2273, 0.2884],
  14819. [ 0.5517, -0.4476, 1.8382, -0.2390, -0.5477, -0.2693, 0.7168, 0.2239],
  14820. [ 0.4387, -0.5354, 1.6025, -0.0815, -0.5076, 0.0810, 0.6287, 0.2390]],
  14821. device='cuda:0', grad_fn=<AddmmBackward>)
  14822. landmarks are: tensor([[[ 0.5784, -0.4085, 1.0859, -1.3929, -0.4037, -1.1158, 0.5605,
  14823. 0.2468],
  14824. [ 0.5235, -0.4273, 1.7499, -0.4306, -0.5827, -0.4614, 0.1651,
  14825. 0.0862],
  14826. [ 0.6174, -0.3118, 1.6402, -0.2074, -0.1958, -1.0927, 0.2267,
  14827. 0.5822],
  14828. [ 0.6054, -0.3767, 1.7788, 0.0774, -0.5711, -0.7694, 0.5887,
  14829. 0.0081],
  14830. [ 0.5320, -0.4488, 1.6633, 0.3315, -0.5018, -0.2459, 0.0828,
  14831. -0.0370],
  14832. [ 0.5425, -0.4067, 1.5543, 0.0241, -0.5596, -0.1381, 0.1005,
  14833. 0.2093],
  14834. [ 0.6108, -0.4008, 1.9088, -0.0253, -0.5769, -0.3075, 0.7905,
  14835. 0.1499],
  14836. [ 0.6191, -0.4297, 1.4612, 0.2391, -0.4961, 0.0313, 1.1166,
  14837. 0.1768]]], device='cuda:0')
  14838. loss_train_step before backward: tensor(0.0268, device='cuda:0', grad_fn=<MseLossBackward>)
  14839. loss_train_step after backward: tensor(0.0268, device='cuda:0', grad_fn=<MseLossBackward>)
  14840. loss_train: 3.2015545638278127
  14841. step: 74
  14842. running loss: 0.04326425086253801
  14843. Train Steps: 74/90 Loss: 0.0433 torch.Size([8, 600, 800])
  14844. torch.Size([8, 8])
  14845. tensor([[0.6153, 0.4117, 0.8688, 0.5167, 0.4895, 0.5647, 0.5524, 0.5136],
  14846. [0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400],
  14847. [0.6256, 0.4199, 0.8638, 0.5800, 0.3987, 0.4383, 0.5600, 0.5950],
  14848. [0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947],
  14849. [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533],
  14850. [0.6257, 0.4034, 0.8287, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901],
  14851. [0.6250, 0.4106, 0.8700, 0.3717, 0.3588, 0.4967, 0.6038, 0.5167],
  14852. [0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274]],
  14853. device='cuda:0', dtype=torch.float64)
  14854. predictions are: tensor([[ 0.4025, -0.5363, 1.6971, 0.1246, -0.1701, 0.1078, 0.2141, 0.2565],
  14855. [ 0.4167, -0.5550, 1.7186, -0.4311, -0.7491, -0.2282, 0.5160, 0.2378],
  14856. [ 0.3892, -0.5351, 1.5867, 0.4916, -0.4973, -0.3103, 0.2211, 0.4570],
  14857. [-0.0709, -0.8842, 1.3462, -1.3077, -0.3917, -1.4341, 0.6078, 0.1435],
  14858. [ 0.5810, -0.4226, 1.7171, -0.1685, -0.4232, 0.1927, 0.2545, 0.2298],
  14859. [ 0.3430, -0.6123, 1.4879, -1.0594, -0.5487, -1.2002, 0.5690, 0.0772],
  14860. [ 0.7413, -0.3250, 1.6692, -0.4937, -0.6821, 0.0267, 0.6439, 0.2146],
  14861. [ 0.1623, -0.6903, 1.6709, -0.8004, -0.1233, -1.2627, 0.6041, 0.2084]],
  14862. device='cuda:0', grad_fn=<AddmmBackward>)
  14863. landmarks are: tensor([[[ 5.5635e-01, -3.8422e-01, 1.7268e+00, 1.0054e-01, -2.4997e-02,
  14864. 3.2255e-01, 2.6581e-01, 8.6245e-02],
  14865. [ 5.7783e-01, -4.3934e-01, 1.8018e+00, -4.6143e-01, -6.6928e-01,
  14866. -1.3811e-01, 5.4896e-01, 2.0831e-01],
  14867. [ 6.0381e-01, -3.4642e-01, 1.7037e+00, 3.9307e-01, -4.4411e-01,
  14868. -2.6128e-01, 3.0069e-01, 4.6236e-01],
  14869. [ 6.0739e-01, -4.0631e-01, 1.3630e+00, -1.5238e+00, -3.1132e-01,
  14870. -1.2390e+00, 6.1950e-01, -9.2270e-04],
  14871. [ 5.4990e-01, -4.2249e-01, 1.8018e+00, -2.9207e-01, -3.0554e-01,
  14872. 5.4350e-02, 4.0462e-01, 2.6990e-01],
  14873. [ 6.0421e-01, -4.2248e-01, 1.5420e+00, -1.2082e+00, -4.7298e-01,
  14874. -1.0311e+00, 6.3800e-01, -2.1963e-02],
  14875. [ 6.0098e-01, -3.8961e-01, 1.7326e+00, -5.6921e-01, -6.2887e-01,
  14876. 8.1601e-03, 5.0277e-01, 1.0054e-01],
  14877. [ 6.5365e-01, -3.7194e-01, 1.6979e+00, -8.6174e-01, -1.6859e-02,
  14878. -1.3621e+00, 6.9257e-01, 1.5008e-01]]], device='cuda:0')
  14879. loss_train_step before backward: tensor(0.0287, device='cuda:0', grad_fn=<MseLossBackward>)
  14880. loss_train_step after backward: tensor(0.0287, device='cuda:0', grad_fn=<MseLossBackward>)
  14881. loss_train: 3.2302527902647853
  14882. step: 75
  14883. running loss: 0.04307003720353047
  14884. Train Steps: 75/90 Loss: 0.0431 torch.Size([8, 600, 800])
  14885. torch.Size([8, 8])
  14886. tensor([[ nan, nan, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563],
  14887. [0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5737, 0.5350],
  14888. [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726],
  14889. [0.6261, 0.3987, 0.8688, 0.4917, 0.4300, 0.5333, 0.7010, 0.5309],
  14890. [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320],
  14891. [0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240],
  14892. [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301],
  14893. [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290]],
  14894. device='cuda:0', dtype=torch.float64)
  14895. predictions are: tensor([[-1.8661, -2.0538, 1.2785, -0.8249, -0.6488, -0.8578, 0.1658, 0.2023],
  14896. [ 0.6903, -0.3068, 1.8189, 0.1420, -0.4976, -0.8075, 0.3031, 0.2389],
  14897. [ 0.7957, -0.2790, 1.3134, -0.9248, -0.5435, -0.9453, 0.2031, 0.1206],
  14898. [ 0.9262, -0.2276, 1.8351, 0.1683, -0.3136, 0.3725, 0.7447, 0.2883],
  14899. [ 0.6961, -0.3637, 1.4982, -1.0795, -0.2947, -1.1262, 0.6684, 0.2113],
  14900. [-0.1521, -0.8515, 1.1088, -1.1953, -0.3039, -1.4054, 0.2218, 0.2412],
  14901. [ 0.8814, -0.2470, 1.7744, 0.1652, -0.3566, 0.0847, 0.4990, 0.2598],
  14902. [ 0.7957, -0.3254, 1.6967, -0.7133, -0.7268, -0.3437, 0.6567, 0.1718]],
  14903. device='cuda:0', grad_fn=<AddmmBackward>)
  14904. landmarks are: tensor([[[-2.2859, -2.2859, 1.4006, -0.8105, -0.6115, -0.8232, 0.0419,
  14905. 0.2837],
  14906. [ 0.5901, -0.4000, 1.8423, -0.0688, -0.5307, -0.9233, 0.3642,
  14907. 0.1852],
  14908. [ 0.4997, -0.4446, 1.1610, -0.9772, -0.6058, -1.0311, 0.1404,
  14909. -0.1031],
  14910. [ 0.6059, -0.4442, 1.7268, -0.0149, -0.2998, 0.1775, 0.9521,
  14911. 0.1661],
  14912. [ 0.6141, -0.4153, 1.4208, -1.2697, -0.2940, -1.0234, 0.8644,
  14913. 0.1715],
  14914. [ 0.5676, -0.4112, 1.1898, -1.2467, -0.2940, -1.4622, 0.2103,
  14915. 0.1343],
  14916. [ 0.6010, -0.4562, 1.7198, -0.0090, -0.3464, 0.0108, 0.6294,
  14917. 0.1627],
  14918. [ 0.5697, -0.4538, 1.5308, -0.8703, -0.6572, -0.3639, 0.5739,
  14919. 0.1576]]], device='cuda:0')
  14920. loss_train_step before backward: tensor(0.0314, device='cuda:0', grad_fn=<MseLossBackward>)
  14921. loss_train_step after backward: tensor(0.0314, device='cuda:0', grad_fn=<MseLossBackward>)
  14922. loss_train: 3.2616537315770984
  14923. step: 76
  14924. running loss: 0.042916496468119715
  14925. Train Steps: 76/90 Loss: 0.0429 torch.Size([8, 600, 800])
  14926. torch.Size([8, 8])
  14927. tensor([[0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938],
  14928. [0.6228, 0.4004, 0.8750, 0.5250, 0.3825, 0.5233, 0.6362, 0.5000],
  14929. [0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947],
  14930. [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320],
  14931. [0.6133, 0.4066, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358],
  14932. [0.6081, 0.3950, 0.8538, 0.4667, 0.3850, 0.4917, 0.5342, 0.4954],
  14933. [0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494],
  14934. [0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6063, 0.6083]],
  14935. device='cuda:0', dtype=torch.float64)
  14936. predictions are: tensor([[ 0.5106, -0.4525, 1.8837, 0.1418, -0.5735, -0.5365, 0.4087, 0.0735],
  14937. [ 0.7139, -0.3620, 1.7725, 0.1793, -0.4467, 0.1089, 0.4679, 0.1903],
  14938. [ 0.2135, -0.6826, 1.3965, -1.3342, -0.2621, -1.3666, 0.6790, 0.1654],
  14939. [ 0.5085, -0.4804, 1.4542, -1.1903, -0.2808, -1.1662, 0.6581, 0.2043],
  14940. [ 0.0263, -0.7549, 0.9504, -1.1005, -0.4628, -1.2502, 0.0552, 0.2227],
  14941. [ 0.5397, -0.4609, 1.7287, -0.2117, -0.5355, -0.0586, 0.2339, 0.0983],
  14942. [ 0.2517, -0.6470, 1.7428, 0.1414, -0.4972, -0.3696, 0.7320, 0.2673],
  14943. [ 0.2752, -0.5864, 1.6242, -0.6104, -0.6357, -0.2963, 0.3731, 0.3826]],
  14944. device='cuda:0', grad_fn=<AddmmBackward>)
  14945. landmarks are: tensor([[[ 6.2730e-01, -4.1045e-01, 1.8480e+00, 1.0824e-01, -5.5381e-01,
  14946. -5.0762e-01, 6.4140e-01, -4.8817e-03],
  14947. [ 5.9082e-01, -4.3664e-01, 1.7557e+00, 1.3903e-01, -5.1917e-01,
  14948. 1.3133e-01, 6.5289e-01, 2.3557e-02],
  14949. [ 6.0739e-01, -4.0631e-01, 1.3630e+00, -1.5238e+00, -3.1132e-01,
  14950. -1.2390e+00, 6.1950e-01, -9.2270e-04],
  14951. [ 6.1413e-01, -4.1527e-01, 1.4208e+00, -1.2697e+00, -2.9400e-01,
  14952. -1.0234e+00, 8.6439e-01, 1.7146e-01],
  14953. [ 5.4700e-01, -4.0808e-01, 8.4919e-01, -1.0773e+00, -5.3072e-01,
  14954. -1.1620e+00, 9.1240e-02, 1.8903e-01],
  14955. [ 5.2269e-01, -4.6151e-01, 1.6575e+00, -1.3041e-01, -5.0762e-01,
  14956. -1.4935e-02, 1.8150e-01, 2.0831e-03],
  14957. [ 6.2236e-01, -4.0323e-01, 1.5940e+00, 2.9299e-01, -5.7691e-01,
  14958. -2.6898e-01, 8.8998e-01, 2.5161e-01],
  14959. [ 5.7742e-01, -3.8684e-01, 1.6286e+00, -5.6921e-01, -6.4619e-01,
  14960. -2.7667e-01, 5.1432e-01, 5.2394e-01]]], device='cuda:0')
  14961. loss_train_step before backward: tensor(0.0231, device='cuda:0', grad_fn=<MseLossBackward>)
  14962. loss_train_step after backward: tensor(0.0231, device='cuda:0', grad_fn=<MseLossBackward>)
  14963. loss_train: 3.2847120529040694
  14964. step: 77
  14965. running loss: 0.04265859808966324
  14966.  
  14967. Train Steps: 77/90 Loss: 0.0427 torch.Size([8, 600, 800])
  14968. torch.Size([8, 8])
  14969. tensor([[0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5737, 0.5350],
  14970. [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567],
  14971. [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083],
  14972. [0.6212, 0.4159, 0.8675, 0.5783, 0.4088, 0.4317, 0.5613, 0.5917],
  14973. [0.6128, 0.4022, 0.8738, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064],
  14974. [0.6122, 0.4006, 0.8850, 0.4217, 0.4088, 0.5517, 0.6063, 0.5517],
  14975. [0.6175, 0.3957, 0.8700, 0.4817, 0.4662, 0.5133, 0.5800, 0.5517],
  14976. [0.6185, 0.4079, 0.8838, 0.4617, 0.4838, 0.5650, 0.6175, 0.5850]],
  14977. device='cuda:0', dtype=torch.float64)
  14978. predictions are: tensor([[ 0.3915, -0.5025, 1.7193, -0.3797, -0.6087, -1.2569, 0.5107, 0.0821],
  14979. [ 0.4523, -0.4895, 1.5831, -0.5257, -0.6798, -0.7597, 0.3999, 0.1821],
  14980. [ 0.3838, -0.5686, 1.6145, -0.2962, -0.6730, -0.5379, 0.5283, 0.3127],
  14981. [ 0.1110, -0.6749, 1.5092, -0.0243, -0.5125, -0.6090, 0.4088, 0.3631],
  14982. [ 0.3467, -0.5665, 1.6822, -0.2644, -0.0797, -0.4045, 0.2067, 0.1177],
  14983. [ 0.7551, -0.3455, 1.6826, -0.6358, -0.4583, 0.0169, 0.7260, 0.1225],
  14984. [ 0.3825, -0.5697, 1.6859, -0.4049, -0.2606, -0.2820, 0.3845, 0.1126],
  14985. [ 0.4185, -0.5269, 1.6859, -0.5249, -0.1563, -0.1265, 0.5731, 0.2411]],
  14986. device='cuda:0', grad_fn=<AddmmBackward>)
  14987. landmarks are: tensor([[[ 0.5901, -0.4000, 1.8423, -0.0688, -0.5307, -0.9233, 0.3642,
  14988. 0.1852],
  14989. [ 0.5432, -0.4336, 1.7095, -0.1766, -0.5942, -0.4845, 0.3007,
  14990. 0.2853],
  14991. [ 0.5713, -0.4105, 1.7557, 0.0467, -0.6520, -0.2690, 0.3988,
  14992. 0.5239],
  14993. [ 0.5836, -0.3649, 1.7210, 0.3854, -0.3979, -0.2921, 0.3065,
  14994. 0.4470],
  14995. [ 0.5446, -0.4280, 1.7499, 0.0543, 0.0156, 0.1301, 0.1918,
  14996. 0.0532],
  14997. [ 0.5417, -0.4355, 1.8018, -0.3383, -0.3979, 0.2622, 0.5143,
  14998. 0.2622],
  14999. [ 0.5662, -0.4581, 1.7326, -0.0611, -0.1323, 0.0851, 0.3931,
  15000. 0.2622],
  15001. [ 0.5707, -0.4017, 1.7961, -0.1535, -0.0515, 0.3238, 0.5663,
  15002. 0.4162]]], device='cuda:0')
  15003. loss_train_step before backward: tensor(0.0481, device='cuda:0', grad_fn=<MseLossBackward>)
  15004. loss_train_step after backward: tensor(0.0481, device='cuda:0', grad_fn=<MseLossBackward>)
  15005. loss_train: 3.33284895028919
  15006. step: 78
  15007. running loss: 0.042728832696015254
  15008. Train Steps: 78/90 Loss: 0.0427 torch.Size([8, 600, 800])
  15009. torch.Size([8, 8])
  15010. tensor([[0.6176, 0.3911, 0.8738, 0.4217, 0.3488, 0.4033, 0.6025, 0.4817],
  15011. [0.6127, 0.4118, 0.8650, 0.5083, 0.4088, 0.5367, 0.5300, 0.5456],
  15012. [ nan, nan, 0.6793, 0.2110, 0.4012, 0.2167, 0.5112, 0.5583],
  15013. [0.6147, 0.4081, 0.8538, 0.3400, 0.3663, 0.3150, 0.5142, 0.4875],
  15014. [0.6143, 0.4034, 0.8800, 0.4833, 0.4512, 0.5367, 0.5289, 0.5097],
  15015. [ nan, nan, 0.7097, 0.2346, 0.4250, 0.1850, 0.5175, 0.5583],
  15016. [ nan, nan, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633],
  15017. [0.6300, 0.4133, 0.8538, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413]],
  15018. device='cuda:0', dtype=torch.float64)
  15019. predictions are: tensor([[ 1.4889, 0.1334, 1.8088, -0.1667, -0.6862, -0.3736, 0.5403, 0.0498],
  15020. [ 1.3320, 0.0668, 1.7552, 0.3750, -0.3791, 0.2351, 0.4607, 0.2350],
  15021. [-1.3916, -1.6578, 0.9473, -1.2229, -0.4628, -1.2002, 0.2467, 0.2536],
  15022. [ 0.9997, -0.1110, 1.7093, -0.5094, -0.6704, -0.7536, 0.3662, 0.0959],
  15023. [ 0.9938, -0.1120, 1.7601, 0.2227, -0.2643, 0.1579, 0.2913, 0.1372],
  15024. [-0.9061, -1.3098, 0.9455, -1.1612, -0.3852, -1.3221, 0.2039, 0.2837],
  15025. [-1.6012, -1.8199, 1.1570, -1.0037, -0.5747, -1.0591, 0.2236, 0.2229],
  15026. [ 1.1578, -0.0632, 1.7333, -1.0920, 0.0493, -1.0529, 1.2628, 0.2168]],
  15027. device='cuda:0', grad_fn=<AddmmBackward>)
  15028. landmarks are: tensor([[[ 0.5669, -0.4794, 1.7499, -0.3383, -0.6751, -0.4229, 0.4970,
  15029. -0.0611],
  15030. [ 0.5443, -0.3840, 1.7095, 0.0620, -0.3979, 0.1929, 0.1622,
  15031. 0.2341],
  15032. [-2.2859, -2.2859, 0.8516, -1.3112, -0.4326, -1.2851, 0.0755,
  15033. 0.2930],
  15034. [ 0.5532, -0.4008, 1.6575, -0.7155, -0.5942, -0.8309, 0.0890,
  15035. -0.0340],
  15036. [ 0.5515, -0.4225, 1.7788, -0.0534, -0.2016, 0.1929, 0.1568,
  15037. 0.0682],
  15038. [-2.2859, -2.2859, 0.9922, -1.2021, -0.3229, -1.4314, 0.1044,
  15039. 0.2930],
  15040. [-2.2859, -2.2859, 1.2030, -1.0288, -0.4961, -1.1081, 0.0813,
  15041. 0.3161],
  15042. [ 0.6240, -0.3768, 1.6575, -1.2851, 0.2949, -1.2467, 1.1276,
  15043. 0.2142]]], device='cuda:0')
  15044. loss_train_step before backward: tensor(0.1332, device='cuda:0', grad_fn=<MseLossBackward>)
  15045. loss_train_step after backward: tensor(0.1332, device='cuda:0', grad_fn=<MseLossBackward>)
  15046. loss_train: 3.4660702189430594
  15047. step: 79
  15048. running loss: 0.043874306568899485
  15049. Train Steps: 79/90 Loss: 0.0439 torch.Size([8, 600, 800])
  15050. torch.Size([8, 8])
  15051. tensor([[0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626],
  15052. [0.6087, 0.3976, 0.8337, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300],
  15053. [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892],
  15054. [0.6299, 0.4008, 0.8450, 0.5350, 0.4213, 0.5000, 0.6350, 0.5100],
  15055. [0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528],
  15056. [0.6142, 0.3982, 0.8650, 0.4883, 0.3912, 0.4317, 0.5315, 0.5350],
  15057. [0.6222, 0.3957, 0.8838, 0.5017, 0.3937, 0.4600, 0.5900, 0.5017],
  15058. [0.6177, 0.4086, 0.8738, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700]],
  15059. device='cuda:0', dtype=torch.float64)
  15060. predictions are: tensor([[ 0.4458, -0.5032, 1.5592, 0.0993, -0.4399, -0.3817, 0.4723, 0.1310],
  15061. [ 0.7403, -0.2708, 1.5286, -0.8189, -0.5177, -0.9784, 0.4248, 0.1831],
  15062. [ 0.1945, -0.5920, 1.5142, -0.3926, -0.6283, -0.8031, 0.0536, 0.2479],
  15063. [ 0.6446, -0.3842, 1.5927, -0.0636, -0.3261, -0.2536, 0.5394, 0.1778],
  15064. [ 0.1360, -0.7198, 1.8964, -0.9149, -0.2398, -0.8776, 1.0256, 0.1518],
  15065. [ 0.2770, -0.6113, 1.6559, -0.1666, -0.4320, -0.4876, 0.1736, 0.1383],
  15066. [ 0.2104, -0.6595, 1.6697, -0.1950, -0.4642, -0.4355, 0.3429, 0.0977],
  15067. [ 0.4589, -0.4922, 1.5949, -0.6927, -0.3965, 0.1450, 0.6767, 0.2705]],
  15068. device='cuda:0', grad_fn=<AddmmBackward>)
  15069. landmarks are: tensor([[[ 0.6105, -0.4293, 1.5824, 0.5239, -0.4730, -0.0380, 0.5025,
  15070. -0.1492],
  15071. [ 0.5255, -0.4495, 1.5651, -0.4999, -0.5711, -0.8463, 0.4566,
  15072. 0.1621],
  15073. [ 0.5454, -0.4053, 1.6633, -0.1766, -0.6058, -0.5923, 0.1577,
  15074. 0.4357],
  15075. [ 0.6236, -0.4344, 1.6171, 0.1852, -0.3402, 0.0236, 0.6471,
  15076. 0.0697],
  15077. [ 0.6388, -0.3623, 1.9173, -0.7386, -0.3517, -0.5846, 1.1495,
  15078. 0.2676],
  15079. [ 0.5513, -0.4467, 1.7095, -0.0303, -0.4788, -0.2921, 0.1692,
  15080. 0.1854],
  15081. [ 0.5879, -0.4583, 1.7961, 0.0313, -0.4672, -0.1612, 0.4393,
  15082. 0.0313],
  15083. [ 0.5671, -0.3988, 1.7499, -0.4614, -0.5423, 0.3007, 0.5894,
  15084. 0.3469]]], device='cuda:0')
  15085. loss_train_step before backward: tensor(0.0359, device='cuda:0', grad_fn=<MseLossBackward>)
  15086. loss_train_step after backward: tensor(0.0359, device='cuda:0', grad_fn=<MseLossBackward>)
  15087. loss_train: 3.502012967132032
  15088. step: 80
  15089. running loss: 0.0437751620891504
  15090. Train Steps: 80/90 Loss: 0.0438 torch.Size([8, 600, 800])
  15091. torch.Size([8, 8])
  15092. tensor([[0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567],
  15093. [0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411],
  15094. [0.6137, 0.4038, 0.8563, 0.4050, 0.3813, 0.2550, 0.5106, 0.4954],
  15095. [0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058],
  15096. [0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945],
  15097. [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690],
  15098. [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817],
  15099. [0.6143, 0.4040, 0.8237, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973]],
  15100. device='cuda:0', dtype=torch.float64)
  15101. predictions are: tensor([[ 0.4675, -0.4482, 1.5720, -0.3319, -0.5601, -0.4890, 0.3313, 0.2283],
  15102. [ 0.4698, -0.4146, 1.3785, -1.0504, -0.3257, -1.3054, 0.3977, 0.1078],
  15103. [ 0.1245, -0.6607, 1.5465, -0.6004, -0.4580, -1.0717, 0.1884, 0.1108],
  15104. [ 0.6066, -0.3798, 1.5851, -0.5191, -0.5383, -0.1436, 0.3088, 0.0781],
  15105. [ 0.0988, -0.7262, 1.8452, -0.4678, -0.2128, -0.6663, 1.0411, 0.2591],
  15106. [ 0.5097, -0.4873, 1.5568, -0.0783, -0.3597, 0.2482, 0.9797, 0.3026],
  15107. [ 0.5244, -0.4431, 1.5430, 0.2526, -0.2678, -0.0672, 0.1750, 0.1631],
  15108. [-0.1690, -0.8363, 1.3229, -0.9650, -0.4343, -1.2110, 0.2810, 0.0549]],
  15109. device='cuda:0', grad_fn=<AddmmBackward>)
  15110. landmarks are: tensor([[[ 0.5432, -0.4336, 1.7095, -0.1766, -0.5942, -0.4845, 0.3007,
  15111. 0.2853],
  15112. [ 0.5713, -0.4321, 1.4975, -0.8134, -0.3094, -1.3345, 0.3779,
  15113. 0.2134],
  15114. [ 0.5485, -0.4209, 1.6691, -0.4152, -0.5249, -1.1081, 0.0725,
  15115. 0.0021],
  15116. [ 0.5245, -0.4361, 1.5940, -0.2921, -0.5480, -0.0919, 0.2432,
  15117. 0.0502],
  15118. [ 0.6471, -0.3719, 1.9866, -0.3921, -0.3806, -0.5538, 1.0070,
  15119. 0.4600],
  15120. [ 0.6306, -0.4153, 1.5141, 0.2224, -0.3691, 0.2622, 1.0033,
  15121. 0.3425],
  15122. [ 0.5193, -0.4699, 1.5767, 0.4008, -0.2420, 0.0774, 0.1178,
  15123. -0.0610],
  15124. [ 0.5515, -0.4201, 1.5189, -0.7463, -0.4037, -1.3082, 0.0869,
  15125. 0.0111]]], device='cuda:0')
  15126. loss_train_step before backward: tensor(0.0330, device='cuda:0', grad_fn=<MseLossBackward>)
  15127. loss_train_step after backward: tensor(0.0330, device='cuda:0', grad_fn=<MseLossBackward>)
  15128. loss_train: 3.53500113543123
  15129. step: 81
  15130. running loss: 0.043641989326311484
  15131.  
  15132. Train Steps: 81/90 Loss: 0.0436 torch.Size([8, 600, 800])
  15133. torch.Size([8, 8])
  15134. tensor([[0.6197, 0.4090, 0.7825, 0.2500, 0.4200, 0.2483, 0.5988, 0.5667],
  15135. [0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133],
  15136. [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284],
  15137. [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5988, 0.5283],
  15138. [0.6308, 0.3990, 0.8688, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133],
  15139. [0.6111, 0.4033, 0.8300, 0.3267, 0.3588, 0.3333, 0.5444, 0.5637],
  15140. [0.6085, 0.4008, 0.8588, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283],
  15141. [0.6043, 0.4022, 0.6887, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136]],
  15142. device='cuda:0', dtype=torch.float64)
  15143. predictions are: tensor([[ 0.2931, -0.5850, 1.3665, -1.0628, -0.4258, -1.0431, 0.5518, 0.2436],
  15144. [ 0.7559, -0.2991, 1.5757, -1.0323, -0.1496, -1.3116, 0.6911, 0.1442],
  15145. [ 0.7068, -0.3175, 1.2592, -1.1584, -0.2464, -1.3062, 0.4130, 0.1583],
  15146. [ 0.2576, -0.6200, 1.7675, 0.3444, -0.5498, 0.0901, 0.4331, 0.1720],
  15147. [ 0.3448, -0.5905, 1.7996, 0.2079, -0.4959, 0.1073, 0.7317, 0.1350],
  15148. [-0.2655, -0.9400, 1.5480, -0.6480, -0.6969, -0.6634, 0.2014, 0.1471],
  15149. [ 0.2986, -0.5727, 1.7204, 0.2381, -0.0618, 0.0382, 0.2197, 0.1804],
  15150. [ 0.2452, -0.6121, 1.0176, -1.1555, -0.5223, -1.0575, 0.2831, 0.1055]],
  15151. device='cuda:0', grad_fn=<AddmmBackward>)
  15152. landmarks are: tensor([[[ 0.5764, -0.3969, 1.3284, -1.1312, -0.3460, -1.1389, 0.4797,
  15153. 0.3315],
  15154. [ 0.6085, -0.4084, 1.5536, -1.1466, -0.0746, -1.4853, 0.6298,
  15155. 0.0851],
  15156. [ 0.5761, -0.4070, 1.2452, -1.2541, -0.1725, -1.4835, 0.4511,
  15157. 0.1545],
  15158. [ 0.5603, -0.4319, 1.7788, 0.1775, -0.5538, -0.0611, 0.4797,
  15159. 0.1544],
  15160. [ 0.6275, -0.4430, 1.7268, 0.1082, -0.4614, 0.0159, 0.6644,
  15161. 0.0851],
  15162. [ 0.5366, -0.4232, 1.5478, -0.7771, -0.6289, -0.7463, 0.2288,
  15163. 0.3177],
  15164. [ 0.5245, -0.4347, 1.6806, 0.1159, 0.0046, 0.0129, 0.1044,
  15165. 0.1544],
  15166. [ 0.5053, -0.4281, 0.8954, -1.3698, -0.5423, -1.1389, 0.2453,
  15167. 0.0862]]], device='cuda:0')
  15168. loss_train_step before backward: tensor(0.0293, device='cuda:0', grad_fn=<MseLossBackward>)
  15169. loss_train_step after backward: tensor(0.0293, device='cuda:0', grad_fn=<MseLossBackward>)
  15170. loss_train: 3.564255858771503
  15171. step: 82
  15172. running loss: 0.04346653486306711
  15173. Train Steps: 82/90 Loss: 0.0435 torch.Size([8, 600, 800])
  15174. torch.Size([8, 8])
  15175. tensor([[0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343],
  15176. [0.6228, 0.4119, 0.7938, 0.2233, 0.4674, 0.1773, 0.6188, 0.5433],
  15177. [ nan, nan, 0.7515, 0.2708, 0.3987, 0.2267, 0.5162, 0.5567],
  15178. [0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904],
  15179. [ nan, nan, 0.6469, 0.1943, 0.4025, 0.2000, 0.5125, 0.5533],
  15180. [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446],
  15181. [0.6153, 0.4119, 0.8463, 0.3833, 0.3600, 0.3200, 0.5106, 0.5563],
  15182. [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5963, 0.6217]],
  15183. device='cuda:0', dtype=torch.float64)
  15184. predictions are: tensor([[ 0.9934, -0.1945, 1.8464, -0.3701, -0.3761, -0.5561, 0.9081, 0.0323],
  15185. [ 1.0571, -0.0771, 1.3978, -0.8396, -0.1539, -1.1052, 0.5071, 0.1309],
  15186. [-1.6352, -1.8696, 1.1451, -0.8900, -0.4603, -0.8711, 0.1760, 0.1800],
  15187. [ 0.6482, -0.3183, 1.6749, -0.8900, -0.1365, -0.8788, 0.6645, 0.0612],
  15188. [-1.4340, -1.7223, 0.8990, -1.0310, -0.4178, -1.1867, 0.1727, 0.2790],
  15189. [ 0.7065, -0.3046, 1.7416, 0.1067, -0.4849, 0.2435, 0.4102, 0.1512],
  15190. [ 0.6215, -0.3323, 1.6901, -0.1141, -0.5998, -0.5355, 0.1440, 0.1141],
  15191. [ 0.9429, -0.1613, 1.1300, -0.8821, -0.3557, -0.9696, 0.4113, 0.3535]],
  15192. device='cuda:0', grad_fn=<AddmmBackward>)
  15193. landmarks are: tensor([[[ 0.6125, -0.4369, 1.9173, -0.5384, -0.2594, -0.8386, 0.9741,
  15194. 0.1821],
  15195. [ 0.5908, -0.3832, 1.3804, -1.2543, -0.1270, -1.4671, 0.5721,
  15196. 0.2237],
  15197. [-2.2859, -2.2859, 1.1854, -1.0352, -0.4441, -1.2390, 0.0986,
  15198. 0.2853],
  15199. [ 0.6126, -0.3871, 1.6229, -1.0773, -0.2132, -1.3698, 0.5829,
  15200. -0.0209],
  15201. [-2.2859, -2.2859, 0.7023, -1.3883, -0.4268, -1.3621, 0.0813,
  15202. 0.2699],
  15203. [ 0.5444, -0.3846, 1.6171, -0.1689, -0.5885, -0.0380, 0.1791,
  15204. 0.2296],
  15205. [ 0.5561, -0.3834, 1.6229, -0.5153, -0.6231, -0.8079, 0.0727,
  15206. 0.2837],
  15207. [ 0.5687, -0.4088, 1.0397, -1.1466, -0.3113, -1.1928, 0.4681,
  15208. 0.5855]]], device='cuda:0')
  15209. loss_train_step before backward: tensor(0.0665, device='cuda:0', grad_fn=<MseLossBackward>)
  15210. loss_train_step after backward: tensor(0.0665, device='cuda:0', grad_fn=<MseLossBackward>)
  15211. loss_train: 3.630719925276935
  15212. step: 83
  15213. running loss: 0.043743613557553435
  15214. Train Steps: 83/90 Loss: 0.0437 torch.Size([8, 600, 800])
  15215. torch.Size([8, 8])
  15216. tensor([[0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433],
  15217. [0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222],
  15218. [ nan, nan, 0.8488, 0.2300, 0.5563, 0.2100, 0.7390, 0.5679],
  15219. [ nan, nan, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702],
  15220. [ nan, nan, 0.7192, 0.2346, 0.4037, 0.2050, 0.5138, 0.5650],
  15221. [0.6202, 0.4064, 0.7879, 0.2179, 0.4567, 0.1725, 0.5955, 0.5478],
  15222. [ nan, nan, 0.7525, 0.2291, 0.3838, 0.3017, 0.6050, 0.5667],
  15223. [0.6199, 0.4065, 0.7598, 0.2385, 0.4317, 0.1981, 0.5933, 0.5221]],
  15224. device='cuda:0', dtype=torch.float64)
  15225. predictions are: tensor([[ 1.1672, -0.0103, 1.4356, -0.7056, -0.6808, -0.2443, 0.4589, 0.0789],
  15226. [ 1.2498, 0.1052, 1.4687, -0.1174, -0.2292, -0.6798, 0.1431, 0.2818],
  15227. [-1.0567, -1.4895, 1.6280, -0.8708, 0.0069, -0.7950, 0.9648, 0.2630],
  15228. [-0.4714, -1.0502, 1.8932, -0.3346, -0.0877, -0.7876, 0.8404, 0.2231],
  15229. [-1.2858, -1.5904, 1.0354, -0.9126, -0.5178, -1.0090, 0.1041, 0.1625],
  15230. [ 1.2956, 0.0862, 1.2792, -0.7547, -0.2526, -1.0581, 0.2798, 0.0854],
  15231. [-1.0856, -1.4630, 1.2056, -0.8829, -0.5616, -0.7047, 0.3875, 0.2448],
  15232. [ 1.5989, 0.2509, 1.3222, -0.8339, -0.3640, -0.9206, 0.3607, 0.0468]],
  15233. device='cuda:0', grad_fn=<AddmmBackward>)
  15234. landmarks are: tensor([[[ 0.5665, -0.3963, 1.3457, -0.9926, -0.6058, -0.6308, 0.5721,
  15235. 0.2237],
  15236. [ 0.6237, -0.2983, 1.3919, -0.4691, -0.0457, -1.2313, 0.2453,
  15237. 0.5882],
  15238. [-2.2859, -2.2859, 1.6344, -1.2236, 0.2834, -1.3159, 1.1276,
  15239. 0.3371],
  15240. [-2.2859, -2.2859, 1.8942, -0.6693, 0.0871, -1.2236, 1.1130,
  15241. 0.3478],
  15242. [-2.2859, -2.2859, 1.0361, -1.2021, -0.4210, -1.3390, 0.0871,
  15243. 0.3238],
  15244. [ 0.5787, -0.4086, 1.3535, -1.2794, -0.1764, -1.4891, 0.4645,
  15245. 0.2442],
  15246. [-2.2859, -2.2859, 1.1898, -1.2278, -0.5134, -0.8925, 0.5085,
  15247. 0.3315],
  15248. [ 0.5774, -0.4082, 1.2235, -1.1844, -0.2919, -1.3709, 0.4544,
  15249. 0.1256]]], device='cuda:0')
  15250. loss_train_step before backward: tensor(0.2647, device='cuda:0', grad_fn=<MseLossBackward>)
  15251. loss_train_step after backward: tensor(0.2647, device='cuda:0', grad_fn=<MseLossBackward>)
  15252. loss_train: 3.8954066587612033
  15253. step: 84
  15254. running loss: 0.04637388879477623
  15255. Train Steps: 84/90 Loss: 0.0464 torch.Size([8, 600, 800])
  15256. torch.Size([8, 8])
  15257. tensor([[0.6078, 0.4033, 0.8019, 0.3055, 0.3450, 0.4200, 0.6025, 0.5550],
  15258. [0.6200, 0.3978, 0.8900, 0.4550, 0.3775, 0.5200, 0.6150, 0.5367],
  15259. [0.6162, 0.4134, 0.6700, 0.2467, 0.3962, 0.2533, 0.5737, 0.5467],
  15260. [0.6097, 0.4000, 0.7325, 0.2667, 0.3450, 0.3517, 0.5284, 0.5045],
  15261. [0.6339, 0.4149, 0.8800, 0.5000, 0.3900, 0.5283, 0.7541, 0.5424],
  15262. [0.6202, 0.4066, 0.8746, 0.3376, 0.3717, 0.3090, 0.5842, 0.5165],
  15263. [0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933],
  15264. [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207]],
  15265. device='cuda:0', dtype=torch.float64)
  15266. predictions are: tensor([[ 0.0092, -0.7831, 1.4251, -0.8901, -0.6195, -0.3991, 0.4450, 0.2552],
  15267. [ 0.1866, -0.6944, 1.7594, -0.0529, -0.4564, 0.0751, 0.3812, 0.1549],
  15268. [ 0.4072, -0.4985, 1.0225, -1.0642, -0.3926, -1.3225, 0.2822, 0.2517],
  15269. [-0.3008, -0.9750, 1.1834, -1.0713, -0.5787, -0.8064, 0.1357, 0.1497],
  15270. [ 0.4684, -0.5414, 1.8716, 0.1877, -0.3996, 0.2120, 0.9094, 0.1500],
  15271. [ 0.7999, -0.2541, 1.8295, -0.6501, -0.4471, -1.0200, 0.4277, 0.0413],
  15272. [ 0.4781, -0.4347, 1.6580, 0.4612, 0.0516, -0.3173, 0.2600, 0.3793],
  15273. [ 0.3030, -0.6041, 1.5122, -1.0825, -0.1982, -1.4243, 0.6538, 0.1128]],
  15274. device='cuda:0', grad_fn=<AddmmBackward>)
  15275. landmarks are: tensor([[[ 0.5215, -0.4232, 1.4182, -0.8747, -0.6924, -0.3460, 0.4970,
  15276. 0.2776],
  15277. [ 0.5776, -0.4484, 1.8249, -0.1843, -0.5423, 0.1159, 0.5547,
  15278. 0.1929],
  15279. [ 0.5603, -0.3764, 0.8088, -1.1466, -0.4557, -1.1158, 0.3642,
  15280. 0.2391],
  15281. [ 0.5303, -0.4384, 1.0975, -1.0542, -0.6924, -0.6616, 0.1548,
  15282. 0.0442],
  15283. [ 0.6421, -0.3695, 1.7788, 0.0236, -0.4845, 0.1544, 1.1971,
  15284. 0.2196],
  15285. [ 0.5787, -0.4077, 1.7537, -0.7263, -0.5692, -0.8586, 0.4126,
  15286. 0.1000],
  15287. [ 0.5959, -0.3579, 1.6055, 0.3623, -0.0573, -0.2074, 0.3122,
  15288. 0.4547],
  15289. [ 0.6132, -0.4248, 1.5247, -1.1620, -0.2882, -1.3159, 0.6545,
  15290. 0.1193]]], device='cuda:0')
  15291. loss_train_step before backward: tensor(0.0374, device='cuda:0', grad_fn=<MseLossBackward>)
  15292. loss_train_step after backward: tensor(0.0374, device='cuda:0', grad_fn=<MseLossBackward>)
  15293. loss_train: 3.9328551990911365
  15294. step: 85
  15295. running loss: 0.04626888469518984
  15296.  
  15297. Train Steps: 85/90 Loss: 0.0463 torch.Size([8, 600, 800])
  15298. torch.Size([8, 8])
  15299. tensor([[0.6138, 0.4101, 0.8800, 0.5083, 0.4637, 0.5950, 0.5587, 0.5077],
  15300. [0.6069, 0.3975, 0.8625, 0.5083, 0.4388, 0.5483, 0.5650, 0.4967],
  15301. [0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350],
  15302. [0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5737, 0.5350],
  15303. [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750],
  15304. [0.6192, 0.4128, 0.8513, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633],
  15305. [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467],
  15306. [ nan, nan, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552]],
  15307. device='cuda:0', dtype=torch.float64)
  15308. predictions are: tensor([[ 0.3836, -0.5184, 1.6498, 0.0154, -0.1134, 0.1239, 0.3044, 0.1965],
  15309. [ 0.4918, -0.4554, 1.6638, 0.0633, -0.2440, 0.0597, 0.2894, 0.1129],
  15310. [ 0.7150, -0.2920, 1.2913, -1.2055, -0.1567, -1.4490, 0.4141, 0.1894],
  15311. [ 0.7872, -0.2240, 1.7332, -0.0770, -0.4979, -1.0182, 0.3480, 0.1206],
  15312. [ 0.5916, -0.3943, 1.0506, -1.1736, -0.5250, -1.0944, 0.4367, 0.2910],
  15313. [ 0.5973, -0.3884, 1.6319, 0.2809, -0.4084, 0.0409, 0.4515, 0.2362],
  15314. [ 0.1417, -0.6834, 1.4807, -0.8859, -0.7096, -0.4316, 0.4162, 0.2001],
  15315. [-2.0467, -2.1798, 1.6085, -1.1863, -0.0272, -1.3415, 0.9383, 0.2650]],
  15316. device='cuda:0', grad_fn=<AddmmBackward>)
  15317. landmarks are: tensor([[[ 0.5491, -0.3918, 1.7788, 0.0620, -0.1439, 0.4624, 0.2946,
  15318. 0.0592],
  15319. [ 0.5174, -0.4497, 1.6979, 0.0620, -0.2594, 0.2468, 0.3238,
  15320. 0.0082],
  15321. [ 0.5768, -0.4031, 1.3838, -1.1527, -0.2188, -1.4216, 0.4379,
  15322. 0.1850],
  15323. [ 0.5901, -0.4000, 1.8423, -0.0688, -0.5307, -0.9233, 0.3642,
  15324. 0.1852],
  15325. [ 0.5805, -0.3818, 1.0282, -1.1774, -0.4903, -0.9310, 0.5894,
  15326. 0.3700],
  15327. [ 0.5742, -0.3792, 1.6460, 0.3084, -0.3460, 0.1467, 0.4162,
  15328. 0.3161],
  15329. [ 0.5484, -0.4591, 1.5940, -0.7771, -0.6693, -0.3460, 0.4219,
  15330. 0.2391],
  15331. [-2.2859, -2.2859, 1.7557, -1.1466, 0.0871, -1.0773, 1.1239,
  15332. 0.2783]]], device='cuda:0')
  15333. loss_train_step before backward: tensor(0.0146, device='cuda:0', grad_fn=<MseLossBackward>)
  15334. loss_train_step after backward: tensor(0.0146, device='cuda:0', grad_fn=<MseLossBackward>)
  15335. loss_train: 3.9474326334893703
  15336. step: 86
  15337. running loss: 0.04590037945917873
  15338. Train Steps: 86/90 Loss: 0.0459 torch.Size([8, 600, 800])
  15339. torch.Size([8, 8])
  15340. tensor([[0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400],
  15341. [0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400],
  15342. [0.6125, 0.4076, 0.8488, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505],
  15343. [0.6084, 0.3981, 0.8588, 0.5233, 0.4600, 0.5367, 0.5680, 0.5006],
  15344. [0.6263, 0.4039, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804],
  15345. [0.6248, 0.4185, 0.8500, 0.5767, 0.4463, 0.4550, 0.5613, 0.5917],
  15346. [0.6243, 0.4128, 0.7762, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417],
  15347. [0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400]],
  15348. device='cuda:0', dtype=torch.float64)
  15349. predictions are: tensor([[ 0.3863, -0.5068, 1.3909, -1.0064, -0.0798, -1.3737, 0.4426, 0.1917],
  15350. [ 0.0985, -0.7115, 1.3916, -1.2851, -0.0720, -1.3754, 0.7241, 0.2478],
  15351. [ 0.3931, -0.5064, 1.5499, -0.4864, -0.6317, -0.7154, 0.0686, 0.1628],
  15352. [ 0.2412, -0.6454, 1.5829, 0.1082, -0.1753, 0.0653, 0.2417, 0.1627],
  15353. [ 0.0780, -0.7639, 1.7483, -0.3355, -0.5922, -0.1877, 0.5523, 0.0529],
  15354. [ 0.2777, -0.6107, 1.5568, 0.3643, -0.1859, -0.3246, 0.4243, 0.4047],
  15355. [ 0.1649, -0.6641, 1.2672, -1.1711, -0.3359, -1.0829, 0.6421, 0.2771],
  15356. [-0.1123, -0.8771, 1.6413, -0.5845, -0.6554, -0.2488, 0.6509, 0.2595]],
  15357. device='cuda:0', grad_fn=<AddmmBackward>)
  15358. landmarks are: tensor([[[ 0.5911, -0.3888, 1.4727, -0.9541, -0.0919, -1.4930, 0.3988,
  15359. 0.2083],
  15360. [ 0.5958, -0.3818, 1.5536, -1.1081, -0.2074, -1.3390, 0.5663,
  15361. 0.2083],
  15362. [ 0.5433, -0.4032, 1.6344, -0.4922, -0.5769, -0.5846, 0.0357,
  15363. 0.2567],
  15364. [ 0.5241, -0.4470, 1.6806, 0.1313, -0.1612, 0.1929, 0.3378,
  15365. 0.0261],
  15366. [ 0.6072, -0.4206, 1.8711, -0.2536, -0.6115, -0.1304, 0.6812,
  15367. -0.0670],
  15368. [ 0.5999, -0.3530, 1.6402, 0.3777, -0.2247, -0.1843, 0.3065,
  15369. 0.4470],
  15370. [ 0.5977, -0.3792, 1.2995, -1.0311, -0.5192, -0.8386, 0.5836,
  15371. 0.2160],
  15372. [ 0.5778, -0.4393, 1.8018, -0.4614, -0.6693, -0.1381, 0.5490,
  15373. 0.2083]]], device='cuda:0')
  15374. loss_train_step before backward: tensor(0.0403, device='cuda:0', grad_fn=<MseLossBackward>)
  15375. loss_train_step after backward: tensor(0.0403, device='cuda:0', grad_fn=<MseLossBackward>)
  15376. loss_train: 3.9877162650227547
  15377. step: 87
  15378. running loss: 0.04583581913819258
  15379. Train Steps: 87/90 Loss: 0.0458 torch.Size([8, 600, 800])
  15380. torch.Size([8, 8])
  15381. tensor([[0.6277, 0.4057, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823],
  15382. [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082],
  15383. [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586],
  15384. [0.6357, 0.4097, 0.9038, 0.3883, 0.4213, 0.2950, 0.6686, 0.5390],
  15385. [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285],
  15386. [0.6193, 0.4108, 0.7438, 0.2700, 0.3650, 0.3683, 0.6238, 0.5717],
  15387. [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467],
  15388. [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896]],
  15389. device='cuda:0', dtype=torch.float64)
  15390. predictions are: tensor([[ 0.1744, -0.6968, 1.5299, -1.1521, -0.1250, -1.4734, 0.6400, 0.1067],
  15391. [ 0.4351, -0.5575, 1.7017, 0.2760, -0.1643, 0.3751, 0.4155, 0.1889],
  15392. [-0.1450, -0.9232, 1.6934, -1.1407, -0.0424, -1.2464, 1.1251, 0.2949],
  15393. [ 0.1820, -0.6939, 1.7591, -0.4505, -0.2610, -1.0295, 0.6851, 0.1865],
  15394. [ 0.4787, -0.4771, 1.3901, -0.4078, -0.5066, -0.3232, 0.0269, 0.1471],
  15395. [ 0.1439, -0.7145, 1.1538, -1.0634, -0.5412, -0.6070, 0.4496, 0.3232],
  15396. [ 0.1047, -0.7455, 1.5241, -0.7351, -0.6339, -0.3523, 0.4122, 0.2333],
  15397. [ 0.3210, -0.5923, 1.4436, -0.4391, -0.5093, -0.9542, 0.0538, 0.3497]],
  15398. device='cuda:0', grad_fn=<AddmmBackward>)
  15399. landmarks are: tensor([[[ 0.6132, -0.4122, 1.5478, -1.0619, -0.2709, -1.4314, 0.5500,
  15400. -0.0583],
  15401. [ 0.5450, -0.4706, 1.7643, 0.0722, -0.3708, 0.3200, 0.4854,
  15402. 0.0612],
  15403. [ 0.6075, -0.4129, 1.6055, -1.0080, -0.2420, -1.0080, 0.9704,
  15404. 0.2944],
  15405. [ 0.6504, -0.3936, 1.8885, -0.4922, -0.3402, -0.9233, 0.8022,
  15406. 0.2035],
  15407. [ 0.5253, -0.4296, 1.3084, -0.3098, -0.6115, -0.2767, 0.0928,
  15408. 0.1552],
  15409. [ 0.5747, -0.3886, 1.1494, -1.0388, -0.6000, -0.5846, 0.5952,
  15410. 0.3546],
  15411. [ 0.5484, -0.4591, 1.5940, -0.7771, -0.6693, -0.3460, 0.4219,
  15412. 0.2391],
  15413. [ 0.5485, -0.3997, 1.4445, -0.4895, -0.6000, -0.8309, 0.1878,
  15414. 0.4374]]], device='cuda:0')
  15415. loss_train_step before backward: tensor(0.0404, device='cuda:0', grad_fn=<MseLossBackward>)
  15416. loss_train_step after backward: tensor(0.0404, device='cuda:0', grad_fn=<MseLossBackward>)
  15417. loss_train: 4.0281651839613914
  15418. step: 88
  15419. running loss: 0.04577460436319763
  15420. Train Steps: 88/90 Loss: 0.0458 torch.Size([8, 600, 800])
  15421. torch.Size([8, 8])
  15422. tensor([[0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719],
  15423. [0.6267, 0.4080, 0.8438, 0.2633, 0.4763, 0.1800, 0.6259, 0.5240],
  15424. [0.6201, 0.3970, 0.8413, 0.4950, 0.4413, 0.5183, 0.6088, 0.5400],
  15425. [0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320],
  15426. [0.6259, 0.4133, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947],
  15427. [0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5787, 0.5600],
  15428. [0.6271, 0.4040, 0.9138, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413],
  15429. [0.6299, 0.4008, 0.8450, 0.5350, 0.4213, 0.5000, 0.6350, 0.5100]],
  15430. device='cuda:0', dtype=torch.float64)
  15431. predictions are: tensor([[ 0.1915, -0.6440, 1.4687, -0.5500, -0.6943, -0.3780, 0.1095, 0.2380],
  15432. [ 0.3804, -0.5327, 1.4882, -1.2403, -0.1517, -1.4743, 0.5692, 0.2154],
  15433. [-0.0196, -0.8108, 1.5399, -0.0779, -0.3252, 0.0614, 0.3196, 0.2752],
  15434. [ 0.3610, -0.5834, 1.6524, -0.2101, -0.6242, -0.3816, 0.5197, 0.1405],
  15435. [ 0.4290, -0.5124, 1.3662, -1.4068, -0.0729, -1.5972, 0.5732, 0.2222],
  15436. [ 0.1196, -0.7140, 1.5856, -0.2194, -0.1171, -0.0294, 0.3316, 0.3566],
  15437. [ 0.1048, -0.7306, 1.9049, -0.7238, -0.1563, -1.2254, 0.9249, 0.1957],
  15438. [ 0.2424, -0.6750, 1.5581, 0.0550, -0.4404, -0.0346, 0.5450, 0.2040]],
  15439. device='cuda:0', grad_fn=<AddmmBackward>)
  15440. landmarks are: tensor([[[ 5.3672e-01, -4.2941e-01, 1.5709e+00, -4.9992e-01, -6.6928e-01,
  15441. -3.0747e-01, 2.4546e-01, 3.5585e-01],
  15442. [ 6.0878e-01, -4.0146e-01, 1.6113e+00, -1.0696e+00, -8.6143e-02,
  15443. -1.4545e+00, 6.0510e-01, 1.3434e-01],
  15444. [ 5.7812e-01, -4.5219e-01, 1.5998e+00, 4.6189e-04, -2.4781e-01,
  15445. 1.0824e-01, 5.2587e-01, 2.0831e-01],
  15446. [ 6.1742e-01, -4.4897e-01, 1.8885e+00, -9.9615e-02, -4.8453e-01,
  15447. -3.6905e-01, 9.8137e-01, 1.7146e-01],
  15448. [ 6.0514e-01, -3.7714e-01, 1.5016e+00, -1.2159e+00, 3.5104e-02,
  15449. -1.5777e+00, 6.0099e-01, -9.2270e-04],
  15450. [ 5.5052e-01, -4.2071e-01, 1.7095e+00, -5.3426e-02, -5.0936e-02,
  15451. 1.0502e-01, 3.8730e-01, 3.0069e-01],
  15452. [ 6.1083e-01, -4.2008e-01, 1.9346e+00, -5.5381e-01, -1.4965e-01,
  15453. -1.0773e+00, 1.0545e+00, 2.1421e-01],
  15454. [ 6.2361e-01, -4.3441e-01, 1.6171e+00, 1.8522e-01, -3.4018e-01,
  15455. 2.3557e-02, 6.4711e-01, 6.9746e-02]]], device='cuda:0')
  15456. loss_train_step before backward: tensor(0.0387, device='cuda:0', grad_fn=<MseLossBackward>)
  15457. loss_train_step after backward: tensor(0.0387, device='cuda:0', grad_fn=<MseLossBackward>)
  15458. loss_train: 4.066847592592239
  15459. step: 89
  15460. running loss: 0.04569491677069932
  15461.  
  15462. Train Steps: 89/90 Loss: 0.0457 torch.Size([8, 600, 800])
  15463. torch.Size([8, 8])
  15464. tensor([[0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350],
  15465. [0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400],
  15466. [0.6249, 0.4142, 0.8350, 0.3283, 0.3613, 0.3700, 0.6188, 0.5400],
  15467. [0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038],
  15468. [0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166],
  15469. [0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767],
  15470. [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852],
  15471. [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600]],
  15472. device='cuda:0', dtype=torch.float64)
  15473. predictions are: tensor([[ 0.2890, -0.5716, 1.3113, -1.2958, -0.1987, -1.5838, 0.4590, 0.2383],
  15474. [ 0.2683, -0.5812, 1.4407, -1.0742, -0.1550, -1.5367, 0.4240, 0.2070],
  15475. [ 0.4971, -0.4523, 1.5860, -0.8335, -0.6447, -0.9323, 0.5781, 0.2279],
  15476. [ 0.2467, -0.6203, 1.7023, -0.1982, -0.2799, -0.1041, 0.2939, 0.1046],
  15477. [ 0.2707, -0.6597, 1.7397, -0.2222, -0.5196, 0.0379, 0.8742, 0.2198],
  15478. [ 0.2934, -0.6259, 1.6778, -0.0822, -0.4994, -0.3037, 0.4800, 0.2324],
  15479. [ 0.1104, -0.7315, 1.6631, -0.0197, -0.1773, -0.1624, 0.3443, 0.3367],
  15480. [ 0.1810, -0.6591, 1.6120, -0.3214, -0.1508, -0.0314, 0.4300, 0.3403]],
  15481. device='cuda:0', grad_fn=<AddmmBackward>)
  15482. landmarks are: tensor([[[ 0.5768, -0.4031, 1.3838, -1.1527, -0.2188, -1.4216, 0.4379,
  15483. 0.1850],
  15484. [ 0.5911, -0.3888, 1.4727, -0.9541, -0.0919, -1.4930, 0.3988,
  15485. 0.2083],
  15486. [ 0.6006, -0.3728, 1.5709, -0.7694, -0.6173, -0.5769, 0.5721,
  15487. 0.2083],
  15488. [ 0.5249, -0.4473, 1.7326, -0.0919, -0.2016, 0.1544, 0.1733,
  15489. 0.0412],
  15490. [ 0.5944, -0.4490, 1.8643, -0.0659, -0.5147, 0.1235, 0.7684,
  15491. 0.1004],
  15492. [ 0.5610, -0.4381, 1.7730, 0.0390, -0.4326, -0.1458, 0.1794,
  15493. 0.3777],
  15494. [ 0.5500, -0.4060, 1.7326, 0.0236, -0.1554, -0.0226, 0.1438,
  15495. 0.4171],
  15496. [ 0.5652, -0.4325, 1.7388, -0.1429, -0.0688, 0.3469, 0.4970,
  15497. 0.3007]]], device='cuda:0')
  15498. loss_train_step before backward: tensor(0.0306, device='cuda:0', grad_fn=<MseLossBackward>)
  15499. loss_train_step after backward: tensor(0.0306, device='cuda:0', grad_fn=<MseLossBackward>)
  15500. loss_train: 4.097452521324158
  15501. step: 90
  15502. running loss: 0.04552725023693509
  15503. Valid Steps: 10/10 Loss: nan 5.8477
  15504. --------------------------------------------------
  15505. Epoch: 4 Train Loss: 0.0455 Valid Loss: nan
  15506. --------------------------------------------------
  15507. size of train loader is: 90
  15508. torch.Size([8, 600, 800])
  15509. torch.Size([8, 8])
  15510. tensor([[0.6081, 0.3950, 0.8538, 0.4667, 0.3850, 0.4917, 0.5342, 0.4954],
  15511. [0.6161, 0.4024, 0.8662, 0.4683, 0.4935, 0.5364, 0.6063, 0.5567],
  15512. [0.6263, 0.4233, 0.7924, 0.4626, 0.3788, 0.2883, 0.5573, 0.6047],
  15513. [0.6250, 0.4013, 0.8525, 0.5417, 0.4037, 0.5117, 0.6325, 0.5017],
  15514. [0.6216, 0.4167, 0.8588, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667],
  15515. [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773],
  15516. [0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5787, 0.5600],
  15517. [0.6068, 0.3963, 0.8650, 0.4317, 0.4037, 0.5083, 0.5253, 0.4999]],
  15518. device='cuda:0', dtype=torch.float64)
  15519. predictions are: tensor([[ 3.4651e-01, -5.8417e-01, 1.7318e+00, -5.1866e-01, -5.5468e-01,
  15520. -4.1584e-01, 3.5147e-01, 5.9733e-02],
  15521. [ 2.0160e-01, -6.7345e-01, 1.7936e+00, -4.8521e-01, -6.1855e-02,
  15522. -4.0095e-01, 5.5372e-01, 2.8475e-01],
  15523. [ 4.4827e-01, -4.2845e-01, 1.4146e+00, -5.0683e-01, -5.0848e-01,
  15524. -1.2103e+00, 2.3973e-01, 5.2721e-01],
  15525. [ 5.1592e-01, -4.7822e-01, 1.7209e+00, -1.5612e-01, -4.5384e-01,
  15526. -2.8481e-01, 7.0564e-01, 1.2522e-01],
  15527. [ 7.6344e-01, -2.9085e-01, 1.7067e+00, 8.7751e-04, -4.5148e-01,
  15528. -2.4599e-01, 6.7884e-01, 2.8083e-01],
  15529. [ 2.3650e-01, -6.3350e-01, 1.7368e+00, -3.9993e-01, -3.5953e-02,
  15530. -4.3623e-01, 3.3012e-01, 2.6307e-01],
  15531. [ 3.1048e-01, -5.8417e-01, 1.7569e+00, -4.4972e-01, -6.0495e-02,
  15532. -4.0298e-01, 5.7915e-01, 3.2222e-01],
  15533. [ 1.8989e-01, -6.7958e-01, 1.7583e+00, -5.6773e-01, -4.8253e-01,
  15534. -2.4003e-01, 4.2579e-01, 8.0821e-02]], device='cuda:0',
  15535. grad_fn=<AddmmBackward>)
  15536. landmarks are: tensor([[[ 0.5227, -0.4615, 1.6575, -0.1304, -0.5076, -0.0149, 0.1815,
  15537. 0.0021],
  15538. [ 0.5598, -0.4273, 1.7152, -0.1227, -0.0065, 0.1917, 0.5143,
  15539. 0.2853],
  15540. [ 0.6069, -0.3309, 1.3742, -0.1493, -0.5365, -0.9541, 0.2884,
  15541. 0.5071],
  15542. [ 0.6010, -0.4321, 1.6517, 0.2160, -0.4210, 0.0774, 0.6356,
  15543. 0.0313],
  15544. [ 0.5853, -0.3614, 1.6806, 0.2930, -0.4499, 0.1005, 0.3815,
  15545. 0.3315],
  15546. [ 0.5428, -0.4244, 1.7095, -0.0380, 0.0232, 0.0725, 0.0866,
  15547. 0.3806],
  15548. [ 0.5505, -0.4207, 1.7095, -0.0534, -0.0509, 0.1050, 0.3873,
  15549. 0.3007],
  15550. [ 0.5168, -0.4556, 1.7095, -0.2921, -0.4210, 0.0620, 0.1404,
  15551. 0.0231]]], device='cuda:0')
  15552. loss_train_step before backward: tensor(0.0548, device='cuda:0', grad_fn=<MseLossBackward>)
  15553. loss_train_step after backward: tensor(0.0548, device='cuda:0', grad_fn=<MseLossBackward>)
  15554. loss_train: 0.054758865386247635
  15555. step: 1
  15556. running loss: 0.054758865386247635
  15557. Train Steps: 1/90 Loss: 0.0548 torch.Size([8, 600, 800])
  15558. torch.Size([8, 8])
  15559. tensor([[0.6197, 0.4090, 0.7825, 0.2500, 0.4200, 0.2483, 0.5988, 0.5667],
  15560. [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517],
  15561. [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170],
  15562. [0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058],
  15563. [0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5988, 0.5700],
  15564. [0.6266, 0.4067, 0.8588, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267],
  15565. [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391],
  15566. [0.6236, 0.3967, 0.8675, 0.5400, 0.3862, 0.4517, 0.5825, 0.5200]],
  15567. device='cuda:0', dtype=torch.float64)
  15568. predictions are: tensor([[ 0.3067, -0.5911, 1.3730, -1.0648, -0.3434, -0.9991, 0.5182, 0.3448],
  15569. [-0.3291, -0.9670, 1.0663, -1.0819, -0.3201, -1.1403, 0.2702, 0.3937],
  15570. [ 0.5365, -0.4417, 1.6636, -1.0241, -0.0103, -1.3002, 0.7841, 0.2095],
  15571. [ 0.7065, -0.3506, 1.7970, -0.1609, -0.5044, 0.1665, 0.2508, 0.1114],
  15572. [-0.1420, -0.8803, 1.4987, -0.7562, -0.4674, -0.4524, 0.4285, 0.3405],
  15573. [ 0.6437, -0.3572, 1.7996, -0.9246, -0.2193, -0.9633, 0.7583, 0.2072],
  15574. [ 0.3890, -0.5329, 1.5026, -0.9721, -0.3509, -1.0259, 0.4017, 0.1881],
  15575. [ 0.6239, -0.4407, 1.8843, 0.4963, -0.4543, 0.0053, 0.5245, 0.1646]],
  15576. device='cuda:0', grad_fn=<AddmmBackward>)
  15577. landmarks are: tensor([[[ 0.5764, -0.3969, 1.3284, -1.1312, -0.3460, -1.1389, 0.4797,
  15578. 0.3315],
  15579. [ 0.5713, -0.3671, 0.8665, -1.0696, -0.3691, -1.2236, 0.3527,
  15580. 0.2622],
  15581. [ 0.6471, -0.3840, 1.5767, -1.0311, -0.0457, -1.5007, 0.6889,
  15582. 0.1020],
  15583. [ 0.5245, -0.4361, 1.5940, -0.2921, -0.5480, -0.0919, 0.2432,
  15584. 0.0502],
  15585. [ 0.5960, -0.3888, 1.4840, -1.0095, -0.6115, -0.6231, 0.4797,
  15586. 0.3469],
  15587. [ 0.6084, -0.4076, 1.6806, -0.9618, -0.2998, -0.9695, 0.6356,
  15588. 0.1467],
  15589. [ 0.5785, -0.4259, 1.4228, -1.0261, -0.4190, -1.2189, 0.4763,
  15590. 0.2043],
  15591. [ 0.5944, -0.4538, 1.7210, 0.2083, -0.5018, -0.1997, 0.4046,
  15592. 0.1159]]], device='cuda:0')
  15593. loss_train_step before backward: tensor(0.0447, device='cuda:0', grad_fn=<MseLossBackward>)
  15594. loss_train_step after backward: tensor(0.0447, device='cuda:0', grad_fn=<MseLossBackward>)
  15595. loss_train: 0.09948705509305
  15596. step: 2
  15597. running loss: 0.049743527546525
  15598. Train Steps: 2/90 Loss: 0.0497 torch.Size([8, 600, 800])
  15599. torch.Size([8, 8])
  15600. tensor([[0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355],
  15601. [0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367],
  15602. [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517],
  15603. [0.6112, 0.4029, 0.8638, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567],
  15604. [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973],
  15605. [0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124],
  15606. [0.6126, 0.4073, 0.8750, 0.5133, 0.3800, 0.4333, 0.4986, 0.5378],
  15607. [0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950]],
  15608. device='cuda:0', dtype=torch.float64)
  15609. predictions are: tensor([[ 0.5210, -0.4563, 1.7610, -0.8858, -0.2437, -0.7401, 0.9069, 0.1881],
  15610. [ 0.2435, -0.5746, 1.7808, -0.4829, -0.3414, -1.1378, 0.3184, 0.3190],
  15611. [-0.1094, -0.8018, 0.9690, -1.3126, -0.3771, -1.2953, 0.2717, 0.4078],
  15612. [ 0.4078, -0.5230, 1.8003, -0.1228, -0.0185, 0.1290, 0.4568, 0.3150],
  15613. [ 0.7459, -0.3426, 1.8601, 0.1211, -0.5755, -0.1134, 0.6215, 0.1065],
  15614. [ 0.4384, -0.4892, 1.4912, -1.1436, -0.3577, -1.0383, 0.4101, 0.1810],
  15615. [ 0.6986, -0.3227, 1.7810, 0.1030, -0.4969, -0.2094, 0.2139, 0.2080],
  15616. [ 0.7046, -0.3568, 1.7441, 0.0391, -0.2582, 0.1621, 0.4744, 0.1544]],
  15617. device='cuda:0', grad_fn=<AddmmBackward>)
  15618. landmarks are: tensor([[[ 6.2236e-01, -4.1045e-01, 1.9173e+00, -7.7706e-01, -1.0299e-01,
  15619. -7.3084e-01, 1.1532e+00, 1.8749e-01],
  15620. [ 5.8995e-01, -3.9323e-01, 1.8307e+00, -3.9215e-01, -4.2679e-01,
  15621. -1.1851e+00, 3.7575e-01, 1.9292e-01],
  15622. [ 5.7131e-01, -3.6712e-01, 8.6651e-01, -1.0696e+00, -3.6905e-01,
  15623. -1.2236e+00, 3.5266e-01, 2.6220e-01],
  15624. [ 5.3712e-01, -4.2517e-01, 1.7037e+00, -6.8822e-02, -3.4180e-02,
  15625. 6.2048e-02, 3.7575e-01, 2.8530e-01],
  15626. [ 6.2401e-01, -4.3212e-01, 1.8423e+00, 1.8522e-01, -5.8845e-01,
  15627. -1.6120e-01, 6.9623e-01, 1.1149e-02],
  15628. [ 5.7898e-01, -4.0793e-01, 1.5929e+00, -1.0630e+00, -4.7294e-01,
  15629. -1.0725e+00, 4.1374e-01, 8.0707e-02],
  15630. [ 5.4353e-01, -4.0454e-01, 1.7557e+00, 8.5142e-02, -5.3072e-01,
  15631. -2.8437e-01, 1.7213e-02, 1.9805e-01],
  15632. [ 5.8199e-01, -4.7544e-01, 1.7095e+00, 1.9292e-01, -2.5358e-01,
  15633. 5.4350e-02, 4.9700e-01, 4.6189e-04]]], device='cuda:0')
  15634. loss_train_step before backward: tensor(0.0220, device='cuda:0', grad_fn=<MseLossBackward>)
  15635. loss_train_step after backward: tensor(0.0220, device='cuda:0', grad_fn=<MseLossBackward>)
  15636. loss_train: 0.12145851366221905
  15637. step: 3
  15638. running loss: 0.040486171220739685
  15639.  
  15640. Train Steps: 3/90 Loss: 0.0405 torch.Size([8, 600, 800])
  15641. torch.Size([8, 8])
  15642. tensor([[0.6276, 0.4120, 0.8738, 0.3133, 0.4225, 0.2217, 0.6203, 0.4892],
  15643. [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817],
  15644. [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123],
  15645. [0.6201, 0.4151, 0.8588, 0.5467, 0.3700, 0.3950, 0.5637, 0.5933],
  15646. [0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431],
  15647. [0.6147, 0.4112, 0.7988, 0.3200, 0.3775, 0.2767, 0.5150, 0.5550],
  15648. [0.6229, 0.4066, 0.7612, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350],
  15649. [0.6100, 0.4016, 0.8600, 0.5067, 0.4612, 0.5233, 0.5086, 0.5519]],
  15650. device='cuda:0', dtype=torch.float64)
  15651. predictions are: tensor([[ 0.5027, -0.4413, 1.8971, -0.8446, -0.2785, -1.2732, 0.7965, 0.1499],
  15652. [ 0.3700, -0.5806, 1.7208, 0.2440, -0.2295, 0.1421, 0.3970, 0.1733],
  15653. [ 0.4950, -0.4589, 1.1742, -1.2359, -0.5166, -1.0659, 0.2792, 0.1711],
  15654. [ 0.6150, -0.3365, 1.7022, 0.1509, -0.4928, -0.4111, 0.5337, 0.3872],
  15655. [ 0.6815, -0.3452, 1.8120, -0.0561, -0.2128, 0.1955, 0.3699, 0.1365],
  15656. [ 0.2446, -0.5919, 1.4378, -0.8585, -0.5198, -0.8801, 0.1190, 0.2465],
  15657. [ 0.7320, -0.2839, 1.4079, -1.1618, -0.3278, -1.2204, 0.5434, 0.2279],
  15658. [ 0.4952, -0.4617, 1.7817, 0.0618, -0.1303, 0.2405, 0.4049, 0.2265]],
  15659. device='cuda:0', grad_fn=<AddmmBackward>)
  15660. landmarks are: tensor([[[ 0.6128, -0.3828, 1.7499, -0.8386, -0.3344, -1.2620, 0.5792,
  15661. -0.0263],
  15662. [ 0.5193, -0.4699, 1.5767, 0.4008, -0.2420, 0.0774, 0.1178,
  15663. -0.0610],
  15664. [ 0.5342, -0.4280, 1.0051, -1.0619, -0.5711, -1.0388, 0.2226,
  15665. 0.0802],
  15666. [ 0.5785, -0.3687, 1.6806, 0.2391, -0.5769, -0.4614, 0.3180,
  15667. 0.4547],
  15668. [ 0.5351, -0.4153, 1.7326, -0.0457, -0.2214, -0.0466, 0.0434,
  15669. 0.2228],
  15670. [ 0.5532, -0.3864, 1.4035, -0.8079, -0.5423, -1.0080, 0.0928,
  15671. 0.2776],
  15672. [ 0.5911, -0.4080, 1.2303, -0.9156, -0.3229, -1.2851, 0.4508,
  15673. 0.1852],
  15674. [ 0.5318, -0.4310, 1.6864, 0.0543, -0.1554, 0.1313, 0.0635,
  15675. 0.2634]]], device='cuda:0')
  15676. loss_train_step before backward: tensor(0.0177, device='cuda:0', grad_fn=<MseLossBackward>)
  15677. loss_train_step after backward: tensor(0.0177, device='cuda:0', grad_fn=<MseLossBackward>)
  15678. loss_train: 0.13915947452187538
  15679. step: 4
  15680. running loss: 0.034789868630468845
  15681. Train Steps: 4/90 Loss: 0.0348 torch.Size([8, 600, 800])
  15682. torch.Size([8, 8])
  15683. tensor([[0.6059, 0.4002, 0.7562, 0.2767, 0.3538, 0.3033, 0.5529, 0.5455],
  15684. [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466],
  15685. [ nan, nan, 0.7425, 0.2117, 0.3937, 0.2433, 0.5438, 0.5567],
  15686. [0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131],
  15687. [0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350],
  15688. [0.6091, 0.3997, 0.8314, 0.4334, 0.3788, 0.4550, 0.5213, 0.5656],
  15689. [0.6200, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391],
  15690. [0.6273, 0.4143, 0.8750, 0.5700, 0.3987, 0.4717, 0.6013, 0.5467]],
  15691. device='cuda:0', dtype=torch.float64)
  15692. predictions are: tensor([[ 0.8880, -0.1930, 1.3037, -0.9538, -0.5272, -0.8709, 0.3894, 0.2118],
  15693. [ 0.7731, -0.2506, 1.6970, 0.1314, -0.1712, 0.3505, 0.3391, 0.2241],
  15694. [-1.9436, -2.0312, 1.3338, -0.8958, -0.3679, -1.0223, 0.3111, 0.2546],
  15695. [ 0.9518, -0.1705, 1.7189, -0.3662, -0.5749, -0.1627, 0.4393, 0.1420],
  15696. [ 0.7302, -0.2546, 1.4106, -1.0633, -0.1283, -1.3372, 0.4802, 0.2443],
  15697. [ 0.9238, -0.1728, 1.6986, -0.2053, -0.4794, -0.0917, 0.3199, 0.2095],
  15698. [ 0.9332, -0.1299, 1.6650, -0.8772, -0.2057, -1.0746, 0.5534, 0.1711],
  15699. [ 0.8415, -0.2139, 1.7384, 0.2925, -0.3925, -0.0505, 0.5245, 0.2025]],
  15700. device='cuda:0', grad_fn=<AddmmBackward>)
  15701. landmarks are: tensor([[[ 0.5129, -0.4374, 1.2072, -1.0080, -0.6520, -0.8848, 0.2679,
  15702. 0.2335],
  15703. [ 0.5178, -0.4480, 1.6748, 0.0620, -0.2767, 0.2083, 0.1067,
  15704. 0.2386],
  15705. [-2.2859, -2.2859, 1.1436, -1.3082, -0.4672, -1.1620, 0.2256,
  15706. 0.2853],
  15707. [ 0.5581, -0.4586, 1.5586, -0.3747, -0.6792, -0.2391, 0.4455,
  15708. 0.0840],
  15709. [ 0.5768, -0.4031, 1.3838, -1.1527, -0.2188, -1.4216, 0.4379,
  15710. 0.1850],
  15711. [ 0.5275, -0.4396, 1.5543, -0.2841, -0.5365, -0.1843, 0.1221,
  15712. 0.3265],
  15713. [ 0.5776, -0.3987, 1.6005, -0.9121, -0.3844, -1.2358, 0.4247,
  15714. 0.2043],
  15715. [ 0.6115, -0.3724, 1.7557, 0.3469, -0.4441, -0.1073, 0.4912,
  15716. 0.2391]]], device='cuda:0')
  15717. loss_train_step before backward: tensor(0.0299, device='cuda:0', grad_fn=<MseLossBackward>)
  15718. loss_train_step after backward: tensor(0.0299, device='cuda:0', grad_fn=<MseLossBackward>)
  15719. loss_train: 0.16904453001916409
  15720. step: 5
  15721. running loss: 0.033808906003832816
  15722. Train Steps: 5/90 Loss: 0.0338 torch.Size([8, 600, 800])
  15723. torch.Size([8, 8])
  15724. tensor([[0.6189, 0.4033, 0.8650, 0.5267, 0.4487, 0.5150, 0.5925, 0.5050],
  15725. [0.6193, 0.4108, 0.7438, 0.2700, 0.3650, 0.3683, 0.6238, 0.5717],
  15726. [0.6161, 0.4024, 0.8662, 0.4683, 0.4935, 0.5364, 0.6063, 0.5567],
  15727. [0.6125, 0.3974, 0.7725, 0.2517, 0.3538, 0.3317, 0.5887, 0.5500],
  15728. [0.6064, 0.4019, 0.8650, 0.4517, 0.4037, 0.5367, 0.5703, 0.5609],
  15729. [0.6200, 0.4118, 0.8287, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799],
  15730. [ nan, nan, 0.8525, 0.2217, 0.5413, 0.2367, 0.7367, 0.5482],
  15731. [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575]],
  15732. device='cuda:0', dtype=torch.float64)
  15733. predictions are: tensor([[ 0.8357, -0.2498, 1.7605, 0.2656, -0.2423, 0.1694, 0.2467, 0.1434],
  15734. [ 0.9518, -0.1559, 1.2529, -0.8951, -0.6224, -0.6567, 0.3689, 0.2523],
  15735. [ 0.6504, -0.3560, 1.8047, -0.0042, -0.0985, 0.1004, 0.2837, 0.2233],
  15736. [ 0.8765, -0.1998, 1.3278, -0.8894, -0.6295, -0.8862, 0.1980, 0.1767],
  15737. [ 0.8432, -0.2417, 1.7446, -0.0057, -0.4000, 0.3157, 0.2729, 0.1645],
  15738. [ 0.6089, -0.3450, 1.5589, -0.3284, -0.5271, -1.1253, 0.0505, 0.2763],
  15739. [-0.9919, -1.4238, 1.7263, -1.0578, 0.1417, -1.2524, 1.0073, 0.2014],
  15740. [ 1.0442, -0.1350, 1.7510, -0.6142, -0.4799, -0.3495, 0.8260, 0.1358]],
  15741. device='cuda:0', grad_fn=<AddmmBackward>)
  15742. landmarks are: tensor([[[ 0.5726, -0.4230, 1.7095, 0.1467, -0.2132, 0.0928, 0.4508,
  15743. 0.0467],
  15744. [ 0.5747, -0.3886, 1.1494, -1.0388, -0.6000, -0.5846, 0.5952,
  15745. 0.3546],
  15746. [ 0.5598, -0.4273, 1.7152, -0.1227, -0.0065, 0.1917, 0.5143,
  15747. 0.2853],
  15748. [ 0.5430, -0.4503, 1.2822, -1.1235, -0.6520, -0.7540, 0.4335,
  15749. 0.2545],
  15750. [ 0.5151, -0.4296, 1.7095, -0.1997, -0.4210, 0.1929, 0.3484,
  15751. 0.3047],
  15752. [ 0.5779, -0.3840, 1.5420, -0.4306, -0.5423, -0.9772, 0.2041,
  15753. 0.3928],
  15754. [-2.2859, -2.2859, 1.6517, -1.2620, 0.2141, -1.1928, 1.1166,
  15755. 0.2463],
  15756. [ 0.6158, -0.4249, 1.8654, -0.9002, -0.3229, -0.3537, 0.9667,
  15757. 0.2890]]], device='cuda:0')
  15758. loss_train_step before backward: tensor(0.0641, device='cuda:0', grad_fn=<MseLossBackward>)
  15759. loss_train_step after backward: tensor(0.0641, device='cuda:0', grad_fn=<MseLossBackward>)
  15760. loss_train: 0.2331299614161253
  15761. step: 6
  15762. running loss: 0.038854993569354214
  15763. Train Steps: 6/90 Loss: 0.0389 torch.Size([8, 600, 800])
  15764. torch.Size([8, 8])
  15765. tensor([[0.6261, 0.4029, 0.8720, 0.3364, 0.3665, 0.3753, 0.6531, 0.5183],
  15766. [0.6218, 0.4098, 0.7238, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350],
  15767. [0.6108, 0.4011, 0.8037, 0.3400, 0.3700, 0.2933, 0.5658, 0.5617],
  15768. [0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274],
  15769. [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478],
  15770. [ nan, nan, 0.6793, 0.2110, 0.4012, 0.2167, 0.5112, 0.5583],
  15771. [0.6115, 0.4005, 0.8838, 0.3867, 0.3763, 0.4700, 0.5800, 0.5550],
  15772. [0.6151, 0.4058, 0.7068, 0.2680, 0.3400, 0.4083, 0.5775, 0.5733]],
  15773. device='cuda:0', dtype=torch.float64)
  15774. predictions are: tensor([[ 0.9756, -0.1992, 1.7893, -0.4266, -0.4919, -0.2887, 0.6990, 0.1151],
  15775. [ 0.9571, -0.1918, 1.1712, -1.0054, -0.2589, -1.1467, 0.5271, 0.2481],
  15776. [ 0.8672, -0.2019, 1.5381, -0.4543, -0.4556, -0.7202, 0.2660, 0.1861],
  15777. [ 0.8332, -0.2177, 1.7749, -0.5389, 0.0245, -1.0307, 0.5606, 0.1816],
  15778. [ 0.7884, -0.2923, 1.8857, -0.3090, -0.4546, -0.6597, 0.4157, 0.1693],
  15779. [-2.0707, -2.1505, 1.0834, -0.9208, -0.3323, -0.9233, 0.0729, 0.2109],
  15780. [ 1.1957, -0.0154, 1.8045, -0.1536, -0.4955, 0.1855, 0.2426, 0.1583],
  15781. [ 0.8495, -0.2346, 1.1987, -0.7389, -0.6078, -0.2832, 0.3070, 0.2877]],
  15782. device='cuda:0', grad_fn=<AddmmBackward>)
  15783. landmarks are: tensor([[[ 0.6059, -0.4249, 1.7420, -0.7321, -0.5931, -0.5523, 0.7307,
  15784. 0.1080],
  15785. [ 0.5861, -0.3929, 1.0570, -1.4314, -0.3286, -1.1235, 0.6182,
  15786. 0.1852],
  15787. [ 0.5355, -0.4335, 1.4266, -0.7155, -0.5769, -0.9310, 0.3275,
  15788. 0.3087],
  15789. [ 0.6537, -0.3719, 1.6979, -0.8617, -0.0169, -1.3621, 0.6926,
  15790. 0.1501],
  15791. [ 0.5792, -0.4052, 1.8214, -0.6587, -0.5384, -0.8924, 0.4381,
  15792. 0.2442],
  15793. [-2.2859, -2.2859, 0.8516, -1.3112, -0.4326, -1.2851, 0.0755,
  15794. 0.2930],
  15795. [ 0.5384, -0.4361, 1.7961, -0.4999, -0.5480, -0.1150, 0.3931,
  15796. 0.2776],
  15797. [ 0.5552, -0.4113, 0.9790, -1.0480, -0.7155, -0.3998, 0.3815,
  15798. 0.3623]]], device='cuda:0')
  15799. loss_train_step before backward: tensor(0.0500, device='cuda:0', grad_fn=<MseLossBackward>)
  15800. loss_train_step after backward: tensor(0.0500, device='cuda:0', grad_fn=<MseLossBackward>)
  15801. loss_train: 0.28310741670429707
  15802. step: 7
  15803. running loss: 0.04044391667204244
  15804.  
  15805. Train Steps: 7/90 Loss: 0.0404 torch.Size([8, 600, 800])
  15806. torch.Size([8, 8])
  15807. tensor([[0.6170, 0.4102, 0.7468, 0.3695, 0.3463, 0.3767, 0.5238, 0.5823],
  15808. [0.6200, 0.4071, 0.7338, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517],
  15809. [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683],
  15810. [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051],
  15811. [0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033],
  15812. [0.6102, 0.4020, 0.8638, 0.3717, 0.3625, 0.5017, 0.6038, 0.5500],
  15813. [0.6148, 0.3996, 0.8488, 0.3867, 0.3488, 0.4067, 0.5863, 0.5000],
  15814. [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461]],
  15815. device='cuda:0', dtype=torch.float64)
  15816. predictions are: tensor([[ 0.7177, -0.2844, 1.2895, -0.7119, -0.6197, -0.5738, 0.1653, 0.3376],
  15817. [ 0.6297, -0.4088, 1.2134, -1.3051, -0.3212, -1.3260, 0.5060, 0.2098],
  15818. [ 0.5482, -0.4027, 1.6650, 0.0851, -0.4617, -0.6359, 0.1926, 0.2705],
  15819. [ 0.6680, -0.3811, 1.8350, -0.1524, -0.4657, -0.6388, 0.6130, 0.1218],
  15820. [ 0.6451, -0.3880, 1.7837, 0.0488, -0.2782, 0.2239, 0.3377, 0.1277],
  15821. [ 0.6347, -0.3732, 1.6429, -0.5697, -0.4933, 0.0440, 0.5640, 0.2210],
  15822. [ 0.7459, -0.3111, 1.7528, -0.4887, -0.6403, -0.4226, 0.3598, 0.0651],
  15823. [ 0.6221, -0.3853, 1.7191, 0.1670, -0.1029, 0.1177, 0.2200, 0.1834]],
  15824. device='cuda:0', grad_fn=<AddmmBackward>)
  15825. landmarks are: tensor([[[ 0.5639, -0.3911, 1.1634, -0.5794, -0.6866, -0.5461, 0.1334,
  15826. 0.4036],
  15827. [ 0.5779, -0.4054, 1.1032, -1.4006, -0.3460, -1.1543, 0.5547,
  15828. 0.2622],
  15829. [ 0.5777, -0.3915, 1.7961, 0.1698, -0.5192, -0.5307, 0.2141,
  15830. 0.3392],
  15831. [ 0.6204, -0.4336, 1.8654, -0.0688, -0.6058, -0.5230, 0.6503,
  15832. 0.0472],
  15833. [ 0.5668, -0.4306, 1.7730, 0.0697, -0.4037, 0.1390, 0.4450,
  15834. 0.0390],
  15835. [ 0.5324, -0.4293, 1.7037, -0.5692, -0.6115, 0.0313, 0.5028,
  15836. 0.2545],
  15837. [ 0.5538, -0.4400, 1.6344, -0.4999, -0.6751, -0.4075, 0.4219,
  15838. 0.0236],
  15839. [ 0.5443, -0.3831, 1.7095, 0.1621, -0.2016, 0.1390, 0.1437,
  15840. 0.2364]]], device='cuda:0')
  15841. loss_train_step before backward: tensor(0.0066, device='cuda:0', grad_fn=<MseLossBackward>)
  15842. loss_train_step after backward: tensor(0.0066, device='cuda:0', grad_fn=<MseLossBackward>)
  15843. loss_train: 0.289715682156384
  15844. step: 8
  15845. running loss: 0.036214460269548
  15846. Train Steps: 8/90 Loss: 0.0362 torch.Size([8, 600, 800])
  15847. torch.Size([8, 8])
  15848. tensor([[0.6202, 0.4066, 0.8746, 0.3376, 0.3717, 0.3090, 0.5842, 0.5165],
  15849. [0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967],
  15850. [ nan, nan, 0.8525, 0.2217, 0.5413, 0.2367, 0.7367, 0.5482],
  15851. [0.6148, 0.4076, 0.8666, 0.4820, 0.4138, 0.5067, 0.5250, 0.5767],
  15852. [0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869],
  15853. [0.6120, 0.4014, 0.6863, 0.2817, 0.3700, 0.2783, 0.5513, 0.5667],
  15854. [ nan, nan, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563],
  15855. [0.6043, 0.4022, 0.6887, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136]],
  15856. device='cuda:0', dtype=torch.float64)
  15857. predictions are: tensor([[ 1.1728, -0.0307, 1.7745, -0.4279, -0.6014, -0.7678, 0.3550, 0.1150],
  15858. [ 1.1965, -0.0361, 1.7375, -0.0554, -0.5987, 0.2345, 0.5089, 0.1112],
  15859. [-1.2754, -1.6184, 1.6297, -0.9376, 0.1132, -0.9360, 0.9832, 0.2434],
  15860. [ 0.9482, -0.2055, 1.7096, 0.2033, -0.3678, 0.2868, 0.2874, 0.2983],
  15861. [ 0.8853, -0.1925, 1.5021, -1.0313, -0.1642, -1.0260, 0.5182, 0.1084],
  15862. [ 0.8353, -0.2145, 1.0419, -0.7815, -0.5458, -0.8351, 0.1290, 0.3244],
  15863. [-0.7671, -1.2577, 1.3239, -0.5202, -0.6450, -0.6157, 0.0111, 0.2329],
  15864. [ 0.9063, -0.2047, 0.9747, -0.9788, -0.5142, -1.0113, 0.2779, 0.1766]],
  15865. device='cuda:0', grad_fn=<AddmmBackward>)
  15866. landmarks are: tensor([[[ 0.5787, -0.4077, 1.7537, -0.7263, -0.5692, -0.8586, 0.4126,
  15867. 0.1000],
  15868. [ 0.4974, -0.4482, 1.6633, -0.3306, -0.6173, 0.1313, 0.2925,
  15869. 0.0081],
  15870. [-2.2859, -2.2859, 1.6517, -1.2620, 0.2141, -1.1928, 1.1166,
  15871. 0.2463],
  15872. [ 0.5539, -0.4031, 1.7168, -0.0594, -0.3748, 0.0543, 0.1390,
  15873. 0.3777],
  15874. [ 0.6127, -0.3944, 1.5189, -1.2467, -0.1323, -1.4622, 0.5646,
  15875. -0.0369],
  15876. [ 0.5410, -0.4321, 0.8838, -0.9849, -0.5769, -1.0003, 0.2603,
  15877. 0.3315],
  15878. [-2.2859, -2.2859, 1.4006, -0.8105, -0.6115, -0.8232, 0.0419,
  15879. 0.2837],
  15880. [ 0.5053, -0.4281, 0.8954, -1.3698, -0.5423, -1.1389, 0.2453,
  15881. 0.0862]]], device='cuda:0')
  15882. loss_train_step before backward: tensor(0.1240, device='cuda:0', grad_fn=<MseLossBackward>)
  15883. loss_train_step after backward: tensor(0.1240, device='cuda:0', grad_fn=<MseLossBackward>)
  15884. loss_train: 0.41375276912003756
  15885. step: 9
  15886. running loss: 0.045972529902226396
  15887. Train Steps: 9/90 Loss: 0.0460 torch.Size([8, 600, 800])
  15888. torch.Size([8, 8])
  15889. tensor([[0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212],
  15890. [ nan, nan, 0.7225, 0.2167, 0.3987, 0.2283, 0.5427, 0.5181],
  15891. [0.6109, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117],
  15892. [0.6213, 0.4001, 0.7712, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183],
  15893. [0.6097, 0.3988, 0.8650, 0.5250, 0.4213, 0.5200, 0.5675, 0.5050],
  15894. [0.6098, 0.3991, 0.8638, 0.4717, 0.4263, 0.4967, 0.5212, 0.5650],
  15895. [0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133],
  15896. [ nan, nan, 0.7515, 0.2708, 0.3987, 0.2267, 0.5162, 0.5567]],
  15897. device='cuda:0', dtype=torch.float64)
  15898. predictions are: tensor([[ 0.9313, -0.2026, 1.4965, -0.8860, -0.4646, -0.8344, 0.7200, 0.1272],
  15899. [-1.7722, -1.9497, 1.0401, -1.0044, -0.4755, -0.9832, 0.1582, 0.2319],
  15900. [ 1.1622, -0.0263, 1.8426, -0.0205, -0.6599, -0.2158, 0.2817, 0.1618],
  15901. [ 1.0712, -0.1175, 1.1899, -1.1806, -0.3783, -1.2676, 0.5125, 0.1718],
  15902. [ 1.0987, -0.0886, 1.7554, 0.2702, -0.3929, 0.3518, 0.2888, 0.1255],
  15903. [ 0.8446, -0.2262, 1.7078, 0.0515, -0.3860, 0.2727, 0.2133, 0.2586],
  15904. [ 1.0712, -0.1029, 1.5134, -0.9648, -0.2017, -1.2557, 0.7378, 0.2042],
  15905. [-1.1009, -1.4880, 1.1267, -0.8914, -0.5174, -0.8465, 0.0194, 0.3053]],
  15906. device='cuda:0', grad_fn=<AddmmBackward>)
  15907. landmarks are: tensor([[[ 0.6125, -0.4345, 1.4308, -1.1384, -0.4213, -1.0031, 0.7190,
  15908. 0.1214],
  15909. [-2.2859, -2.2859, 1.0513, -1.2851, -0.4441, -1.2313, 0.2206,
  15910. 0.1073],
  15911. [ 0.5356, -0.4343, 1.7441, -0.1150, -0.6462, -0.3614, 0.3238,
  15912. 0.0774],
  15913. [ 0.5838, -0.4379, 1.2764, -1.3082, -0.2824, -1.4545, 0.4162,
  15914. 0.1082],
  15915. [ 0.5303, -0.4440, 1.7095, 0.1390, -0.3402, 0.1159, 0.3353,
  15916. 0.0467],
  15917. [ 0.5308, -0.4425, 1.7037, -0.1073, -0.3171, 0.0082, 0.1217,
  15918. 0.3238],
  15919. [ 0.6085, -0.4084, 1.5536, -1.1466, -0.0746, -1.4853, 0.6298,
  15920. 0.0851],
  15921. [-2.2859, -2.2859, 1.1854, -1.0352, -0.4441, -1.2390, 0.0986,
  15922. 0.2853]]], device='cuda:0')
  15923. loss_train_step before backward: tensor(0.0825, device='cuda:0', grad_fn=<MseLossBackward>)
  15924. loss_train_step after backward: tensor(0.0825, device='cuda:0', grad_fn=<MseLossBackward>)
  15925. loss_train: 0.4962860615924001
  15926. step: 10
  15927. running loss: 0.04962860615924001
  15928. Train Steps: 10/90 Loss: 0.0496 torch.Size([8, 600, 800])
  15929. torch.Size([8, 8])
  15930. tensor([[0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285],
  15931. [0.6192, 0.4128, 0.8513, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633],
  15932. [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150],
  15933. [0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136],
  15934. [0.6128, 0.4115, 0.7934, 0.3778, 0.3450, 0.4033, 0.5337, 0.5456],
  15935. [0.6060, 0.3924, 0.8450, 0.5717, 0.4200, 0.5217, 0.5253, 0.4752],
  15936. [ nan, nan, 0.7225, 0.2167, 0.3987, 0.2283, 0.5427, 0.5181],
  15937. [0.6126, 0.4073, 0.8750, 0.5133, 0.3800, 0.4333, 0.4986, 0.5378]],
  15938. device='cuda:0', dtype=torch.float64)
  15939. predictions are: tensor([[ 1.0878, -0.1095, 1.6107, -0.0532, -0.4727, -0.1721, 0.7642, 0.1889],
  15940. [ 0.8507, -0.2066, 1.5986, -0.0028, -0.3621, 0.0362, 0.4501, 0.2864],
  15941. [ 0.8927, -0.2078, 1.7943, -0.2751, -0.5363, -0.4577, 0.5687, 0.1468],
  15942. [ 0.6973, -0.3176, 1.6548, -0.6774, -0.6204, -0.6440, 0.2402, 0.1518],
  15943. [ 0.7442, -0.2523, 1.3724, -0.8019, -0.6210, -0.5154, 0.2633, 0.2536],
  15944. [ 0.6057, -0.3823, 1.5197, 0.0231, -0.3128, -0.0456, 0.1714, 0.1913],
  15945. [-2.1630, -2.2389, 1.0234, -1.2700, -0.4265, -1.2075, 0.2050, 0.1823],
  15946. [ 0.8390, -0.2155, 1.6746, -0.1264, -0.5022, -0.4144, 0.1339, 0.1871]],
  15947. device='cuda:0', grad_fn=<AddmmBackward>)
  15948. landmarks are: tensor([[[ 6.5365e-01, -3.9601e-01, 1.6517e+00, 3.1609e-01, -4.9607e-01,
  15949. 4.6189e-04, 7.6203e-01, 1.5543e-01],
  15950. [ 5.7419e-01, -3.7921e-01, 1.6460e+00, 3.0839e-01, -3.4596e-01,
  15951. 1.4673e-01, 4.1617e-01, 3.1609e-01],
  15952. [ 6.0964e-01, -4.0462e-01, 1.8249e+00, -7.2363e-03, -6.1155e-01,
  15953. -3.5366e-01, 6.1824e-01, 9.2841e-02],
  15954. [ 5.2355e-01, -4.2731e-01, 1.7499e+00, -4.3064e-01, -5.8268e-01,
  15955. -4.6143e-01, 1.6505e-01, 8.6245e-02],
  15956. [ 5.4440e-01, -3.8522e-01, 1.3786e+00, -5.4087e-01, -6.9238e-01,
  15957. -4.2294e-01, 1.7915e-01, 2.3412e-01],
  15958. [ 5.1316e-01, -4.7360e-01, 1.6171e+00, 3.5458e-01, -3.4596e-01,
  15959. 1.2363e-01, 1.4038e-01, -9.1096e-02],
  15960. [-2.2859e+00, -2.2859e+00, 1.0513e+00, -1.2851e+00, -4.4411e-01,
  15961. -1.2313e+00, 2.2057e-01, 1.0729e-01],
  15962. [ 5.4353e-01, -4.0454e-01, 1.7557e+00, 8.5142e-02, -5.3072e-01,
  15963. -2.8437e-01, 1.7213e-02, 1.9805e-01]]], device='cuda:0')
  15964. loss_train_step before backward: tensor(0.0259, device='cuda:0', grad_fn=<MseLossBackward>)
  15965. loss_train_step after backward: tensor(0.0259, device='cuda:0', grad_fn=<MseLossBackward>)
  15966. loss_train: 0.5221390677616
  15967. step: 11
  15968. running loss: 0.047467187978327274
  15969.  
  15970. Train Steps: 11/90 Loss: 0.0475 torch.Size([8, 600, 800])
  15971. torch.Size([8, 8])
  15972. tensor([[0.6256, 0.4199, 0.8638, 0.5800, 0.3987, 0.4383, 0.5600, 0.5950],
  15973. [0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947],
  15974. [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267],
  15975. [0.6125, 0.4035, 0.7825, 0.3100, 0.3463, 0.4900, 0.5832, 0.5637],
  15976. [0.6079, 0.3964, 0.7420, 0.2958, 0.3563, 0.2917, 0.5351, 0.4980],
  15977. [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167],
  15978. [0.6204, 0.4110, 0.7913, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367],
  15979. [0.6216, 0.4167, 0.8588, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667]],
  15980. device='cuda:0', dtype=torch.float64)
  15981. predictions are: tensor([[ 0.3029, -0.5813, 1.6058, 0.2992, -0.4347, -0.2948, 0.2373, 0.4066],
  15982. [ 0.1432, -0.7466, 1.3805, -1.3673, -0.3420, -1.3187, 0.7347, 0.0626],
  15983. [ 0.1686, -0.6956, 1.5923, 0.0621, -0.4416, -0.2541, 0.3519, 0.1528],
  15984. [ 0.2014, -0.6436, 1.2887, -0.8800, -0.6039, 0.0331, 0.3587, 0.2998],
  15985. [ 0.5978, -0.4018, 1.1764, -1.0731, -0.6076, -0.8882, 0.0660, 0.0662],
  15986. [ 0.5845, -0.4183, 1.5968, -0.8787, -0.6331, -0.9194, 0.3063, 0.0992],
  15987. [ 0.5553, -0.4389, 1.4506, -1.0662, -0.4033, -1.0708, 0.5415, 0.1191],
  15988. [ 0.3629, -0.5407, 1.6632, 0.1891, -0.4559, 0.0866, 0.4513, 0.2735]],
  15989. device='cuda:0', grad_fn=<AddmmBackward>)
  15990. landmarks are: tensor([[[ 6.0381e-01, -3.4642e-01, 1.7037e+00, 3.9307e-01, -4.4411e-01,
  15991. -2.6128e-01, 3.0069e-01, 4.6236e-01],
  15992. [ 6.0739e-01, -4.0631e-01, 1.3630e+00, -1.5238e+00, -3.1132e-01,
  15993. -1.2390e+00, 6.1950e-01, -9.2270e-04],
  15994. [ 5.8915e-01, -4.5504e-01, 1.5132e+00, 3.5458e-01, -3.6905e-01,
  15995. -1.5350e-01, 3.8152e-01, 1.4673e-01],
  15996. [ 5.4319e-01, -4.2240e-01, 1.3284e+00, -8.5404e-01, -6.8661e-01,
  15997. -2.2633e-02, 4.0770e-01, 3.1769e-01],
  15998. [ 5.2194e-01, -4.5504e-01, 1.1415e+00, -9.1962e-01, -6.4042e-01,
  15999. -9.3872e-01, 1.8562e-01, 1.4106e-02],
  16000. [ 5.4825e-01, -4.2490e-01, 1.5305e+00, -7.3857e-01, -6.1155e-01,
  16001. -8.6944e-01, 3.3533e-01, 1.0054e-01],
  16002. [ 5.7962e-01, -3.8776e-01, 1.3688e+00, -1.0542e+00, -4.0947e-01,
  16003. -1.1312e+00, 5.8938e-01, 1.9292e-01],
  16004. [ 5.8528e-01, -3.6135e-01, 1.6806e+00, 2.9299e-01, -4.4988e-01,
  16005. 1.0054e-01, 3.8152e-01, 3.3149e-01]]], device='cuda:0')
  16006. loss_train_step before backward: tensor(0.0200, device='cuda:0', grad_fn=<MseLossBackward>)
  16007. loss_train_step after backward: tensor(0.0200, device='cuda:0', grad_fn=<MseLossBackward>)
  16008. loss_train: 0.5421810066327453
  16009. step: 12
  16010. running loss: 0.04518175055272877
  16011. Train Steps: 12/90 Loss: 0.0452 torch.Size([8, 600, 800])
  16012. torch.Size([8, 8])
  16013. tensor([[0.6307, 0.4029, 0.8988, 0.4817, 0.3937, 0.3500, 0.7311, 0.5378],
  16014. [0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5787, 0.5600],
  16015. [ nan, nan, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729],
  16016. [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400],
  16017. [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450],
  16018. [0.6185, 0.4098, 0.8838, 0.4900, 0.4537, 0.5800, 0.6288, 0.5400],
  16019. [0.6206, 0.4001, 0.8900, 0.3933, 0.3588, 0.3567, 0.5837, 0.5083],
  16020. [0.6162, 0.4134, 0.6700, 0.2467, 0.3962, 0.2533, 0.5737, 0.5467]],
  16021. device='cuda:0', dtype=torch.float64)
  16022. predictions are: tensor([[ 0.1219, -0.7227, 1.8549, -0.1560, -0.5318, -0.7085, 0.7696, 0.0714],
  16023. [ 0.8323, -0.2219, 1.6331, -0.1667, -0.2122, 0.0865, 0.1752, 0.2409],
  16024. [-2.4100, -2.4214, 0.8703, -1.1893, -0.4963, -1.1885, 0.1118, 0.2354],
  16025. [ 0.7064, -0.3345, 1.2014, -1.1894, -0.3038, -1.3511, 0.4325, 0.1430],
  16026. [ 0.6720, -0.3316, 1.6616, -0.2564, -0.6401, -0.3004, 0.3643, 0.1793],
  16027. [ 0.5241, -0.4137, 1.6657, -0.1205, -0.2329, 0.3921, 0.3336, 0.2425],
  16028. [ 0.4922, -0.4644, 1.7609, -0.4795, -0.7322, -0.6432, 0.3411, 0.0464],
  16029. [ 0.9622, -0.1731, 0.8906, -1.0965, -0.5967, -1.1497, 0.3895, 0.2532]],
  16030. device='cuda:0', grad_fn=<AddmmBackward>)
  16031. landmarks are: tensor([[[ 0.6273, -0.4249, 1.8654, -0.0611, -0.4672, -0.6693, 1.0910,
  16032. 0.1982],
  16033. [ 0.5505, -0.4207, 1.7095, -0.0534, -0.0509, 0.1050, 0.3873,
  16034. 0.3007],
  16035. [-2.2859, -2.2859, 0.8824, -1.2727, -0.3691, -1.4237, 0.0943,
  16036. 0.3604],
  16037. [ 0.5842, -0.3848, 1.2938, -1.2159, -0.2132, -1.4160, 0.5778,
  16038. 0.2083],
  16039. [ 0.5785, -0.4252, 1.7676, -0.1602, -0.5845, -0.3446, 0.4566,
  16040. 0.2314],
  16041. [ 0.5709, -0.3933, 1.7961, -0.0226, -0.1901, 0.3931, 0.6182,
  16042. 0.2083],
  16043. [ 0.5807, -0.4378, 1.8249, -0.4691, -0.6289, -0.6385, 0.4104,
  16044. 0.0620],
  16045. [ 0.5603, -0.3764, 0.8088, -1.1466, -0.4557, -1.1158, 0.3642,
  16046. 0.2391]]], device='cuda:0')
  16047. loss_train_step before backward: tensor(0.0205, device='cuda:0', grad_fn=<MseLossBackward>)
  16048. loss_train_step after backward: tensor(0.0205, device='cuda:0', grad_fn=<MseLossBackward>)
  16049. loss_train: 0.5627054637297988
  16050. step: 13
  16051. running loss: 0.04328503567152298
  16052. Train Steps: 13/90 Loss: 0.0433 torch.Size([8, 600, 800])
  16053. torch.Size([8, 8])
  16054. tensor([[0.6222, 0.4169, 0.8638, 0.5650, 0.4313, 0.4783, 0.5637, 0.5633],
  16055. [0.6190, 0.4135, 0.8000, 0.4883, 0.3566, 0.3647, 0.5613, 0.5900],
  16056. [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633],
  16057. [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586],
  16058. [0.6339, 0.4102, 0.9088, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390],
  16059. [0.6127, 0.4118, 0.8650, 0.5083, 0.4088, 0.5367, 0.5300, 0.5456],
  16060. [0.6075, 0.4007, 0.8275, 0.4917, 0.4050, 0.5100, 0.5167, 0.5280],
  16061. [0.6286, 0.4060, 0.9188, 0.4333, 0.3675, 0.4167, 0.7034, 0.5528]],
  16062. device='cuda:0', dtype=torch.float64)
  16063. predictions are: tensor([[ 0.4958, -0.4315, 1.4636, 0.0741, -0.4305, -0.2701, 0.1757, 0.2284],
  16064. [ 0.0142, -0.7257, 1.3866, -0.4004, -0.6804, -0.9269, 0.1312, 0.2729],
  16065. [ 0.4388, -0.4977, 1.5452, -0.0589, -0.6108, -0.3443, 0.2965, 0.2244],
  16066. [ 0.1248, -0.7289, 1.5169, -1.3572, -0.3329, -1.3349, 0.9413, 0.0911],
  16067. [ 0.2870, -0.6169, 1.5969, -0.4151, -0.4717, 0.0487, 0.7505, 0.2085],
  16068. [ 0.3186, -0.5650, 1.4995, -0.1835, -0.3716, 0.0049, 0.0432, 0.1541],
  16069. [ 0.1244, -0.6726, 1.3800, -0.3604, -0.4168, -0.0753, -0.0819, 0.1340],
  16070. [ 0.3934, -0.5537, 1.8068, -0.5702, -0.5608, -0.6095, 0.8026, 0.1045]],
  16071. device='cuda:0', grad_fn=<AddmmBackward>)
  16072. landmarks are: tensor([[[ 0.5880, -0.3605, 1.7037, 0.3238, -0.2940, -0.0765, 0.3180,
  16073. 0.3161],
  16074. [ 0.5732, -0.3761, 1.4092, -0.0303, -0.6388, -0.6012, 0.3065,
  16075. 0.4393],
  16076. [ 0.5879, -0.3591, 1.8018, 0.1236, -0.5596, -0.1612, 0.3469,
  16077. 0.3161],
  16078. [ 0.6075, -0.4129, 1.6055, -1.0080, -0.2420, -1.0080, 0.9704,
  16079. 0.2944],
  16080. [ 0.6421, -0.3912, 1.9115, -0.0842, -0.4730, 0.1544, 1.1824,
  16081. 0.2035],
  16082. [ 0.5443, -0.3840, 1.7095, 0.0620, -0.3979, 0.1929, 0.1622,
  16083. 0.2341],
  16084. [ 0.5200, -0.4353, 1.5363, -0.0149, -0.4152, 0.0697, 0.1005,
  16085. 0.1530],
  16086. [ 0.6174, -0.4105, 1.9577, -0.2844, -0.5885, -0.3614, 0.9631,
  16087. 0.2676]]], device='cuda:0')
  16088. loss_train_step before backward: tensor(0.0514, device='cuda:0', grad_fn=<MseLossBackward>)
  16089. loss_train_step after backward: tensor(0.0514, device='cuda:0', grad_fn=<MseLossBackward>)
  16090. loss_train: 0.6141552468761802
  16091. step: 14
  16092. running loss: 0.043868231919727156
  16093. Train Steps: 14/90 Loss: 0.0439 torch.Size([8, 600, 800])
  16094. torch.Size([8, 8])
  16095. tensor([[0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945],
  16096. [0.6148, 0.3996, 0.8488, 0.3867, 0.3488, 0.4067, 0.5863, 0.5000],
  16097. [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5787, 0.5117],
  16098. [0.6163, 0.4114, 0.7650, 0.2017, 0.3763, 0.2867, 0.5631, 0.5071],
  16099. [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760],
  16100. [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133],
  16101. [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114],
  16102. [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083]],
  16103. device='cuda:0', dtype=torch.float64)
  16104. predictions are: tensor([[ 0.1153, -0.7137, 1.8762, -0.4651, -0.3093, -0.7997, 0.9198, 0.2673],
  16105. [ 0.2380, -0.6188, 1.6252, -0.5804, -0.7252, -0.4364, 0.3297, 0.1032],
  16106. [ 0.4029, -0.5418, 1.0334, -1.3869, -0.4907, -1.4458, 0.2220, 0.0935],
  16107. [ 0.4753, -0.5046, 1.0829, -1.3684, -0.6306, -1.0880, 0.3981, 0.0556],
  16108. [ 0.0737, -0.7363, 1.5422, 0.0759, -0.5337, -0.1861, 0.7105, 0.2673],
  16109. [ 0.2822, -0.5751, 1.5885, -0.0442, -0.2840, -0.0404, 0.1796, 0.1596],
  16110. [ 0.2269, -0.6090, 1.6293, -0.0549, -0.2520, 0.0713, 0.2031, 0.1629],
  16111. [-0.0072, -0.7856, 1.5862, -0.1222, -0.6186, -0.3006, 0.3431, 0.3637]],
  16112. device='cuda:0', grad_fn=<AddmmBackward>)
  16113. landmarks are: tensor([[[ 0.6471, -0.3719, 1.9866, -0.3921, -0.3806, -0.5538, 1.0070,
  16114. 0.4600],
  16115. [ 0.5538, -0.4400, 1.6344, -0.4999, -0.6751, -0.4075, 0.4219,
  16116. 0.0236],
  16117. [ 0.5598, -0.4201, 1.1898, -1.3005, -0.3806, -1.3313, 0.3873,
  16118. 0.0774],
  16119. [ 0.5606, -0.3856, 1.2476, -1.3544, -0.5480, -0.9618, 0.3152,
  16120. 0.0562],
  16121. [ 0.6421, -0.4008, 1.6055, 0.2160, -0.5076, -0.0534, 1.1020,
  16122. 0.3745],
  16123. [ 0.5786, -0.4141, 1.7037, 0.1544, -0.1862, 0.0736, 0.4393,
  16124. 0.0851],
  16125. [ 0.5450, -0.4730, 1.7420, 0.1372, -0.1919, 0.2614, 0.4976,
  16126. 0.0764],
  16127. [ 0.5713, -0.4105, 1.7557, 0.0467, -0.6520, -0.2690, 0.3988,
  16128. 0.5239]]], device='cuda:0')
  16129. loss_train_step before backward: tensor(0.0415, device='cuda:0', grad_fn=<MseLossBackward>)
  16130. loss_train_step after backward: tensor(0.0415, device='cuda:0', grad_fn=<MseLossBackward>)
  16131. loss_train: 0.6556800110265613
  16132. step: 15
  16133. running loss: 0.04371200073510408
  16134.  
  16135. Train Steps: 15/90 Loss: 0.0437 torch.Size([8, 600, 800])
  16136. torch.Size([8, 8])
  16137. tensor([[0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896],
  16138. [0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136],
  16139. [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118],
  16140. [0.6204, 0.4013, 0.8075, 0.2400, 0.4313, 0.2050, 0.5800, 0.5150],
  16141. [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092],
  16142. [0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494],
  16143. [0.6161, 0.4076, 0.8900, 0.4667, 0.4125, 0.5917, 0.6262, 0.5367],
  16144. [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683]],
  16145. device='cuda:0', dtype=torch.float64)
  16146. predictions are: tensor([[ 0.1218, -0.6983, 1.2977, -0.7802, -0.6482, -0.9329, 0.1384, 0.3076],
  16147. [ 0.1443, -0.6555, 1.6092, -0.6454, -0.6445, -0.6126, 0.2774, 0.1033],
  16148. [ 0.3040, -0.5442, 1.3830, -0.0120, -0.4510, -0.2879, 0.3900, 0.3919],
  16149. [ 0.0439, -0.7803, 1.1903, -1.3528, -0.3130, -1.4106, 0.5307, 0.1385],
  16150. [ 0.4143, -0.5202, 1.7113, -0.1068, -0.4546, 0.0416, 0.4255, 0.0655],
  16151. [ 0.3109, -0.5906, 1.6933, -0.0469, -0.5008, -0.4294, 0.8705, 0.1147],
  16152. [ 0.1866, -0.6275, 1.6986, -0.3177, -0.2417, 0.3220, 0.5624, 0.1782],
  16153. [ 0.2944, -0.5437, 1.5597, -0.0388, -0.5235, -0.6436, 0.3104, 0.2505]],
  16154. device='cuda:0', grad_fn=<AddmmBackward>)
  16155. landmarks are: tensor([[[ 0.5485, -0.3997, 1.4445, -0.4895, -0.6000, -0.8309, 0.1878,
  16156. 0.4374],
  16157. [ 0.5235, -0.4273, 1.7499, -0.4306, -0.5827, -0.4614, 0.1651,
  16158. 0.0862],
  16159. [ 0.6121, -0.3844, 1.4556, 0.3936, -0.4383, -0.1689, 0.2925,
  16160. 0.5401],
  16161. [ 0.5796, -0.4326, 1.4439, -1.1774, -0.2940, -1.3390, 0.3931,
  16162. 0.0928],
  16163. [ 0.5783, -0.4363, 1.7724, 0.0432, -0.5153, 0.0871, 0.4840,
  16164. 0.0663],
  16165. [ 0.6224, -0.4032, 1.5940, 0.2930, -0.5769, -0.2690, 0.8900,
  16166. 0.2516],
  16167. [ 0.5598, -0.4032, 1.8249, -0.1304, -0.3806, 0.4470, 0.6067,
  16168. 0.1929],
  16169. [ 0.5777, -0.3915, 1.7961, 0.1698, -0.5192, -0.5307, 0.2141,
  16170. 0.3392]]], device='cuda:0')
  16171. loss_train_step before backward: tensor(0.0385, device='cuda:0', grad_fn=<MseLossBackward>)
  16172. loss_train_step after backward: tensor(0.0385, device='cuda:0', grad_fn=<MseLossBackward>)
  16173. loss_train: 0.6941816275939345
  16174. step: 16
  16175. running loss: 0.04338635172462091
  16176. Train Steps: 16/90 Loss: 0.0434 torch.Size([8, 600, 800])
  16177. torch.Size([8, 8])
  16178. tensor([[0.6228, 0.4119, 0.7938, 0.2233, 0.4674, 0.1773, 0.6188, 0.5433],
  16179. [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133],
  16180. [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383],
  16181. [0.6165, 0.4106, 0.7575, 0.1733, 0.3838, 0.2650, 0.5680, 0.5116],
  16182. [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167],
  16183. [0.6190, 0.4135, 0.8000, 0.4883, 0.3566, 0.3647, 0.5613, 0.5900],
  16184. [0.6252, 0.4158, 0.8988, 0.4083, 0.3788, 0.4783, 0.6225, 0.5633],
  16185. [0.6250, 0.4146, 0.8838, 0.3933, 0.3588, 0.4283, 0.6162, 0.5367]],
  16186. device='cuda:0', dtype=torch.float64)
  16187. predictions are: tensor([[ 2.7160e-01, -6.1288e-01, 1.3024e+00, -1.0790e+00, -2.4634e-01,
  16188. -1.3686e+00, 5.0634e-01, 1.6069e-01],
  16189. [ 2.6692e-01, -5.8733e-01, 1.6549e+00, 1.7750e-01, -2.6255e-01,
  16190. 1.4141e-01, 3.5143e-01, 1.6915e-01],
  16191. [ 3.3903e-01, -5.6775e-01, 1.0713e+00, -1.1558e+00, -2.5916e-01,
  16192. -1.4428e+00, 3.9940e-01, 2.0511e-01],
  16193. [ 4.3410e-01, -5.4499e-01, 1.1180e+00, -1.2465e+00, -5.6629e-01,
  16194. -1.0447e+00, 5.5686e-01, 4.9847e-02],
  16195. [ 2.5437e-01, -6.0803e-01, 1.7518e+00, 1.5235e-01, -4.2358e-01,
  16196. 4.0426e-02, 5.6769e-01, 1.3766e-01],
  16197. [ 6.4946e-04, -7.1879e-01, 1.5425e+00, 7.2002e-03, -6.0888e-01,
  16198. -6.5312e-01, 2.5940e-01, 3.6687e-01],
  16199. [ 1.4826e-01, -6.5294e-01, 1.8039e+00, -3.0606e-01, -5.9294e-01,
  16200. -5.2833e-02, 6.3769e-01, 2.8626e-01],
  16201. [ 8.7252e-02, -6.8236e-01, 1.6944e+00, -4.0593e-01, -6.6378e-01,
  16202. -3.7071e-01, 5.3112e-01, 2.8553e-01]], device='cuda:0',
  16203. grad_fn=<AddmmBackward>)
  16204. landmarks are: tensor([[[ 0.5908, -0.3832, 1.3804, -1.2543, -0.1270, -1.4671, 0.5721,
  16205. 0.2237],
  16206. [ 0.5786, -0.4141, 1.7037, 0.1544, -0.1862, 0.0736, 0.4393,
  16207. 0.0851],
  16208. [ 0.5911, -0.3984, 1.1956, -1.0850, -0.0804, -1.5392, 0.4393,
  16209. 0.2006],
  16210. [ 0.5618, -0.3895, 1.2129, -1.4853, -0.5134, -1.0619, 0.3378,
  16211. 0.0772],
  16212. [ 0.5828, -0.4066, 1.7557, 0.0774, -0.3691, -0.0226, 0.4277,
  16213. 0.1005],
  16214. [ 0.5732, -0.3761, 1.4092, -0.0303, -0.6388, -0.6012, 0.3065,
  16215. 0.4393],
  16216. [ 0.6017, -0.3654, 1.8654, -0.3998, -0.5365, -0.0765, 0.5894,
  16217. 0.3161],
  16218. [ 0.6009, -0.3710, 1.7961, -0.4691, -0.6289, -0.3075, 0.5605,
  16219. 0.1929]]], device='cuda:0')
  16220. loss_train_step before backward: tensor(0.0320, device='cuda:0', grad_fn=<MseLossBackward>)
  16221. loss_train_step after backward: tensor(0.0320, device='cuda:0', grad_fn=<MseLossBackward>)
  16222. loss_train: 0.7261817371472716
  16223. step: 17
  16224. running loss: 0.04271657277336892
  16225. Train Steps: 17/90 Loss: 0.0427 torch.Size([8, 600, 800])
  16226. torch.Size([8, 8])
  16227. tensor([[0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882],
  16228. [0.6122, 0.4006, 0.8850, 0.4217, 0.4088, 0.5517, 0.6063, 0.5517],
  16229. [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207],
  16230. [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402],
  16231. [ nan, nan, 0.6900, 0.1917, 0.3937, 0.2367, 0.5240, 0.5246],
  16232. [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118],
  16233. [0.6100, 0.4016, 0.8600, 0.5067, 0.4612, 0.5233, 0.5086, 0.5519],
  16234. [0.6148, 0.4076, 0.8666, 0.4820, 0.4138, 0.5067, 0.5250, 0.5767]],
  16235. device='cuda:0', dtype=torch.float64)
  16236. predictions are: tensor([[ 0.9092, -0.1473, 1.7510, -0.4465, -0.6588, -0.6622, 0.2803, 0.0766],
  16237. [ 0.6855, -0.2995, 1.7302, -0.2893, -0.4286, 0.1628, 0.6664, 0.2222],
  16238. [ 0.5960, -0.4104, 1.4210, -1.1680, -0.3207, -1.3542, 0.9038, 0.1072],
  16239. [ 0.5793, -0.3802, 1.6687, 0.1535, -0.4748, -0.2674, 0.5813, 0.1793],
  16240. [-2.6504, -2.5917, 1.0145, -1.4165, -0.4417, -1.2874, 0.2440, 0.2193],
  16241. [ 0.6755, -0.2863, 1.4774, 0.1994, -0.5063, -0.2944, 0.5160, 0.4400],
  16242. [ 0.4113, -0.4544, 1.6717, 0.0710, -0.1949, -0.0334, 0.2397, 0.2295],
  16243. [ 0.6982, -0.2907, 1.6843, -0.0253, -0.4022, 0.0244, 0.4109, 0.3028]],
  16244. device='cuda:0', grad_fn=<AddmmBackward>)
  16245. landmarks are: tensor([[[ 0.5595, -0.3988, 1.7672, -0.4460, -0.5538, -0.5384, 0.0828,
  16246. -0.0310],
  16247. [ 0.5417, -0.4355, 1.8018, -0.3383, -0.3979, 0.2622, 0.5143,
  16248. 0.2622],
  16249. [ 0.6132, -0.4248, 1.5247, -1.1620, -0.2882, -1.3159, 0.6545,
  16250. 0.1193],
  16251. [ 0.5780, -0.4565, 1.6221, 0.2532, -0.3728, -0.1718, 0.4357,
  16252. 0.2091],
  16253. [-2.2859, -2.2859, 0.9012, -1.4006, -0.4672, -1.1928, 0.1342,
  16254. 0.1373],
  16255. [ 0.6121, -0.3844, 1.4556, 0.3936, -0.4383, -0.1689, 0.2925,
  16256. 0.5401],
  16257. [ 0.5318, -0.4310, 1.6864, 0.0543, -0.1554, 0.1313, 0.0635,
  16258. 0.2634],
  16259. [ 0.5539, -0.4031, 1.7168, -0.0594, -0.3748, 0.0543, 0.1390,
  16260. 0.3777]]], device='cuda:0')
  16261. loss_train_step before backward: tensor(0.0168, device='cuda:0', grad_fn=<MseLossBackward>)
  16262. loss_train_step after backward: tensor(0.0168, device='cuda:0', grad_fn=<MseLossBackward>)
  16263. loss_train: 0.7429619831964374
  16264. step: 18
  16265. running loss: 0.04127566573313541
  16266. Train Steps: 18/90 Loss: 0.0413 torch.Size([8, 600, 800])
  16267. torch.Size([8, 8])
  16268. tensor([[0.6255, 0.4017, 0.8688, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901],
  16269. [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5637, 0.5633],
  16270. [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600],
  16271. [0.6138, 0.4101, 0.8800, 0.5083, 0.4637, 0.5950, 0.5587, 0.5077],
  16272. [0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938],
  16273. [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351],
  16274. [0.6250, 0.4054, 0.8770, 0.4723, 0.4662, 0.5367, 0.6162, 0.5433],
  16275. [0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6188, 0.5433]],
  16276. device='cuda:0', dtype=torch.float64)
  16277. predictions are: tensor([[ 0.1537, -0.7204, 1.6356, -0.8018, -0.6451, -0.8134, 0.7579, 0.1320],
  16278. [ 0.5381, -0.3980, 1.6382, -0.0354, -0.6471, -0.7040, 0.4174, 0.4160],
  16279. [ 0.3666, -0.4972, 1.6796, -0.2601, -0.1986, 0.0435, 0.4511, 0.3033],
  16280. [ 0.2557, -0.5845, 1.7455, 0.0321, -0.1672, 0.1011, 0.3273, 0.2243],
  16281. [ 0.2479, -0.6265, 1.6954, 0.2720, -0.5356, -0.2905, 0.6293, 0.1359],
  16282. [ 0.5597, -0.4278, 1.1547, -1.2702, -0.3355, -1.5681, 0.3867, 0.1769],
  16283. [ 0.3058, -0.5402, 1.7025, -0.1812, -0.2044, 0.0503, 0.5050, 0.2742],
  16284. [ 0.2479, -0.6194, 1.8035, -0.1138, -0.5522, -0.0488, 0.5325, 0.2004]],
  16285. device='cuda:0', grad_fn=<AddmmBackward>)
  16286. landmarks are: tensor([[[ 0.6031, -0.4307, 1.7268, -0.8002, -0.6058, -0.6462, 0.6442,
  16287. -0.0220],
  16288. [ 0.5902, -0.3493, 1.7961, -0.0072, -0.5942, -0.5615, 0.3180,
  16289. 0.3161],
  16290. [ 0.5652, -0.4325, 1.7388, -0.1429, -0.0688, 0.3469, 0.4970,
  16291. 0.3007],
  16292. [ 0.5491, -0.3918, 1.7788, 0.0620, -0.1439, 0.4624, 0.2946,
  16293. 0.0592],
  16294. [ 0.6395, -0.4213, 1.7037, 0.3623, -0.4326, -0.1073, 0.6560,
  16295. -0.0049],
  16296. [ 0.5664, -0.4321, 1.2862, -1.0003, -0.2189, -1.4608, 0.3883,
  16297. 0.1855],
  16298. [ 0.6009, -0.4135, 1.7651, -0.1043, -0.1323, 0.1929, 0.5605,
  16299. 0.2237],
  16300. [ 0.5773, -0.4105, 1.8192, -0.1304, -0.5307, 0.0467, 0.5721,
  16301. 0.2237]]], device='cuda:0')
  16302. loss_train_step before backward: tensor(0.0247, device='cuda:0', grad_fn=<MseLossBackward>)
  16303. loss_train_step after backward: tensor(0.0247, device='cuda:0', grad_fn=<MseLossBackward>)
  16304. loss_train: 0.7676935633644462
  16305. step: 19
  16306. running loss: 0.04040492438760243
  16307.  
  16308. Train Steps: 19/90 Loss: 0.0404 torch.Size([8, 600, 800])
  16309. torch.Size([8, 8])
  16310. tensor([[0.6113, 0.4006, 0.8700, 0.5350, 0.3638, 0.3767, 0.5097, 0.4882],
  16311. [ nan, nan, 0.8850, 0.2817, 0.5112, 0.2183, 0.7184, 0.5436],
  16312. [0.6266, 0.4067, 0.8588, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267],
  16313. [0.6252, 0.4158, 0.8988, 0.4083, 0.3788, 0.4783, 0.6225, 0.5633],
  16314. [0.6200, 0.4071, 0.7338, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517],
  16315. [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100],
  16316. [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334],
  16317. [ nan, nan, 0.6688, 0.2513, 0.4113, 0.2117, 0.5193, 0.5933]],
  16318. device='cuda:0', dtype=torch.float64)
  16319. predictions are: tensor([[ 0.5391, -0.4111, 1.8399, 0.5539, -0.5385, -0.3105, 0.1539, 0.1573],
  16320. [-0.4865, -1.1033, 1.7536, -0.9170, 0.0708, -1.0576, 1.1059, 0.2863],
  16321. [ 0.7090, -0.2659, 1.6790, -0.7104, -0.3165, -0.8219, 0.6182, 0.2112],
  16322. [ 1.0314, -0.0671, 1.9320, 0.0433, -0.5273, 0.1994, 0.5365, 0.3001],
  16323. [ 0.7391, -0.2983, 1.1664, -1.0349, -0.3136, -0.9762, 0.4261, 0.2512],
  16324. [ 0.9982, -0.1675, 1.8577, 0.0253, -0.5594, -0.2516, 0.6394, 0.1182],
  16325. [ 0.6368, -0.3559, 1.2847, -0.8534, -0.4753, -0.6293, 0.4588, 0.2208],
  16326. [-2.1450, -2.2038, 0.8288, -1.1550, -0.3983, -1.0886, 0.1174, 0.3643]],
  16327. device='cuda:0', grad_fn=<AddmmBackward>)
  16328. landmarks are: tensor([[[ 0.5379, -0.4358, 1.7326, 0.1852, -0.6058, -0.5461, 0.0684,
  16329. -0.0310],
  16330. [-2.2859, -2.2859, 1.8018, -0.9849, 0.0755, -1.2774, 1.0326,
  16331. 0.2249],
  16332. [ 0.6084, -0.4076, 1.6806, -0.9618, -0.2998, -0.9695, 0.6356,
  16333. 0.1467],
  16334. [ 0.6017, -0.3654, 1.8654, -0.3998, -0.5365, -0.0765, 0.5894,
  16335. 0.3161],
  16336. [ 0.5779, -0.4054, 1.1032, -1.4006, -0.3460, -1.1543, 0.5547,
  16337. 0.2622],
  16338. [ 0.6220, -0.4436, 1.8711, -0.3691, -0.6173, -0.4999, 0.6702,
  16339. 0.0697],
  16340. [ 0.5730, -0.4562, 1.2195, -1.2440, -0.5497, -0.7711, 0.5704,
  16341. 0.1779],
  16342. [-2.2859, -2.2859, 0.8033, -1.1250, -0.3864, -1.3082, 0.1126,
  16343. 0.4543]]], device='cuda:0')
  16344. loss_train_step before backward: tensor(0.1029, device='cuda:0', grad_fn=<MseLossBackward>)
  16345. loss_train_step after backward: tensor(0.1029, device='cuda:0', grad_fn=<MseLossBackward>)
  16346. loss_train: 0.8706217845901847
  16347. step: 20
  16348. running loss: 0.04353108922950923
  16349. Train Steps: 20/90 Loss: 0.0435 torch.Size([8, 600, 800])
  16350. torch.Size([8, 8])
  16351. tensor([[0.6163, 0.4114, 0.7650, 0.2017, 0.3763, 0.2867, 0.5631, 0.5071],
  16352. [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567],
  16353. [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207],
  16354. [0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162],
  16355. [0.6274, 0.4003, 0.8638, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661],
  16356. [0.6147, 0.4112, 0.7988, 0.3200, 0.3775, 0.2767, 0.5150, 0.5550],
  16357. [0.6329, 0.4196, 0.9238, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748],
  16358. [0.6353, 0.4128, 0.9138, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991]],
  16359. device='cuda:0', dtype=torch.float64)
  16360. predictions are: tensor([[ 0.5395, -0.4382, 1.1486, -1.2259, -0.4905, -0.8644, 0.3337, 0.1198],
  16361. [ 0.6198, -0.3472, 1.6669, -0.0291, -0.5057, -0.2221, 0.2427, 0.3126],
  16362. [ 0.4126, -0.5339, 1.4331, -1.0436, -0.2162, -1.1185, 0.6102, 0.1433],
  16363. [ 0.5501, -0.4162, 1.8176, -0.2285, -0.5117, 0.0819, 0.6617, 0.2014],
  16364. [ 0.6478, -0.3610, 1.6992, 0.5170, -0.4537, 0.1087, 0.4591, 0.1850],
  16365. [ 0.1851, -0.6295, 1.3142, -0.7649, -0.5219, -0.7357, -0.0488, 0.2572],
  16366. [ 0.2418, -0.6364, 1.9933, 0.0325, -0.2490, -0.7881, 0.7903, 0.2832],
  16367. [-0.4642, -1.0908, 1.9517, -0.5161, -0.0796, -0.6657, 1.0323, 0.3870]],
  16368. device='cuda:0', grad_fn=<AddmmBackward>)
  16369. landmarks are: tensor([[[ 0.5606, -0.3856, 1.2476, -1.3544, -0.5480, -0.9618, 0.3152,
  16370. 0.0562],
  16371. [ 0.5432, -0.4336, 1.7095, -0.1766, -0.5942, -0.4845, 0.3007,
  16372. 0.2853],
  16373. [ 0.6132, -0.4248, 1.5247, -1.1620, -0.2882, -1.3159, 0.6545,
  16374. 0.1193],
  16375. [ 0.6059, -0.4442, 1.8921, -0.3422, -0.6230, -0.1461, 0.7489,
  16376. 0.0985],
  16377. [ 0.6120, -0.4371, 1.7037, 0.4701, -0.5827, -0.0226, 0.5354,
  16378. -0.1331],
  16379. [ 0.5532, -0.3864, 1.4035, -0.8079, -0.5423, -1.0080, 0.0928,
  16380. 0.2776],
  16381. [ 0.6372, -0.3479, 1.9808, -0.1689, -0.3171, -0.9310, 1.1057,
  16382. 0.3692],
  16383. [ 0.6487, -0.3792, 1.9346, -0.6539, -0.1208, -0.7848, 1.0143,
  16384. 0.4814]]], device='cuda:0')
  16385. loss_train_step before backward: tensor(0.0475, device='cuda:0', grad_fn=<MseLossBackward>)
  16386. loss_train_step after backward: tensor(0.0475, device='cuda:0', grad_fn=<MseLossBackward>)
  16387. loss_train: 0.9181526629254222
  16388. step: 21
  16389. running loss: 0.04372155537740106
  16390. Train Steps: 21/90 Loss: 0.0437 torch.Size([8, 600, 800])
  16391. torch.Size([8, 8])
  16392. tensor([[0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895],
  16393. [0.6254, 0.3993, 0.8988, 0.4767, 0.3987, 0.5517, 0.6955, 0.5285],
  16394. [0.6142, 0.4127, 0.7575, 0.3067, 0.3438, 0.4383, 0.5778, 0.5207],
  16395. [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578],
  16396. [0.6205, 0.4081, 0.8950, 0.4017, 0.3788, 0.4700, 0.5963, 0.5667],
  16397. [0.6112, 0.4029, 0.8638, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567],
  16398. [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285],
  16399. [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356]],
  16400. device='cuda:0', dtype=torch.float64)
  16401. predictions are: tensor([[ 0.0581, -0.7464, 1.3099, -1.2809, -0.1847, -1.5320, 0.3973, 0.1047],
  16402. [ 0.4789, -0.4865, 2.0377, 0.2317, -0.4262, 0.3754, 0.9245, 0.2274],
  16403. [ 0.2831, -0.5276, 1.3031, -0.8277, -0.5686, -0.3106, 0.3127, 0.3282],
  16404. [-0.0833, -0.8177, 1.0154, -1.2084, -0.3860, -1.3190, 0.2106, 0.2604],
  16405. [ 0.7950, -0.2507, 1.9152, -0.1057, -0.5803, 0.0395, 0.5121, 0.2700],
  16406. [ 0.3878, -0.5124, 1.9157, 0.1641, -0.0578, 0.0620, 0.4092, 0.2649],
  16407. [ 0.6601, -0.3920, 1.8787, 0.6277, -0.4979, 0.0237, 0.6823, 0.2055],
  16408. [ 0.4764, -0.4623, 1.5466, -0.8331, -0.2431, -1.1635, 0.6208, 0.2329]],
  16409. device='cuda:0', grad_fn=<AddmmBackward>)
  16410. landmarks are: tensor([[[ 5.8284e-01, -4.4175e-01, 1.2476e+00, -1.3929e+00, -1.7275e-01,
  16411. -1.5700e+00, 4.6937e-01, -2.4798e-02],
  16412. [ 6.0260e-01, -4.4175e-01, 1.8654e+00, -8.4219e-02, -4.4411e-01,
  16413. 2.6220e-01, 9.2654e-01, 1.5543e-01],
  16414. [ 5.5087e-01, -3.7983e-01, 1.2129e+00, -8.6944e-01, -6.9815e-01,
  16415. -2.6128e-01, 3.8302e-01, 1.1931e-01],
  16416. [ 5.2315e-01, -4.1886e-01, 9.3580e-01, -1.4006e+00, -3.9215e-01,
  16417. -1.3698e+00, 2.5553e-01, 2.9064e-01],
  16418. [ 5.7991e-01, -4.0115e-01, 1.8480e+00, -4.3064e-01, -5.3649e-01,
  16419. -1.1501e-01, 4.6813e-01, 3.3149e-01],
  16420. [ 5.3712e-01, -4.2517e-01, 1.7037e+00, -6.8822e-02, -3.4180e-02,
  16421. 6.2048e-02, 3.7575e-01, 2.8530e-01],
  16422. [ 6.5365e-01, -3.9601e-01, 1.6517e+00, 3.1609e-01, -4.9607e-01,
  16423. 4.6189e-04, 7.6203e-01, 1.5543e-01],
  16424. [ 6.1742e-01, -4.2249e-01, 1.4975e+00, -1.1709e+00, -3.1736e-01,
  16425. -1.1806e+00, 6.5391e-01, 1.8793e-01]]], device='cuda:0')
  16426. loss_train_step before backward: tensor(0.0307, device='cuda:0', grad_fn=<MseLossBackward>)
  16427. loss_train_step after backward: tensor(0.0307, device='cuda:0', grad_fn=<MseLossBackward>)
  16428. loss_train: 0.9488857751712203
  16429. step: 22
  16430. running loss: 0.043131171598691835
  16431. Train Steps: 22/90 Loss: 0.0431 torch.Size([8, 600, 800])
  16432. torch.Size([8, 8])
  16433. tensor([[0.6259, 0.4133, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947],
  16434. [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882],
  16435. [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208],
  16436. [0.6143, 0.4034, 0.8800, 0.4833, 0.4512, 0.5367, 0.5289, 0.5097],
  16437. [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117],
  16438. [0.6221, 0.4107, 0.7788, 0.3033, 0.3950, 0.2817, 0.6075, 0.5517],
  16439. [0.6127, 0.4084, 0.8700, 0.4467, 0.3987, 0.4317, 0.5013, 0.5471],
  16440. [0.6161, 0.4024, 0.8662, 0.4683, 0.4935, 0.5364, 0.6063, 0.5567]],
  16441. device='cuda:0', dtype=torch.float64)
  16442. predictions are: tensor([[ 0.4515, -0.4720, 1.4325, -1.0735, -0.0188, -1.3452, 0.6986, 0.1841],
  16443. [ 0.6504, -0.3483, 1.7911, -0.2205, -0.5954, -0.4333, 0.2837, 0.1217],
  16444. [ 0.1686, -0.6797, 1.4327, -0.8897, -0.3231, -1.0884, 0.6743, 0.1728],
  16445. [ 0.2799, -0.5739, 1.7917, 0.1795, -0.2080, 0.2609, 0.3030, 0.1324],
  16446. [ 0.0126, -0.7755, 1.7338, -0.6765, -0.5620, -0.7782, 0.6809, 0.1863],
  16447. [ 0.5867, -0.3887, 1.3940, -0.7872, -0.4405, -0.8869, 0.5255, 0.3176],
  16448. [ 0.5458, -0.4143, 1.7265, -0.0424, -0.4915, -0.1369, 0.3132, 0.2854],
  16449. [ 0.4597, -0.4834, 1.8259, 0.0604, -0.1184, 0.2290, 0.6608, 0.2708]],
  16450. device='cuda:0', grad_fn=<AddmmBackward>)
  16451. landmarks are: tensor([[[ 6.0514e-01, -3.7714e-01, 1.5016e+00, -1.2159e+00, 3.5104e-02,
  16452. -1.5777e+00, 6.0099e-01, -9.2270e-04],
  16453. [ 5.5953e-01, -3.9877e-01, 1.7672e+00, -4.4604e-01, -5.5381e-01,
  16454. -5.3841e-01, 8.2802e-02, -3.0981e-02],
  16455. [ 5.8851e-01, -4.4288e-01, 1.4266e+00, -9.9261e-01, -4.3834e-01,
  16456. -1.2313e+00, 4.2276e-01, 1.1948e-01],
  16457. [ 5.5155e-01, -4.2249e-01, 1.7788e+00, -5.3426e-02, -2.0162e-01,
  16458. 1.9292e-01, 1.5683e-01, 6.8210e-02],
  16459. [ 5.7991e-01, -4.3295e-01, 1.7210e+00, -7.6936e-01, -5.7113e-01,
  16460. -8.7714e-01, 3.9885e-01, 7.7444e-02],
  16461. [ 5.8747e-01, -3.8876e-01, 1.3111e+00, -8.8483e-01, -4.6143e-01,
  16462. -9.8491e-01, 5.2009e-01, 2.6220e-01],
  16463. [ 5.4405e-01, -3.9969e-01, 1.7326e+00, -2.2279e-01, -4.4411e-01,
  16464. -2.9207e-01, 2.9551e-02, 2.4088e-01],
  16465. [ 5.5978e-01, -4.2731e-01, 1.7152e+00, -1.2271e-01, -6.4698e-03,
  16466. 1.9169e-01, 5.1432e-01, 2.8530e-01]]], device='cuda:0')
  16467. loss_train_step before backward: tensor(0.0256, device='cuda:0', grad_fn=<MseLossBackward>)
  16468. loss_train_step after backward: tensor(0.0256, device='cuda:0', grad_fn=<MseLossBackward>)
  16469. loss_train: 0.974448068998754
  16470. step: 23
  16471. running loss: 0.042367307347771915
  16472.  
  16473. Train Steps: 23/90 Loss: 0.0424 torch.Size([8, 600, 800])
  16474. torch.Size([8, 8])
  16475. tensor([[0.6202, 0.4079, 0.8025, 0.2500, 0.3763, 0.3217, 0.6125, 0.5533],
  16476. [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283],
  16477. [0.6154, 0.4112, 0.7037, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600],
  16478. [0.6076, 0.3953, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954],
  16479. [0.6234, 0.4179, 0.7825, 0.3450, 0.3813, 0.2867, 0.5675, 0.5617],
  16480. [0.6204, 0.4110, 0.7913, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367],
  16481. [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6188, 0.5283],
  16482. [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679]],
  16483. device='cuda:0', dtype=torch.float64)
  16484. predictions are: tensor([[ 0.5464, -0.4343, 1.5575, -0.8826, -0.4123, -0.6552, 0.5500, 0.2263],
  16485. [ 0.8880, -0.2405, 1.9189, -0.1182, -0.5029, -0.3676, 0.4924, 0.2121],
  16486. [ 0.0587, -0.7420, 0.9993, -1.1595, -0.2452, -1.3513, 0.2003, 0.2672],
  16487. [ 0.3428, -0.5721, 1.6978, -0.4107, -0.5528, -0.2838, 0.3479, 0.0575],
  16488. [ 0.1762, -0.6208, 1.4553, -0.6422, -0.3622, -0.7972, 0.3625, 0.3719],
  16489. [ 0.4635, -0.4896, 1.5542, -0.8813, -0.2341, -1.0112, 0.6000, 0.1013],
  16490. [ 0.6053, -0.4251, 1.9318, -0.4953, -0.3275, -0.9114, 0.6566, 0.1574],
  16491. [ 0.3713, -0.5982, 1.8847, 0.3658, -0.2720, 0.4730, 0.9611, 0.2410]],
  16492. device='cuda:0', grad_fn=<AddmmBackward>)
  16493. landmarks are: tensor([[[ 0.5788, -0.4020, 1.4208, -1.1312, -0.5480, -0.8002, 0.5432,
  16494. 0.2699],
  16495. [ 0.5880, -0.3887, 1.8423, -0.3306, -0.6231, -0.5230, 0.4046,
  16496. 0.1544],
  16497. [ 0.5565, -0.3864, 0.9647, -1.2159, -0.3286, -1.4391, 0.1852,
  16498. 0.3007],
  16499. [ 0.5206, -0.4603, 1.4670, -0.4768, -0.6577, -0.3998, 0.1836,
  16500. 0.0021],
  16501. [ 0.5935, -0.3558, 1.3284, -0.6924, -0.5249, -0.9618, 0.3353,
  16502. 0.3084],
  16503. [ 0.5796, -0.3878, 1.3688, -1.0542, -0.4095, -1.1312, 0.5894,
  16504. 0.1929],
  16505. [ 0.6062, -0.3778, 1.8423, -0.6462, -0.4383, -1.0465, 0.5721,
  16506. 0.1544],
  16507. [ 0.6257, -0.4249, 1.5998, 0.1236, -0.3806, 0.3084, 0.9887,
  16508. 0.3371]]], device='cuda:0')
  16509. loss_train_step before backward: tensor(0.0266, device='cuda:0', grad_fn=<MseLossBackward>)
  16510. loss_train_step after backward: tensor(0.0266, device='cuda:0', grad_fn=<MseLossBackward>)
  16511. loss_train: 1.001086718402803
  16512. step: 24
  16513. running loss: 0.04171194660011679
  16514. Train Steps: 24/90 Loss: 0.0417 torch.Size([8, 600, 800])
  16515. torch.Size([8, 8])
  16516. tensor([[0.6139, 0.4019, 0.7137, 0.2150, 0.4375, 0.1533, 0.5293, 0.5006],
  16517. [0.6250, 0.4146, 0.8838, 0.3933, 0.3588, 0.4283, 0.6162, 0.5367],
  16518. [0.6250, 0.4054, 0.8770, 0.4723, 0.4662, 0.5367, 0.6162, 0.5433],
  16519. [0.6068, 0.3963, 0.8650, 0.4317, 0.4037, 0.5083, 0.5253, 0.4999],
  16520. [0.6268, 0.4052, 0.8175, 0.2250, 0.4688, 0.1917, 0.6375, 0.5267],
  16521. [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5787, 0.5117],
  16522. [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083],
  16523. [0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200]],
  16524. device='cuda:0', dtype=torch.float64)
  16525. predictions are: tensor([[ 0.1870, -0.6556, 1.0506, -1.2502, -0.2470, -1.4751, 0.2263, 0.1088],
  16526. [ 0.9500, -0.1776, 1.8741, -0.2319, -0.6673, -0.1799, 0.6115, 0.2558],
  16527. [ 0.5001, -0.4606, 1.8854, 0.0563, -0.1951, 0.3606, 0.6289, 0.2175],
  16528. [ 0.3553, -0.5747, 1.8311, -0.1599, -0.5016, 0.1822, 0.3915, 0.0840],
  16529. [ 0.4626, -0.5170, 1.5021, -1.0981, -0.1401, -1.4030, 0.7077, 0.1556],
  16530. [ 0.5643, -0.4305, 1.2420, -1.1086, -0.3508, -1.2393, 0.2733, 0.1198],
  16531. [ 0.5204, -0.5165, 1.8178, 0.2130, -0.5849, -0.0536, 0.5791, 0.3442],
  16532. [ 0.3039, -0.5933, 1.8124, -0.8292, -0.2101, -1.0178, 0.6856, 0.2138]],
  16533. device='cuda:0', grad_fn=<AddmmBackward>)
  16534. landmarks are: tensor([[[ 0.5496, -0.4295, 1.0109, -1.2928, -0.2651, -1.5777, 0.1589,
  16535. 0.0261],
  16536. [ 0.6009, -0.3710, 1.7961, -0.4691, -0.6289, -0.3075, 0.5605,
  16537. 0.1929],
  16538. [ 0.6009, -0.4135, 1.7651, -0.1043, -0.1323, 0.1929, 0.5605,
  16539. 0.2237],
  16540. [ 0.5168, -0.4556, 1.7095, -0.2921, -0.4210, 0.0620, 0.1404,
  16541. 0.0231],
  16542. [ 0.6092, -0.4143, 1.4901, -1.2467, -0.1208, -1.4006, 0.6587,
  16543. 0.1467],
  16544. [ 0.5598, -0.4201, 1.1898, -1.3005, -0.3806, -1.3313, 0.3873,
  16545. 0.0774],
  16546. [ 0.5713, -0.4105, 1.7557, 0.0467, -0.6520, -0.2690, 0.3988,
  16547. 0.5239],
  16548. [ 0.6089, -0.3948, 1.7383, -0.8617, -0.2536, -1.2390, 0.6009,
  16549. 0.1159]]], device='cuda:0')
  16550. loss_train_step before backward: tensor(0.0186, device='cuda:0', grad_fn=<MseLossBackward>)
  16551. loss_train_step after backward: tensor(0.0186, device='cuda:0', grad_fn=<MseLossBackward>)
  16552. loss_train: 1.0197177669033408
  16553. step: 25
  16554. running loss: 0.040788710676133634
  16555. Train Steps: 25/90 Loss: 0.0408 torch.Size([8, 600, 800])
  16556. torch.Size([8, 8])
  16557. tensor([[0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385],
  16558. [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332],
  16559. [0.6198, 0.4115, 0.7762, 0.2717, 0.3713, 0.3200, 0.5837, 0.5683],
  16560. [0.6248, 0.4185, 0.8500, 0.5767, 0.4463, 0.4550, 0.5613, 0.5917],
  16561. [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466],
  16562. [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329],
  16563. [ nan, nan, 0.7268, 0.2333, 0.4125, 0.1933, 0.5112, 0.5383],
  16564. [0.6099, 0.4030, 0.8638, 0.5117, 0.4983, 0.4965, 0.5086, 0.5388]],
  16565. device='cuda:0', dtype=torch.float64)
  16566. predictions are: tensor([[ 0.6903, -0.3480, 1.6033, -1.1603, -0.2271, -1.2476, 0.8479, 0.1206],
  16567. [ 0.9546, -0.2341, 1.9506, -0.1557, -0.5827, -0.3543, 0.8376, 0.1278],
  16568. [ 0.7325, -0.3103, 1.4645, -1.0865, -0.5483, -0.8621, 0.4646, 0.2128],
  16569. [ 0.9658, -0.1793, 1.7586, 0.3576, -0.2545, -0.2151, 0.4456, 0.2952],
  16570. [ 0.6139, -0.3888, 1.7524, 0.0334, -0.2684, 0.1579, 0.3302, 0.1551],
  16571. [ 0.9975, -0.1532, 1.7208, -0.8041, -0.6586, -0.6190, 0.5044, 0.1213],
  16572. [-1.4734, -1.7306, 1.0105, -1.3044, -0.3757, -1.4271, 0.1577, 0.2327],
  16573. [ 0.5141, -0.4748, 1.8513, -0.0350, -0.0525, -0.0404, 0.2714, 0.1178]],
  16574. device='cuda:0', grad_fn=<AddmmBackward>)
  16575. landmarks are: tensor([[[ 0.6174, -0.3936, 1.4586, -1.1709, -0.2420, -1.1389, 0.8296,
  16576. 0.2012],
  16577. [ 0.6520, -0.4056, 1.9173, -0.0765, -0.5596, -0.4537, 0.7949,
  16578. 0.1768],
  16579. [ 0.5768, -0.3852, 1.2995, -1.0311, -0.5711, -0.8079, 0.4104,
  16580. 0.3392],
  16581. [ 0.5999, -0.3530, 1.6402, 0.3777, -0.2247, -0.1843, 0.3065,
  16582. 0.4470],
  16583. [ 0.5178, -0.4480, 1.6748, 0.0620, -0.2767, 0.2083, 0.1067,
  16584. 0.2386],
  16585. [ 0.5680, -0.4393, 1.5920, -0.6672, -0.6453, -0.5457, 0.5149,
  16586. 0.1753],
  16587. [-2.2859, -2.2859, 1.0712, -1.2085, -0.3806, -1.3929, 0.0755,
  16588. 0.2006],
  16589. [ 0.5309, -0.4246, 1.7037, 0.0774, 0.0158, 0.0075, 0.0635,
  16590. 0.2026]]], device='cuda:0')
  16591. loss_train_step before backward: tensor(0.0308, device='cuda:0', grad_fn=<MseLossBackward>)
  16592. loss_train_step after backward: tensor(0.0308, device='cuda:0', grad_fn=<MseLossBackward>)
  16593. loss_train: 1.0505003025755286
  16594. step: 26
  16595. running loss: 0.04040385779136649
  16596. Train Steps: 26/90 Loss: 0.0404 torch.Size([8, 600, 800])
  16597. torch.Size([8, 8])
  16598. tensor([[0.6200, 0.4112, 0.8862, 0.4100, 0.3638, 0.4917, 0.6088, 0.6050],
  16599. [0.6273, 0.4105, 0.8988, 0.4517, 0.3912, 0.2550, 0.5894, 0.4811],
  16600. [0.6208, 0.4082, 0.8538, 0.3067, 0.3588, 0.3717, 0.6112, 0.5517],
  16601. [0.6224, 0.4097, 0.7438, 0.2267, 0.3850, 0.2850, 0.5988, 0.5250],
  16602. [0.6182, 0.3998, 0.8793, 0.4191, 0.3552, 0.4285, 0.6038, 0.5312],
  16603. [0.6286, 0.4086, 0.8408, 0.2801, 0.4163, 0.2800, 0.6725, 0.5393],
  16604. [0.6277, 0.4103, 0.8087, 0.5717, 0.4188, 0.4750, 0.5663, 0.6083],
  16605. [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852]],
  16606. device='cuda:0', dtype=torch.float64)
  16607. predictions are: tensor([[ 0.3545, -0.5765, 1.8130, -0.4768, -0.5115, -0.0557, 0.6404, 0.3101],
  16608. [ 0.8690, -0.2657, 1.8614, -0.3410, -0.3643, -1.1425, 0.4606, -0.0299],
  16609. [ 0.5647, -0.4242, 1.6549, -0.8860, -0.5752, -0.6408, 0.4819, 0.1474],
  16610. [ 0.5902, -0.3918, 1.2179, -1.2495, -0.3850, -1.0606, 0.3993, 0.0960],
  16611. [ 0.7421, -0.3249, 1.7175, -0.5679, -0.5654, -0.3985, 0.4766, 0.1935],
  16612. [ 0.3850, -0.5400, 1.6495, -1.0489, -0.2069, -1.1659, 0.7564, 0.1570],
  16613. [ 0.6308, -0.3960, 1.5943, 0.2649, -0.3435, -0.1843, 0.4090, 0.3099],
  16614. [ 0.4256, -0.5386, 1.7622, -0.0179, -0.1365, -0.0189, 0.2498, 0.2253]],
  16615. device='cuda:0', grad_fn=<AddmmBackward>)
  16616. landmarks are: tensor([[[ 0.5777, -0.3868, 1.8076, -0.3921, -0.6058, -0.0149, 0.5259,
  16617. 0.5085],
  16618. [ 0.6116, -0.3898, 1.8654, -0.1997, -0.4788, -1.1081, 0.4367,
  16619. -0.0637],
  16620. [ 0.5814, -0.4003, 1.6575, -0.8694, -0.6289, -0.5692, 0.5374,
  16621. 0.2622],
  16622. [ 0.5891, -0.3937, 1.1494, -1.2390, -0.5076, -0.9695, 0.4797,
  16623. 0.1390],
  16624. [ 0.5697, -0.4393, 1.7754, -0.3503, -0.6453, -0.3067, 0.5028,
  16625. 0.1677],
  16626. [ 0.6174, -0.3984, 1.5975, -0.9921, -0.3633, -0.9926, 0.8205,
  16627. 0.2050],
  16628. [ 0.6134, -0.3910, 1.4497, 0.3546, -0.3517, -0.0919, 0.3296,
  16629. 0.5239],
  16630. [ 0.5500, -0.4060, 1.7326, 0.0236, -0.1554, -0.0226, 0.1438,
  16631. 0.4171]]], device='cuda:0')
  16632. loss_train_step before backward: tensor(0.0119, device='cuda:0', grad_fn=<MseLossBackward>)
  16633.  
  16634. loss_train_step after backward: tensor(0.0119, device='cuda:0', grad_fn=<MseLossBackward>)
  16635. loss_train: 1.0623777024447918
  16636. step: 27
  16637. running loss: 0.039347322312770064
  16638. Train Steps: 27/90 Loss: 0.0393 torch.Size([8, 600, 800])
  16639. torch.Size([8, 8])
  16640. tensor([[0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200],
  16641. [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083],
  16642. [0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750],
  16643. [0.6251, 0.4163, 0.8662, 0.4467, 0.3625, 0.3567, 0.6038, 0.5533],
  16644. [0.6177, 0.4086, 0.8738, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700],
  16645. [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617],
  16646. [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600],
  16647. [0.6277, 0.4057, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823]],
  16648. device='cuda:0', dtype=torch.float64)
  16649. predictions are: tensor([[ 1.0085, -0.1871, 1.8592, -0.1330, -0.5335, -0.2461, 0.5151, 0.1121],
  16650. [ 0.3415, -0.6384, 1.7394, -0.0560, -0.5300, -0.2992, 0.4617, 0.3580],
  16651. [ 0.3285, -0.5683, 1.3489, -1.0629, -0.3567, -1.1883, 0.1364, 0.1996],
  16652. [ 0.9330, -0.2222, 1.7962, -0.3420, -0.4949, -0.8077, 0.4360, 0.1853],
  16653. [ 0.1978, -0.6507, 1.7339, -0.5420, -0.4382, 0.2163, 0.6497, 0.2670],
  16654. [ 0.5605, -0.4510, 1.8137, -0.1079, -0.2602, 0.0344, 0.3814, 0.2185],
  16655. [ 0.6950, -0.3404, 1.2603, -1.1607, -0.5204, -0.9337, 0.4292, 0.1559],
  16656. [ 0.6403, -0.3981, 1.6608, -1.2186, -0.1359, -1.5327, 0.6665, -0.0350]],
  16657. device='cuda:0', grad_fn=<AddmmBackward>)
  16658. landmarks are: tensor([[[ 0.5865, -0.3973, 1.8423, -0.0688, -0.5192, -0.2305, 0.4162,
  16659. 0.1159],
  16660. [ 0.5713, -0.4105, 1.7557, 0.0467, -0.6520, -0.2690, 0.3988,
  16661. 0.5239],
  16662. [ 0.5695, -0.3927, 1.3226, -0.9002, -0.4672, -1.1928, 0.1737,
  16663. 0.3700],
  16664. [ 0.6016, -0.3633, 1.7152, -0.2228, -0.6115, -0.6385, 0.5028,
  16665. 0.2699],
  16666. [ 0.5671, -0.3988, 1.7499, -0.4614, -0.5423, 0.3007, 0.5894,
  16667. 0.3469],
  16668. [ 0.5637, -0.4143, 1.7519, -0.0787, -0.3055, -0.0149, 0.3758,
  16669. 0.3084],
  16670. [ 0.5466, -0.4080, 1.0668, -1.1764, -0.6289, -0.7617, 0.4855,
  16671. 0.3007],
  16672. [ 0.6132, -0.4122, 1.5478, -1.0619, -0.2709, -1.4314, 0.5500,
  16673. -0.0583]]], device='cuda:0')
  16674. loss_train_step before backward: tensor(0.0192, device='cuda:0', grad_fn=<MseLossBackward>)
  16675. loss_train_step after backward: tensor(0.0192, device='cuda:0', grad_fn=<MseLossBackward>)
  16676. loss_train: 1.081544168293476
  16677. step: 28
  16678. running loss: 0.03862657743905272
  16679. Train Steps: 28/90 Loss: 0.0386 torch.Size([8, 600, 800])
  16680. torch.Size([8, 8])
  16681. tensor([[0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166],
  16682. [0.6111, 0.3995, 0.8788, 0.4567, 0.3813, 0.4833, 0.5450, 0.5700],
  16683. [0.6265, 0.4088, 0.8025, 0.1850, 0.4163, 0.2500, 0.6290, 0.4947],
  16684. [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817],
  16685. [0.6273, 0.4105, 0.8988, 0.4517, 0.3912, 0.2550, 0.5894, 0.4811],
  16686. [0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166],
  16687. [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483],
  16688. [0.6136, 0.4085, 0.6688, 0.2317, 0.3862, 0.2367, 0.5517, 0.5783]],
  16689. device='cuda:0', dtype=torch.float64)
  16690. predictions are: tensor([[ 7.0282e-01, -3.5619e-01, 1.7135e+00, -4.1727e-02, -3.3237e-01,
  16691. 1.0901e-02, 6.6772e-01, 2.1084e-01],
  16692. [ 3.8480e-01, -5.4974e-01, 1.7183e+00, -4.0325e-01, -5.3389e-01,
  16693. -2.2765e-01, 2.6194e-01, 2.6075e-01],
  16694. [ 7.6077e-01, -3.0297e-01, 1.5522e+00, -1.4613e+00, -3.3689e-01,
  16695. -1.3400e+00, 7.1667e-01, 7.7529e-02],
  16696. [ 7.0844e-01, -3.7311e-01, 1.6807e+00, 2.1612e-01, -2.8488e-01,
  16697. -1.6373e-01, 1.3775e-01, 1.9161e-01],
  16698. [ 9.8953e-01, -1.7994e-01, 1.8835e+00, -4.4454e-01, -4.4358e-01,
  16699. -1.2113e+00, 4.9894e-01, 5.9100e-02],
  16700. [ 2.3141e-01, -6.8316e-01, 1.7980e+00, -3.3082e-01, -5.0640e-01,
  16701. 9.5241e-05, 7.1420e-01, 2.7764e-01],
  16702. [ 4.2223e-01, -5.0048e-01, 1.6832e+00, -6.2590e-01, -5.5156e-01,
  16703. -1.7219e-01, 3.4742e-01, 2.4207e-01],
  16704. [ 7.0103e-01, -3.0725e-01, 9.7545e-01, -1.3389e+00, -4.3456e-01,
  16705. -1.3520e+00, 1.6919e-01, 2.4624e-01]], device='cuda:0',
  16706. grad_fn=<AddmmBackward>)
  16707. landmarks are: tensor([[[ 6.0425e-01, -4.2731e-01, 1.6920e+00, 1.8595e-01, -2.7171e-01,
  16708. 1.4059e-01, 7.9965e-01, 1.0043e-01],
  16709. [ 5.3684e-01, -4.4057e-01, 1.7730e+00, -1.7660e-01, -5.2494e-01,
  16710. -5.3426e-02, 2.3141e-01, 3.4688e-01],
  16711. [ 6.0785e-01, -3.9761e-01, 1.4208e+00, -1.4314e+00, -3.6328e-01,
  16712. -1.1312e+00, 6.1950e-01, -9.2270e-04],
  16713. [ 5.1928e-01, -4.6990e-01, 1.5767e+00, 4.0077e-01, -2.4203e-01,
  16714. 7.7444e-02, 1.1776e-01, -6.1038e-02],
  16715. [ 6.1161e-01, -3.8976e-01, 1.8654e+00, -1.9969e-01, -4.7875e-01,
  16716. -1.1081e+00, 4.3668e-01, -6.3661e-02],
  16717. [ 5.9436e-01, -4.4897e-01, 1.8643e+00, -6.5918e-02, -5.1472e-01,
  16718. 1.2348e-01, 7.6842e-01, 1.0043e-01],
  16719. [ 5.5456e-01, -4.6205e-01, 1.7788e+00, -4.2294e-01, -5.1917e-01,
  16720. -2.2633e-02, 4.2771e-01, 2.4681e-01],
  16721. [ 5.4827e-01, -3.9908e-01, 8.0300e-01, -1.2159e+00, -5.0185e-01,
  16722. -1.1928e+00, 2.6243e-01, 3.8516e-01]]], device='cuda:0')
  16723. loss_train_step before backward: tensor(0.0218, device='cuda:0', grad_fn=<MseLossBackward>)
  16724. loss_train_step after backward: tensor(0.0218, device='cuda:0', grad_fn=<MseLossBackward>)
  16725. loss_train: 1.1033018305897713
  16726. step: 29
  16727. running loss: 0.03804489070999211
  16728. Train Steps: 29/90 Loss: 0.0380 torch.Size([8, 600, 800])
  16729. torch.Size([8, 8])
  16730. tensor([[0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947],
  16731. [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170],
  16732. [ nan, nan, 0.7268, 0.2333, 0.4125, 0.1933, 0.5112, 0.5383],
  16733. [0.6204, 0.4049, 0.7975, 0.2700, 0.3937, 0.2567, 0.5700, 0.5183],
  16734. [0.6198, 0.3997, 0.8582, 0.5361, 0.4117, 0.5016, 0.5942, 0.5134],
  16735. [0.6157, 0.4102, 0.8513, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890],
  16736. [0.6177, 0.4086, 0.8738, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700],
  16737. [0.6339, 0.4102, 0.8588, 0.3133, 0.4425, 0.2117, 0.6417, 0.5089]],
  16738. device='cuda:0', dtype=torch.float64)
  16739. predictions are: tensor([[ 0.9041, -0.2328, 1.4413, -1.2928, -0.3322, -1.1886, 0.6267, 0.0711],
  16740. [ 0.7592, -0.2912, 1.6034, -0.9775, -0.1320, -1.3031, 0.6796, 0.1741],
  16741. [-1.2913, -1.6136, 0.9918, -1.1419, -0.4308, -1.3234, 0.0654, 0.2949],
  16742. [ 0.8360, -0.2591, 1.3810, -0.9939, -0.5174, -1.0080, 0.2762, 0.1503],
  16743. [ 0.7807, -0.3532, 1.7547, 0.2277, -0.4817, 0.1521, 0.4084, 0.1783],
  16744. [ 0.7861, -0.2807, 1.6459, -0.4515, -0.6758, -0.4393, 0.1187, 0.3166],
  16745. [ 0.4825, -0.4640, 1.7166, -0.4412, -0.5613, 0.3670, 0.6141, 0.3111],
  16746. [ 0.8659, -0.2588, 1.6902, -0.7670, -0.2796, -1.1250, 0.6830, 0.1375]],
  16747. device='cuda:0', grad_fn=<AddmmBackward>)
  16748. landmarks are: tensor([[[ 6.0739e-01, -4.0631e-01, 1.3630e+00, -1.5238e+00, -3.1132e-01,
  16749. -1.2390e+00, 6.1950e-01, -9.2270e-04],
  16750. [ 6.4707e-01, -3.8397e-01, 1.5767e+00, -1.0311e+00, -4.5727e-02,
  16751. -1.5007e+00, 6.8892e-01, 1.0199e-01],
  16752. [-2.2859e+00, -2.2859e+00, 1.0712e+00, -1.2085e+00, -3.8060e-01,
  16753. -1.3929e+00, 7.5520e-02, 2.0062e-01],
  16754. [ 5.7985e-01, -4.1555e-01, 1.3977e+00, -1.0388e+00, -4.6721e-01,
  16755. -1.1004e+00, 3.4688e-01, 1.0824e-01],
  16756. [ 5.7673e-01, -4.3957e-01, 1.6782e+00, 1.9046e-01, -3.8437e-01,
  16757. 3.0841e-02, 4.5876e-01, 8.5521e-02],
  16758. [ 5.5813e-01, -3.9120e-01, 1.6460e+00, -5.2302e-01, -6.1732e-01,
  16759. -5.9230e-01, 6.8107e-02, 4.3475e-01],
  16760. [ 5.6715e-01, -3.9885e-01, 1.7499e+00, -4.6143e-01, -5.4226e-01,
  16761. 3.0069e-01, 5.8938e-01, 3.4688e-01],
  16762. [ 6.4212e-01, -3.9120e-01, 1.6806e+00, -8.3865e-01, -2.4203e-01,
  16763. -1.3082e+00, 6.7795e-01, 6.4585e-02]]], device='cuda:0')
  16764. loss_train_step before backward: tensor(0.0340, device='cuda:0', grad_fn=<MseLossBackward>)
  16765. loss_train_step after backward: tensor(0.0340, device='cuda:0', grad_fn=<MseLossBackward>)
  16766. loss_train: 1.1373316459357738
  16767. step: 30
  16768. running loss: 0.0379110548645258
  16769.  
  16770. Train Steps: 30/90 Loss: 0.0379 torch.Size([8, 600, 800])
  16771. torch.Size([8, 8])
  16772. tensor([[0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656],
  16773. [0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783],
  16774. [ nan, nan, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621],
  16775. [ nan, nan, 0.6469, 0.1943, 0.4025, 0.2000, 0.5125, 0.5533],
  16776. [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633],
  16777. [0.6193, 0.4034, 0.7757, 0.2347, 0.3733, 0.2919, 0.5930, 0.4926],
  16778. [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679],
  16779. [0.6145, 0.4008, 0.8750, 0.5383, 0.3975, 0.4650, 0.5563, 0.5533]],
  16780. device='cuda:0', dtype=torch.float64)
  16781. predictions are: tensor([[ 0.9908, -0.1592, 1.7502, -0.3743, -0.6669, -0.1327, 0.2648, 0.2209],
  16782. [-0.1769, -0.8809, 1.1025, -1.1944, -0.3271, -1.3661, 0.1683, 0.3038],
  16783. [-0.9808, -1.4234, 1.6078, -1.3676, 0.0597, -1.3673, 0.7059, 0.2835],
  16784. [-1.0429, -1.4503, 0.9365, -1.3842, -0.4058, -1.4962, 0.0476, 0.3043],
  16785. [ 1.3095, 0.0057, 1.8528, 0.1808, -0.6550, -0.1029, 0.3944, 0.2550],
  16786. [ 1.3513, 0.0706, 1.4051, -1.2822, -0.5891, -1.0075, 0.4432, -0.0362],
  16787. [ 1.3579, 0.0215, 1.8816, 0.1004, -0.4905, 0.3744, 0.8456, 0.2072],
  16788. [ 1.0344, -0.1677, 1.8004, 0.2043, -0.4927, -0.0642, 0.3500, 0.2197]],
  16789. device='cuda:0', grad_fn=<AddmmBackward>)
  16790. landmarks are: tensor([[[ 0.5283, -0.4429, 1.5940, -0.2844, -0.5827, -0.1458, 0.2823,
  16791. 0.3267],
  16792. [ 0.5742, -0.4474, 0.9834, -1.0159, -0.3229, -1.3159, 0.2314,
  16793. 0.3854],
  16794. [-2.2859, -2.2859, 1.7210, -0.9772, 0.1852, -1.3698, 0.7986,
  16795. 0.3104],
  16796. [-2.2859, -2.2859, 0.7023, -1.3883, -0.4268, -1.3621, 0.0813,
  16797. 0.2699],
  16798. [ 0.5879, -0.3591, 1.8018, 0.1236, -0.5596, -0.1612, 0.3469,
  16799. 0.3161],
  16800. [ 0.5746, -0.4225, 1.2968, -1.2019, -0.5615, -0.9374, 0.4533,
  16801. -0.0108],
  16802. [ 0.6257, -0.4249, 1.5998, 0.1236, -0.3806, 0.3084, 0.9887,
  16803. 0.3371],
  16804. [ 0.5526, -0.4347, 1.7557, 0.2006, -0.4499, -0.1381, 0.2834,
  16805. 0.2699]]], device='cuda:0')
  16806. loss_train_step before backward: tensor(0.1383, device='cuda:0', grad_fn=<MseLossBackward>)
  16807. loss_train_step after backward: tensor(0.1383, device='cuda:0', grad_fn=<MseLossBackward>)
  16808. loss_train: 1.2756343595683575
  16809. step: 31
  16810. running loss: 0.04114949546994701
  16811. Train Steps: 31/90 Loss: 0.0411 torch.Size([8, 600, 800])
  16812. torch.Size([8, 8])
  16813. tensor([[0.6202, 0.4079, 0.8025, 0.2500, 0.3763, 0.3217, 0.6125, 0.5533],
  16814. [0.6117, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103],
  16815. [0.6357, 0.4097, 0.9038, 0.3883, 0.4213, 0.2950, 0.6686, 0.5390],
  16816. [0.6277, 0.4103, 0.8087, 0.5717, 0.4188, 0.4750, 0.5663, 0.6083],
  16817. [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167],
  16818. [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167],
  16819. [0.6254, 0.3993, 0.8988, 0.4767, 0.3987, 0.5517, 0.6955, 0.5285],
  16820. [0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166]],
  16821. device='cuda:0', dtype=torch.float64)
  16822. predictions are: tensor([[ 0.7756, -0.2943, 1.5082, -1.1577, -0.5133, -0.9230, 0.4684, 0.1807],
  16823. [ 0.3804, -0.5261, 0.9433, -1.4339, -0.4707, -1.3062, 0.1094, 0.1210],
  16824. [ 0.6888, -0.3697, 1.7758, -0.6519, -0.3946, -1.1142, 0.5529, 0.1363],
  16825. [ 0.6840, -0.3646, 1.5596, 0.2019, -0.4157, -0.2509, 0.2369, 0.3986],
  16826. [ 0.8092, -0.3029, 1.7801, -0.0520, -0.4425, -0.0876, 0.3524, 0.1252],
  16827. [ 0.6672, -0.3370, 1.2757, -1.0965, -0.4721, -1.1377, 0.3404, 0.3186],
  16828. [ 0.3779, -0.5995, 1.8432, -0.2514, -0.4791, 0.1413, 0.7156, 0.2372],
  16829. [ 0.3615, -0.6130, 1.7762, -0.2651, -0.5145, 0.0278, 0.5792, 0.2315]],
  16830. device='cuda:0', grad_fn=<AddmmBackward>)
  16831. landmarks are: tensor([[[ 0.5788, -0.4020, 1.4208, -1.1312, -0.5480, -0.8002, 0.5432,
  16832. 0.2699],
  16833. [ 0.5393, -0.4299, 0.7453, -1.3775, -0.5596, -1.1081, 0.1527,
  16834. 0.0712],
  16835. [ 0.6504, -0.3936, 1.8885, -0.4922, -0.3402, -0.9233, 0.8022,
  16836. 0.2035],
  16837. [ 0.6134, -0.3910, 1.4497, 0.3546, -0.3517, -0.0919, 0.3296,
  16838. 0.5239],
  16839. [ 0.5828, -0.4066, 1.7557, 0.0774, -0.3691, -0.0226, 0.4277,
  16840. 0.1005],
  16841. [ 0.5631, -0.4129, 1.2129, -0.9233, -0.4152, -1.0311, 0.4566,
  16842. 0.5624],
  16843. [ 0.6026, -0.4417, 1.8654, -0.0842, -0.4441, 0.2622, 0.9265,
  16844. 0.1554],
  16845. [ 0.5944, -0.4490, 1.8643, -0.0659, -0.5147, 0.1235, 0.7684,
  16846. 0.1004]]], device='cuda:0')
  16847. loss_train_step before backward: tensor(0.0165, device='cuda:0', grad_fn=<MseLossBackward>)
  16848. loss_train_step after backward: tensor(0.0165, device='cuda:0', grad_fn=<MseLossBackward>)
  16849. loss_train: 1.2920843735337257
  16850. step: 32
  16851. running loss: 0.04037763667292893
  16852. Train Steps: 32/90 Loss: 0.0404 torch.Size([8, 600, 800])
  16853. torch.Size([8, 8])
  16854. tensor([[0.6137, 0.4038, 0.8563, 0.4050, 0.3813, 0.2550, 0.5106, 0.4954],
  16855. [0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947],
  16856. [0.6177, 0.4086, 0.8738, 0.3950, 0.3775, 0.5600, 0.6225, 0.5700],
  16857. [0.6043, 0.4022, 0.6887, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136],
  16858. [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366],
  16859. [ nan, nan, 0.6412, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700],
  16860. [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600],
  16861. [0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938]],
  16862. device='cuda:0', dtype=torch.float64)
  16863. predictions are: tensor([[ 0.7272, -0.3183, 1.7046, -0.5656, -0.4559, -1.0896, 0.1794, 0.1725],
  16864. [ 0.4934, -0.5156, 1.8159, -0.1265, -0.5439, -0.0894, 0.4923, 0.2124],
  16865. [ 0.6612, -0.3344, 1.7433, -0.5302, -0.4852, 0.2510, 0.5717, 0.3081],
  16866. [ 0.7566, -0.2804, 1.0319, -1.3667, -0.4885, -1.1770, 0.3090, 0.1345],
  16867. [ 0.6595, -0.3936, 1.5598, 0.0052, -0.4803, -0.1097, 0.5725, 0.2791],
  16868. [-0.9094, -1.3789, 1.0018, -1.3468, -0.2830, -1.5701, 0.1808, 0.3358],
  16869. [ 0.7135, -0.3099, 1.2355, -1.1843, -0.5799, -0.8957, 0.4245, 0.2362],
  16870. [ 0.8254, -0.3237, 1.8059, 0.2547, -0.4740, -0.2230, 0.5220, 0.1259]],
  16871. device='cuda:0', grad_fn=<AddmmBackward>)
  16872. landmarks are: tensor([[[ 5.4850e-01, -4.2094e-01, 1.6691e+00, -4.1524e-01, -5.2494e-01,
  16873. -1.1081e+00, 7.2521e-02, 2.0831e-03],
  16874. [ 5.8031e-01, -4.3672e-01, 1.8423e+00, -3.0331e-02, -5.9423e-01,
  16875. 2.3557e-02, 6.5034e-01, -9.2270e-04],
  16876. [ 5.6715e-01, -3.9885e-01, 1.7499e+00, -4.6143e-01, -5.4226e-01,
  16877. 3.0069e-01, 5.8938e-01, 3.4688e-01],
  16878. [ 5.0531e-01, -4.2810e-01, 8.9538e-01, -1.3698e+00, -5.4226e-01,
  16879. -1.1389e+00, 2.4525e-01, 8.6245e-02],
  16880. [ 6.2895e-01, -4.3453e-01, 1.3794e+00, 3.6792e-01, -4.8453e-01,
  16881. 3.8953e-02, 9.2654e-01, 1.9283e-01],
  16882. [-2.2859e+00, -2.2859e+00, 6.7598e-01, -1.4083e+00, -3.2864e-01,
  16883. -1.4160e+00, 2.4873e-01, 3.4688e-01],
  16884. [ 5.4660e-01, -4.0805e-01, 1.0668e+00, -1.1764e+00, -6.2887e-01,
  16885. -7.6166e-01, 4.8545e-01, 3.0069e-01],
  16886. [ 6.3949e-01, -4.2132e-01, 1.7037e+00, 3.6228e-01, -4.3256e-01,
  16887. -1.0731e-01, 6.5602e-01, -4.8817e-03]]], device='cuda:0')
  16888. loss_train_step before backward: tensor(0.0586, device='cuda:0', grad_fn=<MseLossBackward>)
  16889. loss_train_step after backward: tensor(0.0586, device='cuda:0', grad_fn=<MseLossBackward>)
  16890. loss_train: 1.3506991527974606
  16891. step: 33
  16892. running loss: 0.0409302773574988
  16893. Train Steps: 33/90 Loss: 0.0409 torch.Size([8, 600, 800])
  16894. torch.Size([8, 8])
  16895. tensor([[0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058],
  16896. [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869],
  16897. [0.6179, 0.4040, 0.7412, 0.1850, 0.3825, 0.2783, 0.5837, 0.5600],
  16898. [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426],
  16899. [0.6190, 0.4135, 0.8000, 0.4883, 0.3566, 0.3647, 0.5613, 0.5900],
  16900. [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617],
  16901. [0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033],
  16902. [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364]],
  16903. device='cuda:0', dtype=torch.float64)
  16904. predictions are: tensor([[ 0.3825, -0.5192, 1.5531, -0.5483, -0.5774, -0.2306, 0.2690, 0.1452],
  16905. [ 0.3526, -0.5745, 1.5906, 0.1417, -0.4948, -0.4045, 0.2288, 0.1579],
  16906. [ 0.5171, -0.4502, 1.0703, -1.4119, -0.5305, -1.0881, 0.4877, 0.2627],
  16907. [ 0.4865, -0.4936, 1.6517, -0.4755, -0.5666, -0.7995, 0.4503, 0.1849],
  16908. [ 0.8655, -0.1874, 1.4872, -0.2307, -0.6180, -0.7052, 0.3166, 0.3255],
  16909. [ 0.4165, -0.5100, 1.6322, -0.2517, -0.3476, -0.0160, 0.4465, 0.2920],
  16910. [ 0.4606, -0.5201, 1.6801, -0.0939, -0.3939, 0.0481, 0.4526, 0.2095],
  16911. [ 0.4394, -0.4942, 1.3631, -1.2998, -0.2485, -1.3518, 0.7349, 0.1318]],
  16912. device='cuda:0', grad_fn=<AddmmBackward>)
  16913. landmarks are: tensor([[[ 0.5245, -0.4361, 1.5940, -0.2921, -0.5480, -0.0919, 0.2432,
  16914. 0.0502],
  16915. [ 0.5320, -0.4488, 1.6633, 0.3315, -0.5018, -0.2459, 0.0828,
  16916. -0.0370],
  16917. [ 0.5680, -0.4201, 1.1379, -1.4314, -0.5192, -1.0003, 0.4104,
  16918. 0.3007],
  16919. [ 0.5788, -0.4135, 1.8214, -0.2408, -0.6039, -0.7854, 0.4115,
  16920. 0.2203],
  16921. [ 0.5732, -0.3761, 1.4092, -0.0303, -0.6388, -0.6012, 0.3065,
  16922. 0.4393],
  16923. [ 0.5637, -0.4143, 1.7519, -0.0787, -0.3055, -0.0149, 0.3758,
  16924. 0.3084],
  16925. [ 0.5668, -0.4306, 1.7730, 0.0697, -0.4037, 0.1390, 0.4450,
  16926. 0.0390],
  16927. [ 0.6158, -0.3960, 1.4092, -1.2774, -0.2074, -1.1851, 0.8491,
  16928. 0.1917]]], device='cuda:0')
  16929. loss_train_step before backward: tensor(0.0138, device='cuda:0', grad_fn=<MseLossBackward>)
  16930.  
  16931. loss_train_step after backward: tensor(0.0138, device='cuda:0', grad_fn=<MseLossBackward>)
  16932. loss_train: 1.3644524635747075
  16933. step: 34
  16934. running loss: 0.04013095481102081
  16935. Train Steps: 34/90 Loss: 0.0401 torch.Size([8, 600, 800])
  16936. torch.Size([8, 8])
  16937. tensor([[0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333],
  16938. [0.6109, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117],
  16939. [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285],
  16940. [0.6110, 0.4047, 0.8700, 0.4483, 0.3713, 0.3967, 0.5088, 0.5517],
  16941. [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377],
  16942. [0.6145, 0.4007, 0.8775, 0.4533, 0.4562, 0.5533, 0.6088, 0.5533],
  16943. [0.6277, 0.4103, 0.8087, 0.5717, 0.4188, 0.4750, 0.5663, 0.6083],
  16944. [0.6115, 0.4081, 0.6725, 0.2433, 0.4088, 0.1933, 0.5167, 0.5544]],
  16945. device='cuda:0', dtype=torch.float64)
  16946. predictions are: tensor([[ 0.5411, -0.4536, 1.7278, -0.3899, -0.5399, -0.1135, 0.6217, 0.1842],
  16947. [ 0.3986, -0.5291, 1.6807, -0.3776, -0.6092, -0.5501, 0.3713, 0.1663],
  16948. [ 0.6661, -0.3267, 1.3198, -0.6602, -0.6227, -0.4715, 0.1822, 0.1441],
  16949. [ 0.3818, -0.5150, 1.5559, -0.4041, -0.6201, -0.5724, 0.1614, 0.2175],
  16950. [ 0.3520, -0.6014, 1.6483, 0.0400, -0.4569, -0.2901, 0.5195, 0.1823],
  16951. [ 0.7070, -0.3190, 1.6861, -0.3544, -0.1871, 0.0213, 0.5974, 0.2302],
  16952. [ 0.5853, -0.4072, 1.4730, 0.1342, -0.3969, -0.3106, 0.5202, 0.4160],
  16953. [ 0.1553, -0.6436, 0.8552, -1.3981, -0.4279, -1.4607, 0.2782, 0.2042]],
  16954. device='cuda:0', grad_fn=<AddmmBackward>)
  16955. landmarks are: tensor([[[ 0.6003, -0.3747, 1.8885, -0.1997, -0.5018, -0.0149, 0.5490,
  16956. 0.1775],
  16957. [ 0.5356, -0.4343, 1.7441, -0.1150, -0.6462, -0.3614, 0.3238,
  16958. 0.0774],
  16959. [ 0.5253, -0.4296, 1.3084, -0.3098, -0.6115, -0.2767, 0.0928,
  16960. 0.1552],
  16961. [ 0.5363, -0.4168, 1.7326, -0.2151, -0.5711, -0.4537, 0.0640,
  16962. 0.2622],
  16963. [ 0.5775, -0.4607, 1.6741, 0.1962, -0.4036, -0.1212, 0.4588,
  16964. 0.1979],
  16965. [ 0.5526, -0.4351, 1.7672, -0.1920, -0.1785, 0.2699, 0.5259,
  16966. 0.2699],
  16967. [ 0.6134, -0.3910, 1.4497, 0.3546, -0.3517, -0.0919, 0.3296,
  16968. 0.5239],
  16969. [ 0.5385, -0.4010, 0.8203, -1.1620, -0.3979, -1.3929, 0.1005,
  16970. 0.2747]]], device='cuda:0')
  16971. loss_train_step before backward: tensor(0.0201, device='cuda:0', grad_fn=<MseLossBackward>)
  16972. loss_train_step after backward: tensor(0.0201, device='cuda:0', grad_fn=<MseLossBackward>)
  16973. loss_train: 1.3845558362081647
  16974. step: 35
  16975. running loss: 0.03955873817737613
  16976. Train Steps: 35/90 Loss: 0.0396 torch.Size([8, 600, 800])
  16977. torch.Size([8, 8])
  16978. tensor([[0.6293, 0.3982, 0.8700, 0.5300, 0.3763, 0.4717, 0.7050, 0.5297],
  16979. [0.6136, 0.4117, 0.8700, 0.5167, 0.4188, 0.5083, 0.5147, 0.5495],
  16980. [0.6280, 0.4101, 0.9050, 0.4533, 0.3775, 0.3217, 0.6338, 0.4915],
  16981. [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092],
  16982. [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332],
  16983. [0.6114, 0.4018, 0.7213, 0.1967, 0.3763, 0.2700, 0.5875, 0.5533],
  16984. [0.6250, 0.4106, 0.8700, 0.3717, 0.3588, 0.4967, 0.6038, 0.5167],
  16985. [0.6212, 0.4159, 0.8675, 0.5783, 0.4088, 0.4317, 0.5613, 0.5917]],
  16986. device='cuda:0', dtype=torch.float64)
  16987. predictions are: tensor([[ 0.2897, -0.6276, 1.5647, -0.0474, -0.4891, -0.2550, 0.5759, 0.1923],
  16988. [ 0.5408, -0.4012, 1.5358, 0.0289, -0.3157, -0.0208, 0.1720, 0.2009],
  16989. [ 0.5524, -0.4431, 1.6840, -0.4082, -0.5403, -0.9415, 0.4238, 0.0553],
  16990. [ 0.4974, -0.4879, 1.6445, -0.0998, -0.4748, -0.0273, 0.3334, 0.1735],
  16991. [ 0.5754, -0.4577, 1.6794, -0.2655, -0.5683, -0.5576, 0.6392, 0.2151],
  16992. [ 0.2067, -0.6470, 0.9056, -1.5566, -0.5469, -1.1697, 0.3692, 0.1897],
  16993. [ 0.6249, -0.3607, 1.5125, -0.7481, -0.5763, -0.1067, 0.5937, 0.2508],
  16994. [ 0.3513, -0.5097, 1.4105, 0.1475, -0.3911, -0.3318, 0.2325, 0.4419]],
  16995. device='cuda:0', grad_fn=<AddmmBackward>)
  16996. landmarks are: tensor([[[ 0.6207, -0.4466, 1.7326, 0.1621, -0.5480, -0.1073, 0.9704,
  16997. 0.1608],
  16998. [ 0.5482, -0.3841, 1.7326, 0.1005, -0.3517, 0.0620, 0.0912,
  16999. 0.2522],
  17000. [ 0.6148, -0.3918, 1.8942, -0.1920, -0.5423, -0.8002, 0.6414,
  17001. -0.0156],
  17002. [ 0.5783, -0.4363, 1.7724, 0.0432, -0.5153, 0.0871, 0.4840,
  17003. 0.0663],
  17004. [ 0.6520, -0.4056, 1.9173, -0.0765, -0.5596, -0.4537, 0.7949,
  17005. 0.1768],
  17006. [ 0.5380, -0.4302, 1.0455, -1.3775, -0.5480, -1.0388, 0.4277,
  17007. 0.2699],
  17008. [ 0.6010, -0.3896, 1.7326, -0.5692, -0.6289, 0.0082, 0.5028,
  17009. 0.1005],
  17010. [ 0.5836, -0.3649, 1.7210, 0.3854, -0.3979, -0.2921, 0.3065,
  17011. 0.4470]]], device='cuda:0')
  17012. loss_train_step before backward: tensor(0.0228, device='cuda:0', grad_fn=<MseLossBackward>)
  17013. loss_train_step after backward: tensor(0.0228, device='cuda:0', grad_fn=<MseLossBackward>)
  17014. loss_train: 1.4073562482371926
  17015. step: 36
  17016. running loss: 0.039093229117699795
  17017. Train Steps: 36/90 Loss: 0.0391 torch.Size([8, 600, 800])
  17018. torch.Size([8, 8])
  17019. tensor([[0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517],
  17020. [0.6179, 0.4008, 0.7505, 0.2678, 0.4368, 0.1891, 0.5831, 0.5263],
  17021. [0.6124, 0.4075, 0.7696, 0.4153, 0.3475, 0.3767, 0.5157, 0.5427],
  17022. [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833],
  17023. [0.6254, 0.3993, 0.8988, 0.4767, 0.3987, 0.5517, 0.6955, 0.5285],
  17024. [ nan, nan, 0.7097, 0.2346, 0.4250, 0.1850, 0.5175, 0.5583],
  17025. [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167],
  17026. [0.6293, 0.3982, 0.8700, 0.5300, 0.3763, 0.4717, 0.7050, 0.5297]],
  17027. device='cuda:0', dtype=torch.float64)
  17028. predictions are: tensor([[ 0.6089, -0.3586, 0.9983, -1.0574, -0.4327, -1.0262, 0.4522, 0.3104],
  17029. [ 0.7212, -0.3187, 1.2293, -1.0060, -0.4093, -1.1530, 0.3823, 0.1102],
  17030. [ 0.9051, -0.1809, 1.4924, -0.3170, -0.7052, -0.3290, 0.1440, 0.1669],
  17031. [ 0.1922, -0.6328, 0.9536, -0.8334, -0.5797, -0.8038, 0.2082, 0.3640],
  17032. [ 0.9224, -0.2444, 2.0366, 0.1023, -0.4859, 0.4914, 0.9817, 0.1094],
  17033. [-2.0891, -2.1434, 1.0060, -1.0435, -0.4197, -1.1671, 0.2153, 0.2539],
  17034. [ 0.3065, -0.5146, 1.1554, -0.6968, -0.1279, -1.0569, 0.2675, 0.3804],
  17035. [ 0.7788, -0.3376, 1.8395, 0.3493, -0.5287, 0.0480, 0.6786, 0.0816]],
  17036. device='cuda:0', grad_fn=<AddmmBackward>)
  17037. landmarks are: tensor([[[ 0.5713, -0.3671, 0.8665, -1.0696, -0.3691, -1.2236, 0.3527,
  17038. 0.2622],
  17039. [ 0.5680, -0.4345, 1.1806, -1.0490, -0.2683, -1.4127, 0.4074,
  17040. 0.1449],
  17041. [ 0.5427, -0.4035, 1.2688, -0.3675, -0.6808, -0.5461, 0.0959,
  17042. 0.2206],
  17043. [ 0.5614, -0.4080, 0.7741, -0.8848, -0.5423, -0.9156, 0.3584,
  17044. 0.4085],
  17045. [ 0.6026, -0.4417, 1.8654, -0.0842, -0.4441, 0.2622, 0.9265,
  17046. 0.1554],
  17047. [-2.2859, -2.2859, 0.9922, -1.2021, -0.3229, -1.4314, 0.1044,
  17048. 0.2930],
  17049. [ 0.6161, -0.3075, 1.1678, -0.6500, 0.0813, -1.4006, 0.2545,
  17050. 0.5624],
  17051. [ 0.6207, -0.4466, 1.7326, 0.1621, -0.5480, -0.1073, 0.9704,
  17052. 0.1608]]], device='cuda:0')
  17053. loss_train_step before backward: tensor(0.0270, device='cuda:0', grad_fn=<MseLossBackward>)
  17054. loss_train_step after backward: tensor(0.0270, device='cuda:0', grad_fn=<MseLossBackward>)
  17055. loss_train: 1.4343694495037198
  17056. step: 37
  17057. running loss: 0.038766741878478915
  17058. Train Steps: 37/90 Loss: 0.0388 torch.Size([8, 600, 800])
  17059. torch.Size([8, 8])
  17060. tensor([[0.6113, 0.4088, 0.6859, 0.2208, 0.4363, 0.1700, 0.5188, 0.5533],
  17061. [0.6161, 0.4055, 0.8675, 0.3867, 0.3713, 0.4033, 0.5195, 0.5162],
  17062. [0.6117, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103],
  17063. [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133],
  17064. [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366],
  17065. [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450],
  17066. [0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250],
  17067. [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5413, 0.5433]],
  17068. device='cuda:0', dtype=torch.float64)
  17069. predictions are: tensor([[-1.0315, -1.4505, 0.8585, -1.1804, -0.3616, -1.3684, 0.2596, 0.2981],
  17070. [ 0.3760, -0.4742, 1.6197, -0.2871, -0.5844, -0.3497, 0.1494, 0.1322],
  17071. [ 0.2731, -0.5802, 0.7607, -1.3135, -0.4940, -1.0363, 0.2531, 0.2214],
  17072. [ 0.6946, -0.3087, 1.6473, 0.2415, -0.1956, 0.1741, 0.4767, 0.1525],
  17073. [ 0.7139, -0.3329, 1.4853, 0.1988, -0.4484, 0.0563, 0.7421, 0.2211],
  17074. [ 0.5019, -0.4580, 1.7091, -0.0458, -0.5463, -0.2297, 0.5223, 0.2264],
  17075. [ 0.7135, -0.3331, 1.8205, -0.2745, -0.5317, -0.6166, 0.6577, 0.1887],
  17076. [ 0.4978, -0.4209, 1.3890, -0.5404, -0.6099, -0.6694, 0.2540, 0.2692]],
  17077. device='cuda:0', grad_fn=<AddmmBackward>)
  17078. landmarks are: tensor([[[ 0.5377, -0.3978, 0.8824, -1.2663, -0.2709, -1.5007, 0.1102,
  17079. 0.2699],
  17080. [ 0.5598, -0.4129, 1.7210, -0.4999, -0.5711, -0.4229, 0.1136,
  17081. 0.0983],
  17082. [ 0.5393, -0.4299, 0.7453, -1.3775, -0.5596, -1.1081, 0.1527,
  17083. 0.0712],
  17084. [ 0.5786, -0.4141, 1.7037, 0.1544, -0.1862, 0.0736, 0.4393,
  17085. 0.0851],
  17086. [ 0.6289, -0.4345, 1.3794, 0.3679, -0.4845, 0.0390, 0.9265,
  17087. 0.1928],
  17088. [ 0.5785, -0.4252, 1.7676, -0.1602, -0.5845, -0.3446, 0.4566,
  17089. 0.2314],
  17090. [ 0.6111, -0.3828, 1.8885, -0.3844, -0.5654, -0.8079, 0.5663,
  17091. 0.1390],
  17092. [ 0.5894, -0.3479, 1.7730, -0.6847, -0.5538, -0.7155, 0.2141,
  17093. 0.2237]]], device='cuda:0')
  17094. loss_train_step before backward: tensor(0.0676, device='cuda:0', grad_fn=<MseLossBackward>)
  17095.  
  17096. loss_train_step after backward: tensor(0.0676, device='cuda:0', grad_fn=<MseLossBackward>)
  17097. loss_train: 1.501950633712113
  17098. step: 38
  17099. running loss: 0.03952501667663455
  17100. Train Steps: 38/90 Loss: 0.0395 torch.Size([8, 600, 800])
  17101. torch.Size([8, 8])
  17102. tensor([[0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482],
  17103. [ nan, nan, 0.8938, 0.2850, 0.4662, 0.3117, 0.7406, 0.5528],
  17104. [0.6159, 0.4085, 0.6900, 0.2283, 0.4088, 0.1950, 0.5123, 0.5397],
  17105. [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600],
  17106. [0.6200, 0.4059, 0.8700, 0.4900, 0.4163, 0.5000, 0.6162, 0.5467],
  17107. [0.6091, 0.3997, 0.8314, 0.4334, 0.3788, 0.4550, 0.5213, 0.5656],
  17108. [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837],
  17109. [0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090]],
  17110. device='cuda:0', dtype=torch.float64)
  17111. predictions are: tensor([[ 0.6832, -0.3513, 1.7042, -0.1977, -0.6291, -0.4616, 0.5176, 0.1659],
  17112. [-0.7401, -1.2939, 1.7845, -0.6514, -0.1847, -0.8358, 1.0248, 0.2050],
  17113. [-0.5645, -1.1050, 0.7261, -1.0714, -0.4450, -1.3088, 0.0884, 0.2421],
  17114. [ 0.6336, -0.3166, 1.6069, 0.0188, -0.1684, 0.2818, 0.4854, 0.2356],
  17115. [ 0.6863, -0.3263, 1.6596, 0.1793, -0.3551, 0.1340, 0.4881, 0.2052],
  17116. [ 0.6414, -0.3296, 1.4893, -0.1288, -0.5712, -0.1099, 0.2392, 0.1740],
  17117. [ 0.4698, -0.4321, 0.7184, -0.8961, -0.6497, -0.8492, 0.1134, 0.3681],
  17118. [ 0.7837, -0.2923, 1.5828, -0.4301, -0.5045, -0.8921, 0.5768, 0.0597]],
  17119. device='cuda:0', grad_fn=<AddmmBackward>)
  17120. landmarks are: tensor([[[ 0.5791, -0.4127, 1.8442, -0.3985, -0.6031, -0.6154, 0.4473,
  17121. 0.2464],
  17122. [-2.2859, -2.2859, 1.8423, -0.9695, -0.1323, -0.8463, 1.1349,
  17123. 0.2676],
  17124. [ 0.5591, -0.3990, 0.9012, -1.2313, -0.3979, -1.3852, 0.0804,
  17125. 0.2071],
  17126. [ 0.5652, -0.4325, 1.7388, -0.1429, -0.0688, 0.3469, 0.4970,
  17127. 0.3007],
  17128. [ 0.5776, -0.4109, 1.7326, -0.0226, -0.3633, 0.0236, 0.5605,
  17129. 0.2391],
  17130. [ 0.5275, -0.4396, 1.5543, -0.2841, -0.5365, -0.1843, 0.1221,
  17131. 0.3265],
  17132. [ 0.5629, -0.3871, 0.7243, -0.9581, -0.5827, -0.9849, 0.1288,
  17133. 0.4103],
  17134. [ 0.6165, -0.4249, 1.7961, -0.5384, -0.4499, -0.9695, 0.6401,
  17135. 0.0652]]], device='cuda:0')
  17136. loss_train_step before backward: tensor(0.0914, device='cuda:0', grad_fn=<MseLossBackward>)
  17137. loss_train_step after backward: tensor(0.0914, device='cuda:0', grad_fn=<MseLossBackward>)
  17138. loss_train: 1.5933644147589803
  17139. step: 39
  17140. running loss: 0.04085549781433283
  17141. Train Steps: 39/90 Loss: 0.0409 torch.Size([8, 600, 800])
  17142. torch.Size([8, 8])
  17143. tensor([[0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411],
  17144. [0.6284, 0.4127, 0.8538, 0.5867, 0.4363, 0.5083, 0.6038, 0.5433],
  17145. [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683],
  17146. [0.6223, 0.4028, 0.8988, 0.4200, 0.3763, 0.5733, 0.6375, 0.5167],
  17147. [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390],
  17148. [0.6179, 0.4040, 0.7412, 0.1850, 0.3825, 0.2783, 0.5837, 0.5600],
  17149. [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167],
  17150. [0.6166, 0.4008, 0.8563, 0.5667, 0.4388, 0.4933, 0.5575, 0.5567]],
  17151. device='cuda:0', dtype=torch.float64)
  17152. predictions are: tensor([[ 0.5238, -0.3950, 1.2829, -0.8842, -0.3780, -1.2504, 0.3666, 0.1162],
  17153. [ 0.2301, -0.5965, 1.5215, 0.3388, -0.3021, -0.0835, 0.4180, 0.1480],
  17154. [ 0.3454, -0.4980, 1.4978, 0.1416, -0.5490, -0.5785, 0.3343, 0.2566],
  17155. [ 0.5529, -0.4195, 1.7595, -0.3212, -0.4820, 0.2027, 0.8572, 0.1335],
  17156. [ 0.3021, -0.5822, 1.6934, -0.0044, -0.4441, -0.0950, 0.3973, 0.1461],
  17157. [ 0.3611, -0.5352, 0.9858, -1.2857, -0.5901, -1.0377, 0.4483, 0.2478],
  17158. [ 0.2816, -0.5578, 1.3447, -0.7849, -0.6097, -0.5858, 0.5139, 0.3761],
  17159. [ 0.3448, -0.5351, 1.4953, 0.3681, -0.2450, -0.1087, 0.2777, 0.2207]],
  17160. device='cuda:0', grad_fn=<AddmmBackward>)
  17161. landmarks are: tensor([[[ 0.5713, -0.4321, 1.4975, -0.8134, -0.3094, -1.3345, 0.3779,
  17162. 0.2134],
  17163. [ 0.6166, -0.3795, 1.6575, 0.4239, -0.2709, 0.0620, 0.5028,
  17164. 0.2237],
  17165. [ 0.5777, -0.3915, 1.7961, 0.1698, -0.5192, -0.5307, 0.2141,
  17166. 0.3392],
  17167. [ 0.5886, -0.4253, 1.8654, -0.3460, -0.5480, 0.3623, 0.6587,
  17168. 0.1005],
  17169. [ 0.5250, -0.4661, 1.7383, -0.0765, -0.4268, -0.0226, 0.2535,
  17170. 0.2035],
  17171. [ 0.5680, -0.4201, 1.1379, -1.4314, -0.5192, -1.0003, 0.4104,
  17172. 0.3007],
  17173. [ 0.5768, -0.3899, 1.3861, -0.7771, -0.5885, -0.5461, 0.5028,
  17174. 0.5624],
  17175. [ 0.5620, -0.4346, 1.6691, 0.3315, -0.2594, -0.0072, 0.2891,
  17176. 0.2853]]], device='cuda:0')
  17177. loss_train_step before backward: tensor(0.0166, device='cuda:0', grad_fn=<MseLossBackward>)
  17178. loss_train_step after backward: tensor(0.0166, device='cuda:0', grad_fn=<MseLossBackward>)
  17179. loss_train: 1.610002956353128
  17180. step: 40
  17181. running loss: 0.0402500739088282
  17182. Train Steps: 40/90 Loss: 0.0403 torch.Size([8, 600, 800])
  17183. torch.Size([8, 8])
  17184. tensor([[0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334],
  17185. [0.6132, 0.4118, 0.8200, 0.3633, 0.3563, 0.5400, 0.5787, 0.5136],
  17186. [0.6053, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168],
  17187. [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207],
  17188. [0.6176, 0.4030, 0.8850, 0.4850, 0.3688, 0.4050, 0.5312, 0.5783],
  17189. [0.6229, 0.4066, 0.7612, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350],
  17190. [0.6274, 0.4003, 0.8638, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661],
  17191. [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103]],
  17192. device='cuda:0', dtype=torch.float64)
  17193. predictions are: tensor([[ 0.5868, -0.4193, 1.2456, -0.9690, -0.5729, -0.7326, 0.5721, 0.2117],
  17194. [ 0.3966, -0.4664, 1.5337, -0.4057, -0.4873, 0.2182, 0.5452, 0.3014],
  17195. [-0.0431, -0.8257, 0.8331, -1.1309, -0.4279, -1.1973, 0.2385, 0.2226],
  17196. [ 0.3674, -0.5902, 1.4668, -0.8268, -0.2956, -1.1755, 0.6494, 0.1242],
  17197. [ 0.3705, -0.5327, 1.7229, 0.2829, -0.4762, -0.2262, 0.2493, 0.2158],
  17198. [ 0.4006, -0.4981, 1.2489, -0.7730, -0.3574, -1.0831, 0.3577, 0.2296],
  17199. [ 0.3442, -0.6000, 1.7299, 0.7331, -0.4478, 0.0127, 0.5833, 0.1481],
  17200. [ 0.3361, -0.5357, 1.5195, -0.5111, -0.5457, -0.3380, 0.4007, 0.2688]],
  17201. device='cuda:0', grad_fn=<AddmmBackward>)
  17202. landmarks are: tensor([[[ 0.5730, -0.4562, 1.2195, -1.2440, -0.5497, -0.7711, 0.5704,
  17203. 0.1779],
  17204. [ 0.5466, -0.3840, 1.5016, -0.6077, -0.6404, 0.2083, 0.3871,
  17205. 0.0862],
  17206. [ 0.5102, -0.4223, 0.8999, -1.4011, -0.4383, -1.3082, 0.2267,
  17207. 0.1013],
  17208. [ 0.6132, -0.4248, 1.5247, -1.1620, -0.2882, -1.3159, 0.6545,
  17209. 0.1193],
  17210. [ 0.5667, -0.4246, 1.8018, -0.0457, -0.5827, -0.4152, 0.1679,
  17211. 0.3854],
  17212. [ 0.5911, -0.4080, 1.2303, -0.9156, -0.3229, -1.2851, 0.4508,
  17213. 0.1852],
  17214. [ 0.6120, -0.4371, 1.7037, 0.4701, -0.5827, -0.0226, 0.5354,
  17215. -0.1331],
  17216. [ 0.5886, -0.3784, 1.4554, -0.9079, -0.6577, -0.4845, 0.3440,
  17217. 0.0712]]], device='cuda:0')
  17218. loss_train_step before backward: tensor(0.0313, device='cuda:0', grad_fn=<MseLossBackward>)
  17219. loss_train_step after backward: tensor(0.0313, device='cuda:0', grad_fn=<MseLossBackward>)
  17220. loss_train: 1.6413122108206153
  17221. step: 41
  17222. running loss: 0.040032005141966226
  17223. Train Steps: 41/90 Loss: 0.0400 torch.Size([8, 600, 800])
  17224. torch.Size([8, 8])
  17225. tensor([[0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633],
  17226. [0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767],
  17227. [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617],
  17228. [0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567],
  17229. [0.6263, 0.4039, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804],
  17230. [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413],
  17231. [ nan, nan, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609],
  17232. [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758]],
  17233. device='cuda:0', dtype=torch.float64)
  17234. predictions are: tensor([[ 0.7870, -0.2595, 1.2943, -1.1052, -0.3462, -1.0785, 0.7152, 0.2656],
  17235. [ 0.4789, -0.4572, 1.4920, 0.2105, -0.5343, -0.1205, 0.1726, 0.2225],
  17236. [ 0.6183, -0.3261, 1.4758, 0.0889, -0.4249, 0.0699, 0.1858, 0.2688],
  17237. [ 0.7325, -0.2322, 1.4474, -0.0852, -0.2401, 0.0818, 0.2022, 0.2692],
  17238. [ 0.6747, -0.3397, 1.6252, -0.1277, -0.6790, -0.1723, 0.4741, 0.0560],
  17239. [ 0.2175, -0.6533, 1.6239, -0.7552, -0.1979, -1.1170, 0.7941, 0.1465],
  17240. [-1.5191, -1.7626, 1.4842, -0.8116, -0.1347, -1.1538, 0.5951, 0.3030],
  17241. [ 0.6647, -0.3226, 1.5790, 0.1086, -0.5153, 0.0848, 0.4939, 0.1292]],
  17242. device='cuda:0', grad_fn=<AddmmBackward>)
  17243. landmarks are: tensor([[[ 0.6042, -0.4105, 1.5478, -1.2082, -0.1208, -1.0927, 0.9704,
  17244. 0.3157],
  17245. [ 0.5610, -0.4381, 1.7730, 0.0390, -0.4326, -0.1458, 0.1794,
  17246. 0.3777],
  17247. [ 0.5637, -0.4143, 1.7519, -0.0787, -0.3055, -0.0149, 0.3758,
  17248. 0.3084],
  17249. [ 0.5538, -0.4139, 1.7557, -0.1843, -0.0459, 0.1242, 0.4219,
  17250. 0.2853],
  17251. [ 0.6072, -0.4206, 1.8711, -0.2536, -0.6115, -0.1304, 0.6812,
  17252. -0.0670],
  17253. [ 0.6108, -0.4201, 1.8711, -0.7848, -0.0053, -1.2236, 1.0362,
  17254. 0.2142],
  17255. [-2.2859, -2.2859, 1.8249, -0.8002, 0.0409, -1.2543, 0.8059,
  17256. 0.3050],
  17257. [ 0.6072, -0.4250, 1.8249, -0.0072, -0.4037, 0.1082, 0.6730,
  17258. -0.0881]]], device='cuda:0')
  17259. loss_train_step before backward: tensor(0.0408, device='cuda:0', grad_fn=<MseLossBackward>)
  17260. loss_train_step after backward: tensor(0.0408, device='cuda:0', grad_fn=<MseLossBackward>)
  17261. loss_train: 1.682096759788692
  17262. step: 42
  17263. running loss: 0.040049922852111716
  17264.  
  17265. Train Steps: 42/90 Loss: 0.0400 torch.Size([8, 600, 800])
  17266. torch.Size([8, 8])
  17267. tensor([[0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585],
  17268. [0.6228, 0.4004, 0.8750, 0.5250, 0.3825, 0.5233, 0.6362, 0.5000],
  17269. [0.6151, 0.4058, 0.7068, 0.2680, 0.3400, 0.4083, 0.5775, 0.5733],
  17270. [0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407],
  17271. [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283],
  17272. [0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6188, 0.5433],
  17273. [0.6200, 0.4118, 0.8287, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799],
  17274. [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150]],
  17275. device='cuda:0', dtype=torch.float64)
  17276. predictions are: tensor([[ 0.5004, -0.4306, 1.6711, -0.3141, -0.1863, 0.1056, 0.4462, 0.1948],
  17277. [ 0.3833, -0.5706, 1.7636, 0.1744, -0.3820, -0.0903, 0.6075, 0.1104],
  17278. [ 0.7137, -0.3340, 1.0047, -1.1597, -0.6519, -0.5842, 0.4302, 0.3509],
  17279. [ 0.2005, -0.6191, 1.5591, -0.4497, -0.5971, -1.0111, 0.1772, 0.1835],
  17280. [ 0.3202, -0.6021, 1.6128, 0.3261, -0.4215, -0.4935, 0.3956, 0.1628],
  17281. [ 0.4754, -0.4941, 1.8339, -0.1144, -0.4388, -0.0946, 0.6142, 0.1738],
  17282. [ 0.4219, -0.4805, 1.3762, -0.5846, -0.5242, -1.1056, 0.2978, 0.3035],
  17283. [ 0.3386, -0.5356, 1.7691, -0.1034, -0.1531, 0.1973, 0.6431, 0.2470]],
  17284. device='cuda:0', grad_fn=<AddmmBackward>)
  17285. landmarks are: tensor([[[ 0.5115, -0.4332, 1.7557, -0.3152, -0.2536, 0.3931, 0.4139,
  17286. 0.2936],
  17287. [ 0.5908, -0.4366, 1.7557, 0.1390, -0.5192, 0.1313, 0.6529,
  17288. 0.0236],
  17289. [ 0.5552, -0.4113, 0.9790, -1.0480, -0.7155, -0.3998, 0.3815,
  17290. 0.3623],
  17291. [ 0.5470, -0.3952, 1.6377, -0.4253, -0.6289, -0.8079, 0.0249,
  17292. 0.2116],
  17293. [ 0.5900, -0.4377, 1.6113, 0.3623, -0.5018, -0.3229, 0.3700,
  17294. 0.1544],
  17295. [ 0.5773, -0.4105, 1.8192, -0.1304, -0.5307, 0.0467, 0.5721,
  17296. 0.2237],
  17297. [ 0.5779, -0.3840, 1.5420, -0.4306, -0.5423, -0.9772, 0.2041,
  17298. 0.3928],
  17299. [ 0.6014, -0.3883, 1.8192, -0.1150, -0.2882, 0.4008, 0.5201,
  17300. 0.0928]]], device='cuda:0')
  17301. loss_train_step before backward: tensor(0.0167, device='cuda:0', grad_fn=<MseLossBackward>)
  17302. loss_train_step after backward: tensor(0.0167, device='cuda:0', grad_fn=<MseLossBackward>)
  17303. loss_train: 1.6988404570147395
  17304. step: 43
  17305. running loss: 0.03950791760499394
  17306. Train Steps: 43/90 Loss: 0.0395 torch.Size([8, 600, 800])
  17307. torch.Size([8, 8])
  17308. tensor([[0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207],
  17309. [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758],
  17310. [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600],
  17311. [0.6128, 0.4118, 0.8638, 0.5333, 0.4625, 0.5267, 0.5193, 0.5475],
  17312. [0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767],
  17313. [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690],
  17314. [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483],
  17315. [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5988, 0.5667]],
  17316. device='cuda:0', dtype=torch.float64)
  17317. predictions are: tensor([[ 0.7240, -0.3647, 1.4197, -1.1986, -0.3715, -1.4628, 0.6403, 0.0908],
  17318. [ 0.3467, -0.5586, 1.7756, -0.1293, -0.4171, -0.0376, 0.5929, 0.1243],
  17319. [ 0.5438, -0.4304, 1.6915, 0.2665, -0.4031, -0.1974, 0.4155, 0.2774],
  17320. [ 0.4697, -0.4414, 1.6029, 0.1337, -0.1995, -0.1209, 0.1611, 0.2092],
  17321. [ 0.1763, -0.6875, 1.6918, -0.0366, -0.4339, -0.2375, 0.3021, 0.2240],
  17322. [ 0.3239, -0.5864, 1.5668, 0.0814, -0.3454, 0.1129, 0.7768, 0.3226],
  17323. [ 0.5229, -0.4454, 1.6353, -0.4913, -0.5194, -0.0920, 0.3297, 0.1946],
  17324. [ 0.4346, -0.4854, 1.6011, -0.7661, -0.5156, -0.7754, 0.4753, 0.2727]],
  17325. device='cuda:0', grad_fn=<AddmmBackward>)
  17326. landmarks are: tensor([[[ 0.6132, -0.4248, 1.5247, -1.1620, -0.2882, -1.3159, 0.6545,
  17327. 0.1193],
  17328. [ 0.6072, -0.4250, 1.8249, -0.0072, -0.4037, 0.1082, 0.6730,
  17329. -0.0881],
  17330. [ 0.5883, -0.3594, 1.7557, 0.2545, -0.4152, -0.0611, 0.3353,
  17331. 0.3007],
  17332. [ 0.5447, -0.3838, 1.7037, 0.1775, -0.1497, 0.1467, 0.1128,
  17333. 0.2431],
  17334. [ 0.5610, -0.4381, 1.7730, 0.0390, -0.4326, -0.1458, 0.1794,
  17335. 0.3777],
  17336. [ 0.6306, -0.4153, 1.5141, 0.2224, -0.3691, 0.2622, 1.0033,
  17337. 0.3425],
  17338. [ 0.5546, -0.4620, 1.7788, -0.4229, -0.5192, -0.0226, 0.4277,
  17339. 0.2468],
  17340. [ 0.5783, -0.3933, 1.6748, -0.6154, -0.5769, -0.6462, 0.4797,
  17341. 0.3315]]], device='cuda:0')
  17342. loss_train_step before backward: tensor(0.0154, device='cuda:0', grad_fn=<MseLossBackward>)
  17343. loss_train_step after backward: tensor(0.0154, device='cuda:0', grad_fn=<MseLossBackward>)
  17344. loss_train: 1.7142196567729115
  17345. step: 44
  17346. running loss: 0.03895953765392981
  17347. Train Steps: 44/90 Loss: 0.0390 torch.Size([8, 600, 800])
  17348. torch.Size([8, 8])
  17349. tensor([[ nan, nan, 0.8525, 0.2217, 0.5413, 0.2367, 0.7367, 0.5482],
  17350. [0.6138, 0.4054, 0.8750, 0.4750, 0.4363, 0.5017, 0.5086, 0.5822],
  17351. [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153],
  17352. [0.6185, 0.4079, 0.8838, 0.4617, 0.4838, 0.5650, 0.6175, 0.5850],
  17353. [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500],
  17354. [0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6188, 0.5433],
  17355. [0.6055, 0.4015, 0.7425, 0.2033, 0.4113, 0.1883, 0.5217, 0.4823],
  17356. [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649]],
  17357. device='cuda:0', dtype=torch.float64)
  17358. predictions are: tensor([[-1.8981, -2.0774, 1.7373, -1.0459, 0.1501, -1.2919, 0.9771, 0.2549],
  17359. [ 0.6542, -0.3248, 1.6869, 0.0199, -0.3070, 0.0515, 0.1334, 0.2514],
  17360. [ 0.7948, -0.2716, 1.4147, -0.7527, -0.6649, -0.6969, 0.3543, 0.1676],
  17361. [ 0.5800, -0.3888, 1.8255, -0.1103, -0.1156, 0.1581, 0.5761, 0.3198],
  17362. [ 0.7730, -0.2810, 1.3485, -0.9682, -0.6464, -0.4463, 0.5706, 0.2872],
  17363. [ 0.5955, -0.4088, 1.9148, 0.0193, -0.5055, 0.0592, 0.5711, 0.1925],
  17364. [ 0.5736, -0.3881, 0.9811, -1.1537, -0.4690, -1.4067, 0.1148, 0.1811],
  17365. [ 0.9616, -0.1831, 1.7293, 0.5908, -0.5086, -0.0755, 0.4916, 0.0809]],
  17366. device='cuda:0', grad_fn=<AddmmBackward>)
  17367. landmarks are: tensor([[[-2.2859, -2.2859, 1.6517, -1.2620, 0.2141, -1.1928, 1.1166,
  17368. 0.2463],
  17369. [ 0.5491, -0.4132, 1.7557, -0.0919, -0.2709, 0.0313, 0.0635,
  17370. 0.4032],
  17371. [ 0.5614, -0.4514, 1.3252, -0.7240, -0.6237, -0.7215, 0.4299,
  17372. 0.0942],
  17373. [ 0.5707, -0.4017, 1.7961, -0.1535, -0.0515, 0.3238, 0.5663,
  17374. 0.4162],
  17375. [ 0.5763, -0.4147, 1.3226, -1.0619, -0.6635, -0.4152, 0.5374,
  17376. 0.2545],
  17377. [ 0.5773, -0.4105, 1.8192, -0.1304, -0.5307, 0.0467, 0.5721,
  17378. 0.2237],
  17379. [ 0.5109, -0.4316, 1.1436, -1.3467, -0.3864, -1.4160, 0.1239,
  17380. -0.0580],
  17381. [ 0.6105, -0.4360, 1.6171, 0.5162, -0.5076, 0.0159, 0.5171,
  17382. -0.1385]]], device='cuda:0')
  17383. loss_train_step before backward: tensor(0.0156, device='cuda:0', grad_fn=<MseLossBackward>)
  17384. loss_train_step after backward: tensor(0.0156, device='cuda:0', grad_fn=<MseLossBackward>)
  17385. loss_train: 1.7298358231782913
  17386. step: 45
  17387. running loss: 0.0384407960706287
  17388. Train Steps: 45/90 Loss: 0.0384 torch.Size([8, 600, 800])
  17389. torch.Size([8, 8])
  17390. tensor([[ nan, nan, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552],
  17391. [0.6263, 0.4065, 0.9038, 0.4317, 0.3588, 0.4550, 0.6325, 0.5250],
  17392. [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356],
  17393. [0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567],
  17394. [0.6248, 0.4032, 0.7738, 0.1900, 0.4813, 0.1400, 0.5941, 0.4904],
  17395. [0.6168, 0.4029, 0.8523, 0.3417, 0.3588, 0.5000, 0.6125, 0.5400],
  17396. [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466],
  17397. [0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350]],
  17398. device='cuda:0', dtype=torch.float64)
  17399. predictions are: tensor([[-2.2043, -2.2418, 1.7081, -0.8358, -0.0276, -1.0152, 0.8084, 0.2407],
  17400. [ 0.5243, -0.4429, 1.9431, -0.0100, -0.6092, 0.0366, 0.5699, 0.1601],
  17401. [ 0.6887, -0.3278, 1.3823, -0.8020, -0.3371, -1.0292, 0.6584, 0.2193],
  17402. [ 0.6709, -0.3032, 1.7437, 0.0247, -0.1854, 0.2688, 0.3536, 0.2386],
  17403. [ 0.9483, -0.1798, 1.2172, -0.9029, -0.1352, -1.4696, 0.4495, 0.1328],
  17404. [ 0.6183, -0.3735, 1.6392, -0.4924, -0.5940, 0.1497, 0.6687, 0.2686],
  17405. [ 0.6845, -0.2976, 1.3550, -0.4397, -0.6997, -0.2449, 0.1676, 0.2389],
  17406. [ 0.7395, -0.2648, 1.2949, -0.8578, -0.2455, -1.1228, 0.3436, 0.2488]],
  17407. device='cuda:0', grad_fn=<AddmmBackward>)
  17408. landmarks are: tensor([[[-2.2859, -2.2859, 1.7557, -1.1466, 0.0871, -1.0773, 1.1239,
  17409. 0.2783],
  17410. [ 0.6070, -0.4085, 1.8885, -0.2921, -0.6289, -0.1843, 0.6356,
  17411. 0.1390],
  17412. [ 0.6174, -0.4225, 1.4975, -1.1709, -0.3174, -1.1806, 0.6539,
  17413. 0.1879],
  17414. [ 0.5538, -0.4139, 1.7557, -0.1843, -0.0459, 0.1242, 0.4219,
  17415. 0.2853],
  17416. [ 0.5999, -0.4236, 1.2880, -1.4083, -0.0630, -1.6393, 0.4584,
  17417. -0.0208],
  17418. [ 0.5631, -0.4249, 1.6509, -0.7078, -0.6289, 0.0236, 0.5432,
  17419. 0.2083],
  17420. [ 0.5076, -0.4443, 1.2337, -0.5023, -0.6808, -0.3614, 0.0866,
  17421. 0.2386],
  17422. [ 0.5768, -0.4031, 1.3838, -1.1527, -0.2188, -1.4216, 0.4379,
  17423. 0.1850]]], device='cuda:0')
  17424. loss_train_step before backward: tensor(0.0250, device='cuda:0', grad_fn=<MseLossBackward>)
  17425. loss_train_step after backward: tensor(0.0250, device='cuda:0', grad_fn=<MseLossBackward>)
  17426. loss_train: 1.7548013236373663
  17427. step: 46
  17428. running loss: 0.038147854861681874
  17429.  
  17430. Train Steps: 46/90 Loss: 0.0381 torch.Size([8, 600, 800])
  17431. torch.Size([8, 8])
  17432. tensor([[0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461],
  17433. [0.6259, 0.4156, 0.8812, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960],
  17434. [0.6199, 0.4102, 0.8950, 0.4417, 0.4012, 0.5367, 0.6112, 0.5967],
  17435. [ nan, nan, 0.7425, 0.2117, 0.3937, 0.2433, 0.5438, 0.5567],
  17436. [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183],
  17437. [0.6060, 0.3924, 0.8450, 0.5717, 0.4200, 0.5217, 0.5253, 0.4752],
  17438. [0.6293, 0.3982, 0.8700, 0.5300, 0.3763, 0.4717, 0.7050, 0.5297],
  17439. [0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507]],
  17440. device='cuda:0', dtype=torch.float64)
  17441. predictions are: tensor([[ 0.5808, -0.4027, 1.7523, 0.1851, -0.2221, 0.0181, 0.2749, 0.2247],
  17442. [ 0.1685, -0.6521, 1.8133, -0.9287, -0.1096, -1.2804, 0.6237, 0.1164],
  17443. [ 0.5971, -0.3892, 1.8052, -0.2200, -0.3549, 0.2286, 0.5063, 0.3570],
  17444. [-1.4883, -1.7690, 1.2798, -1.1738, -0.4054, -1.2219, 0.2785, 0.2763],
  17445. [ 0.4992, -0.4773, 1.9323, -0.2267, -0.4326, 0.3214, 0.7804, 0.2136],
  17446. [ 0.7130, -0.3289, 1.7113, 0.3647, -0.3429, 0.0021, 0.2070, 0.1561],
  17447. [ 0.6170, -0.4312, 1.7823, 0.1004, -0.4980, -0.1354, 0.7443, 0.1665],
  17448. [ 1.4584, 0.1012, 1.0517, -1.3167, -0.5159, -1.1998, 0.3566, 0.2365]],
  17449. device='cuda:0', grad_fn=<AddmmBackward>)
  17450. landmarks are: tensor([[[ 0.5443, -0.3831, 1.7095, 0.1621, -0.2016, 0.1390, 0.1437,
  17451. 0.2364],
  17452. [ 0.6052, -0.3663, 1.7845, -0.8156, -0.0804, -1.4237, 0.5866,
  17453. 0.0051],
  17454. [ 0.5772, -0.3913, 1.8480, -0.2459, -0.4326, 0.1929, 0.5374,
  17455. 0.4701],
  17456. [-2.2859, -2.2859, 1.1436, -1.3082, -0.4672, -1.1620, 0.2256,
  17457. 0.2853],
  17458. [ 0.5934, -0.4276, 1.8192, -0.1458, -0.4499, 0.3777, 0.6702,
  17459. 0.1082],
  17460. [ 0.5132, -0.4736, 1.6171, 0.3546, -0.3460, 0.1236, 0.1404,
  17461. -0.0911],
  17462. [ 0.6207, -0.4466, 1.7326, 0.1621, -0.5480, -0.1073, 0.9704,
  17463. 0.1608],
  17464. [ 0.5185, -0.4252, 0.9647, -1.2928, -0.4788, -1.2390, 0.2617,
  17465. 0.2576]]], device='cuda:0')
  17466. loss_train_step before backward: tensor(0.0435, device='cuda:0', grad_fn=<MseLossBackward>)
  17467. loss_train_step after backward: tensor(0.0435, device='cuda:0', grad_fn=<MseLossBackward>)
  17468. loss_train: 1.7982591893523932
  17469. step: 47
  17470. running loss: 0.038260833816008366
  17471. Train Steps: 47/90 Loss: 0.0383 torch.Size([8, 600, 800])
  17472. torch.Size([8, 8])
  17473. tensor([[0.6055, 0.4015, 0.7425, 0.2033, 0.4113, 0.1883, 0.5217, 0.4823],
  17474. [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672],
  17475. [ nan, nan, 0.7268, 0.2333, 0.4125, 0.1933, 0.5112, 0.5383],
  17476. [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332],
  17477. [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167],
  17478. [0.6104, 0.4029, 0.8738, 0.4900, 0.4088, 0.4533, 0.5070, 0.5510],
  17479. [0.6042, 0.3990, 0.6831, 0.2875, 0.3500, 0.3133, 0.5143, 0.5510],
  17480. [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268]],
  17481. device='cuda:0', dtype=torch.float64)
  17482. predictions are: tensor([[ 0.5740, -0.4570, 1.2343, -1.2853, -0.3430, -1.3086, 0.2643, 0.1285],
  17483. [ 0.7208, -0.4197, 1.9882, 0.2828, -0.5127, -0.1009, 0.7220, 0.0963],
  17484. [-1.0776, -1.5501, 1.2808, -1.1677, -0.3018, -1.2895, 0.2042, 0.2265],
  17485. [ 0.6298, -0.4848, 2.0928, -0.0428, -0.5081, -0.2292, 0.8535, 0.1820],
  17486. [ 0.9159, -0.2347, 1.5106, -0.9483, -0.3872, -0.8938, 0.5857, 0.3830],
  17487. [ 0.3516, -0.5812, 1.9792, 0.1283, -0.3523, 0.1517, 0.2751, 0.1903],
  17488. [ 0.7296, -0.3684, 1.1634, -1.1615, -0.5584, -0.6658, 0.1552, 0.3173],
  17489. [ 0.5119, -0.5072, 1.9833, -0.0225, -0.1524, 0.3139, 0.7481, 0.1937]],
  17490. device='cuda:0', grad_fn=<AddmmBackward>)
  17491. landmarks are: tensor([[[ 0.5109, -0.4316, 1.1436, -1.3467, -0.3864, -1.4160, 0.1239,
  17492. -0.0580],
  17493. [ 0.6128, -0.4375, 1.7788, 0.2699, -0.6346, -0.2536, 0.5463,
  17494. -0.1278],
  17495. [-2.2859, -2.2859, 1.0712, -1.2085, -0.3806, -1.3929, 0.0755,
  17496. 0.2006],
  17497. [ 0.6520, -0.4056, 1.9173, -0.0765, -0.5596, -0.4537, 0.7949,
  17498. 0.1768],
  17499. [ 0.5631, -0.4129, 1.2129, -0.9233, -0.4152, -1.0311, 0.4566,
  17500. 0.5624],
  17501. [ 0.5334, -0.4252, 1.7499, -0.0226, -0.3979, -0.1920, 0.0558,
  17502. 0.2589],
  17503. [ 0.5049, -0.4428, 0.8692, -0.9581, -0.6693, -0.8386, 0.0897,
  17504. 0.2589],
  17505. [ 0.5828, -0.4682, 1.7031, -0.0497, -0.2458, 0.0818, 0.6381,
  17506. 0.1474]]], device='cuda:0')
  17507. loss_train_step before backward: tensor(0.0539, device='cuda:0', grad_fn=<MseLossBackward>)
  17508. loss_train_step after backward: tensor(0.0539, device='cuda:0', grad_fn=<MseLossBackward>)
  17509. loss_train: 1.852201757952571
  17510. step: 48
  17511. running loss: 0.038587536624011896
  17512. Train Steps: 48/90 Loss: 0.0386 torch.Size([8, 600, 800])
  17513. torch.Size([8, 8])
  17514. tensor([[0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733],
  17515. [0.6196, 0.4088, 0.8888, 0.4583, 0.4500, 0.5683, 0.6138, 0.5883],
  17516. [ nan, nan, 0.6793, 0.2110, 0.4012, 0.2167, 0.5112, 0.5583],
  17517. [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290],
  17518. [0.6163, 0.4006, 0.8788, 0.4683, 0.3663, 0.4883, 0.5887, 0.5017],
  17519. [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633],
  17520. [0.6201, 0.4064, 0.8688, 0.5050, 0.4225, 0.5100, 0.6138, 0.5500],
  17521. [0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222]],
  17522. device='cuda:0', dtype=torch.float64)
  17523. predictions are: tensor([[ 0.3295, -0.5814, 1.8805, -0.0697, -0.5494, -0.1917, 0.4035, 0.2892],
  17524. [ 0.7263, -0.3292, 1.8847, -0.2392, -0.1815, 0.3605, 0.5436, 0.2755],
  17525. [-1.5498, -1.8469, 1.1289, -1.4847, -0.3832, -1.3001, 0.2906, 0.2223],
  17526. [ 0.7428, -0.3750, 1.4495, -1.3152, -0.2427, -1.3279, 0.7417, 0.0372],
  17527. [ 0.4533, -0.5365, 1.8424, -0.2498, -0.5483, 0.0561, 0.5313, 0.1250],
  17528. [ 0.5942, -0.4407, 1.8397, 0.1174, -0.5398, -0.1064, 0.4411, 0.2406],
  17529. [ 0.7884, -0.3214, 1.8269, 0.0239, -0.3524, 0.1700, 0.4697, 0.1303],
  17530. [ 0.9849, -0.1455, 1.5928, -0.6527, -0.0722, -1.1593, 0.4756, 0.3225]],
  17531. device='cuda:0', grad_fn=<AddmmBackward>)
  17532. landmarks are: tensor([[[ 0.6121, -0.3138, 1.8423, 0.0082, -0.6462, -0.3075, 0.3469,
  17533. 0.3623],
  17534. [ 0.5759, -0.3978, 1.8192, -0.1689, -0.2074, 0.3392, 0.5490,
  17535. 0.4316],
  17536. [-2.2859, -2.2859, 0.8516, -1.3112, -0.4326, -1.2851, 0.0755,
  17537. 0.2930],
  17538. [ 0.6125, -0.4273, 1.2807, -1.3253, -0.2574, -1.2542, 0.6864,
  17539. 0.1575],
  17540. [ 0.5609, -0.4354, 1.7730, -0.1227, -0.5942, -0.0303, 0.4335,
  17541. 0.0313],
  17542. [ 0.5879, -0.3591, 1.8018, 0.1236, -0.5596, -0.1612, 0.3469,
  17543. 0.3161],
  17544. [ 0.5784, -0.4088, 1.7268, 0.0467, -0.3344, 0.0697, 0.5490,
  17545. 0.2545],
  17546. [ 0.6237, -0.2983, 1.3919, -0.4691, -0.0457, -1.2313, 0.2453,
  17547. 0.5882]]], device='cuda:0')
  17548. loss_train_step before backward: tensor(0.0276, device='cuda:0', grad_fn=<MseLossBackward>)
  17549. loss_train_step after backward: tensor(0.0276, device='cuda:0', grad_fn=<MseLossBackward>)
  17550. loss_train: 1.8798270765691996
  17551. step: 49
  17552. running loss: 0.03836381788916734
  17553. Train Steps: 49/90 Loss: 0.0384 torch.Size([8, 600, 800])
  17554. torch.Size([8, 8])
  17555. tensor([[0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413],
  17556. [0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5737, 0.5350],
  17557. [0.6271, 0.4040, 0.9138, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413],
  17558. [0.6203, 0.4073, 0.8189, 0.2398, 0.4400, 0.2054, 0.5929, 0.5501],
  17559. [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717],
  17560. [0.6102, 0.4005, 0.8688, 0.5100, 0.4813, 0.5400, 0.5404, 0.5064],
  17561. [ nan, nan, 0.8888, 0.3100, 0.5262, 0.2817, 0.7145, 0.6003],
  17562. [0.6079, 0.3964, 0.7420, 0.2958, 0.3563, 0.2917, 0.5351, 0.4980]],
  17563. device='cuda:0', dtype=torch.float64)
  17564. predictions are: tensor([[ 0.0420, -0.7969, 1.9109, -0.8322, -0.1011, -0.8858, 0.9601, 0.1443],
  17565. [ 0.6866, -0.3095, 1.7257, -0.0317, -0.5664, -0.5953, 0.2935, 0.1853],
  17566. [ 0.2108, -0.6950, 1.9938, -0.5592, -0.2313, -0.8180, 0.9008, 0.1145],
  17567. [ 0.5528, -0.4017, 1.3345, -1.0168, -0.2567, -1.0105, 0.4153, 0.2544],
  17568. [ 0.6255, -0.3738, 1.1034, -1.1191, -0.5814, -0.7942, 0.3894, 0.3216],
  17569. [ 0.7603, -0.3225, 1.7301, 0.1145, -0.2304, 0.4093, 0.1185, 0.0922],
  17570. [-1.5151, -1.8184, 1.7782, -0.8214, -0.0171, -0.7889, 0.8844, 0.3772],
  17571. [ 0.9260, -0.1868, 1.1814, -1.0478, -0.6314, -0.6874, 0.0614, 0.0822]],
  17572. device='cuda:0', grad_fn=<AddmmBackward>)
  17573. landmarks are: tensor([[[ 0.6108, -0.4201, 1.8711, -0.7848, -0.0053, -1.2236, 1.0362,
  17574. 0.2142],
  17575. [ 0.5901, -0.4000, 1.8423, -0.0688, -0.5307, -0.9233, 0.3642,
  17576. 0.1852],
  17577. [ 0.6108, -0.4201, 1.9346, -0.5538, -0.1497, -1.0773, 1.0545,
  17578. 0.2142],
  17579. [ 0.5792, -0.4048, 1.4965, -1.1781, -0.2534, -1.3371, 0.4528,
  17580. 0.2549],
  17581. [ 0.5746, -0.3882, 1.1436, -1.2005, -0.4903, -1.0157, 0.4393,
  17582. 0.3546],
  17583. [ 0.5328, -0.4361, 1.7268, 0.0697, -0.0630, 0.2083, 0.2103,
  17584. 0.0532],
  17585. [-2.2859, -2.2859, 1.8192, -0.8540, 0.1448, -0.9849, 1.0143,
  17586. 0.4867],
  17587. [ 0.5219, -0.4550, 1.1415, -0.9196, -0.6404, -0.9387, 0.1856,
  17588. 0.0141]]], device='cuda:0')
  17589. loss_train_step before backward: tensor(0.0425, device='cuda:0', grad_fn=<MseLossBackward>)
  17590. loss_train_step after backward: tensor(0.0425, device='cuda:0', grad_fn=<MseLossBackward>)
  17591. loss_train: 1.9223244320601225
  17592. step: 50
  17593. running loss: 0.03844648864120245
  17594.  
  17595. Train Steps: 50/90 Loss: 0.0384 torch.Size([8, 600, 800])
  17596. torch.Size([8, 8])
  17597. tensor([[0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986],
  17598. [ nan, nan, 0.8850, 0.3000, 0.5363, 0.2250, 0.7343, 0.5771],
  17599. [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364],
  17600. [0.6260, 0.4153, 0.9000, 0.4533, 0.4025, 0.2633, 0.6223, 0.4967],
  17601. [0.6250, 0.4146, 0.8838, 0.3933, 0.3588, 0.4283, 0.6162, 0.5367],
  17602. [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167],
  17603. [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116],
  17604. [0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496]],
  17605. device='cuda:0', dtype=torch.float64)
  17606. predictions are: tensor([[ 0.5079, -0.4253, 1.7334, -0.5518, -0.2087, -1.0789, 0.5205, 0.1830],
  17607. [-2.1959, -2.2539, 1.6896, -0.8953, 0.1348, -0.9927, 0.8918, 0.3466],
  17608. [ 0.7811, -0.3041, 1.3094, -1.1007, -0.1904, -1.1196, 0.6205, 0.1609],
  17609. [ 0.6017, -0.3955, 1.7478, -0.2562, -0.3932, -0.8967, 0.5031, 0.0947],
  17610. [ 0.5943, -0.3944, 1.5952, -0.4495, -0.6258, -0.1813, 0.4599, 0.2885],
  17611. [ 0.5591, -0.3863, 1.4813, -0.8121, -0.5755, -0.7021, 0.1551, 0.1755],
  17612. [ 0.7742, -0.2851, 1.6700, -0.2640, -0.4691, 0.5642, 0.4217, 0.1421],
  17613. [ 0.5294, -0.4214, 1.5105, -0.9518, -0.3016, -0.8068, 0.4605, 0.2259]],
  17614. device='cuda:0', grad_fn=<AddmmBackward>)
  17615. landmarks are: tensor([[[ 0.6054, -0.3641, 1.8711, -0.5153, -0.2305, -1.3005, 0.5948,
  17616. 0.0171],
  17617. [-2.2859, -2.2859, 1.8018, -0.9002, 0.1910, -1.2467, 1.1057,
  17618. 0.3799],
  17619. [ 0.6158, -0.3960, 1.4092, -1.2774, -0.2074, -1.1851, 0.8491,
  17620. 0.1917],
  17621. [ 0.6055, -0.3676, 1.8711, -0.1920, -0.4268, -1.0696, 0.5887,
  17622. 0.0081],
  17623. [ 0.6009, -0.3710, 1.7961, -0.4691, -0.6289, -0.3075, 0.5605,
  17624. 0.1929],
  17625. [ 0.5483, -0.4249, 1.5305, -0.7386, -0.6115, -0.8694, 0.3353,
  17626. 0.1005],
  17627. [ 0.5376, -0.3903, 1.7095, -0.3229, -0.4730, 0.4701, 0.3871,
  17628. 0.0772],
  17629. [ 0.5790, -0.4031, 1.6915, -0.9564, -0.4152, -1.1063, 0.4425,
  17630. 0.2528]]], device='cuda:0')
  17631. loss_train_step before backward: tensor(0.0130, device='cuda:0', grad_fn=<MseLossBackward>)
  17632. loss_train_step after backward: tensor(0.0130, device='cuda:0', grad_fn=<MseLossBackward>)
  17633. loss_train: 1.9353253524750471
  17634. step: 51
  17635. running loss: 0.037947555930883274
  17636. Train Steps: 51/90 Loss: 0.0379 torch.Size([8, 600, 800])
  17637. torch.Size([8, 8])
  17638. tensor([[0.6282, 0.4034, 0.7830, 0.2080, 0.4532, 0.2080, 0.6404, 0.5323],
  17639. [ nan, nan, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729],
  17640. [0.6189, 0.4049, 0.8888, 0.4417, 0.4213, 0.5200, 0.5988, 0.5633],
  17641. [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672],
  17642. [0.6117, 0.4019, 0.8538, 0.4067, 0.3513, 0.3583, 0.5663, 0.5133],
  17643. [0.6204, 0.4110, 0.7913, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367],
  17644. [ nan, nan, 0.6793, 0.2110, 0.4012, 0.2167, 0.5112, 0.5583],
  17645. [0.6104, 0.4029, 0.8738, 0.4900, 0.4088, 0.4533, 0.5070, 0.5510]],
  17646. device='cuda:0', dtype=torch.float64)
  17647. predictions are: tensor([[ 0.7340, -0.3327, 1.4879, -1.1681, -0.0782, -1.3344, 0.7709, 0.1514],
  17648. [-0.8746, -1.3776, 1.0971, -1.3145, -0.2678, -1.3062, 0.2473, 0.3377],
  17649. [ 1.0725, -0.1333, 1.8916, -0.2227, -0.3438, 0.3868, 0.5975, 0.2453],
  17650. [ 1.0992, -0.1554, 1.8354, 0.2487, -0.5583, -0.2261, 0.6530, 0.0657],
  17651. [ 0.5148, -0.4407, 1.8467, -0.5045, -0.5949, -0.5906, 0.3270, 0.1523],
  17652. [ 0.8868, -0.2235, 1.5971, -1.0558, -0.2848, -1.0580, 0.6227, 0.1622],
  17653. [-2.1556, -2.2283, 1.1792, -1.2546, -0.2982, -1.1205, 0.3752, 0.3032],
  17654. [ 0.7901, -0.2864, 1.8163, 0.0887, -0.4164, 0.0258, 0.2030, 0.1948]],
  17655. device='cuda:0', grad_fn=<AddmmBackward>)
  17656. landmarks are: tensor([[[ 0.6158, -0.4225, 1.3307, -1.3253, -0.1924, -1.3252, 0.6721,
  17657. 0.1727],
  17658. [-2.2859, -2.2859, 0.8824, -1.2727, -0.3691, -1.4237, 0.0943,
  17659. 0.3604],
  17660. [ 0.5726, -0.4159, 1.8192, -0.2459, -0.3402, 0.1159, 0.4797,
  17661. 0.3161],
  17662. [ 0.6128, -0.4375, 1.7788, 0.2699, -0.6346, -0.2536, 0.5463,
  17663. -0.1278],
  17664. [ 0.5393, -0.4294, 1.6575, -0.4075, -0.6635, -0.6308, 0.3296,
  17665. 0.0851],
  17666. [ 0.5796, -0.3878, 1.3688, -1.0542, -0.4095, -1.1312, 0.5894,
  17667. 0.1929],
  17668. [-2.2859, -2.2859, 0.8516, -1.3112, -0.4326, -1.2851, 0.0755,
  17669. 0.2930],
  17670. [ 0.5334, -0.4252, 1.7499, -0.0226, -0.3979, -0.1920, 0.0558,
  17671. 0.2589]]], device='cuda:0')
  17672. loss_train_step before backward: tensor(0.0704, device='cuda:0', grad_fn=<MseLossBackward>)
  17673. loss_train_step after backward: tensor(0.0704, device='cuda:0', grad_fn=<MseLossBackward>)
  17674. loss_train: 2.0057187657803297
  17675. step: 52
  17676. running loss: 0.038571514726544805
  17677. Train Steps: 52/90 Loss: 0.0386 torch.Size([8, 600, 800])
  17678. torch.Size([8, 8])
  17679. tensor([[0.6259, 0.4156, 0.8812, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960],
  17680. [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167],
  17681. [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200],
  17682. [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284],
  17683. [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402],
  17684. [0.6143, 0.4034, 0.8800, 0.4833, 0.4512, 0.5367, 0.5289, 0.5097],
  17685. [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500],
  17686. [0.6278, 0.4253, 0.8875, 0.5017, 0.4113, 0.2750, 0.5413, 0.6196]],
  17687. device='cuda:0', dtype=torch.float64)
  17688. predictions are: tensor([[-0.0896, -0.8350, 1.8853, -0.9877, -0.0115, -1.3491, 0.7882, 0.0851],
  17689. [ 0.2664, -0.6331, 1.5328, -0.9256, -0.5753, -0.5936, 0.5557, 0.4303],
  17690. [ 0.2831, -0.6234, 1.7790, -0.2026, -0.5764, -0.2057, 0.4492, 0.1349],
  17691. [ 0.2364, -0.6552, 1.3299, -1.3268, -0.1236, -1.4516, 0.5103, 0.1620],
  17692. [ 0.4781, -0.5272, 1.7003, 0.0702, -0.4299, -0.1375, 0.4306, 0.1112],
  17693. [ 0.4198, -0.5174, 1.7888, -0.1580, -0.2072, 0.1132, 0.0995, 0.0633],
  17694. [ 0.4876, -0.4862, 1.3988, -1.0940, -0.6574, -0.4401, 0.6009, 0.2054],
  17695. [ 0.3655, -0.5576, 1.7252, -0.1082, -0.3145, -0.9504, 0.4736, 0.3978]],
  17696. device='cuda:0', grad_fn=<AddmmBackward>)
  17697. landmarks are: tensor([[[ 0.6052, -0.3663, 1.7845, -0.8156, -0.0804, -1.4237, 0.5866,
  17698. 0.0051],
  17699. [ 0.5768, -0.3899, 1.3861, -0.7771, -0.5885, -0.5461, 0.5028,
  17700. 0.5624],
  17701. [ 0.5865, -0.3973, 1.8423, -0.0688, -0.5192, -0.2305, 0.4162,
  17702. 0.1159],
  17703. [ 0.5761, -0.4070, 1.2452, -1.2541, -0.1725, -1.4835, 0.4511,
  17704. 0.1545],
  17705. [ 0.5780, -0.4565, 1.6221, 0.2532, -0.3728, -0.1718, 0.4357,
  17706. 0.2091],
  17707. [ 0.5515, -0.4225, 1.7788, -0.0534, -0.2016, 0.1929, 0.1568,
  17708. 0.0682],
  17709. [ 0.5763, -0.4147, 1.3226, -1.0619, -0.6635, -0.4152, 0.5374,
  17710. 0.2545],
  17711. [ 0.6139, -0.3216, 1.8134, 0.0313, -0.3864, -1.0157, 0.2144,
  17712. 0.5762]]], device='cuda:0')
  17713. loss_train_step before backward: tensor(0.0279, device='cuda:0', grad_fn=<MseLossBackward>)
  17714. loss_train_step after backward: tensor(0.0279, device='cuda:0', grad_fn=<MseLossBackward>)
  17715. loss_train: 2.0335935931652784
  17716. step: 53
  17717. running loss: 0.038369690437080724
  17718. Train Steps: 53/90 Loss: 0.0384 torch.Size([8, 600, 800])
  17719. torch.Size([8, 8])
  17720. tensor([[0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431],
  17721. [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986],
  17722. [0.6179, 0.4008, 0.8600, 0.4015, 0.3932, 0.2515, 0.5711, 0.5438],
  17723. [0.6213, 0.4001, 0.7712, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183],
  17724. [0.6219, 0.4114, 0.8175, 0.2817, 0.3925, 0.2783, 0.5900, 0.5350],
  17725. [0.6207, 0.4110, 0.8738, 0.5000, 0.4800, 0.5633, 0.6300, 0.5433],
  17726. [0.6102, 0.4005, 0.8688, 0.5100, 0.4813, 0.5400, 0.5404, 0.5064],
  17727. [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235]],
  17728. device='cuda:0', dtype=torch.float64)
  17729. predictions are: tensor([[ 0.4297, -0.4967, 1.6639, -0.1452, -0.3522, -0.0779, 0.1092, 0.1016],
  17730. [-0.0545, -0.8022, 1.8791, -0.6957, -0.2669, -1.4643, 0.7409, 0.1956],
  17731. [ 0.1286, -0.6560, 1.5928, -0.6839, -0.3915, -1.2792, 0.3740, 0.2566],
  17732. [ 0.2547, -0.6250, 1.2035, -1.4607, -0.3059, -1.5024, 0.5547, 0.1377],
  17733. [-0.0957, -0.8061, 1.3868, -1.1668, -0.4368, -1.0446, 0.5488, 0.2840],
  17734. [ 0.9410, -0.2207, 1.7451, -0.2033, -0.2315, 0.1644, 0.6318, 0.2716],
  17735. [ 0.5294, -0.4676, 1.7483, -0.0365, -0.1930, -0.0027, 0.2473, 0.0986],
  17736. [ 0.0932, -0.7754, 1.6980, -0.1735, -0.4596, 0.0964, 0.6184, 0.1822]],
  17737. device='cuda:0', grad_fn=<AddmmBackward>)
  17738. landmarks are: tensor([[[ 0.5351, -0.4153, 1.7326, -0.0457, -0.2214, -0.0466, 0.0434,
  17739. 0.2228],
  17740. [ 0.6054, -0.3641, 1.8711, -0.5153, -0.2305, -1.3005, 0.5948,
  17741. 0.0171],
  17742. [ 0.5680, -0.4345, 1.6864, -0.4315, -0.4698, -1.1241, 0.3518,
  17743. 0.2261],
  17744. [ 0.5838, -0.4379, 1.2764, -1.3082, -0.2824, -1.4545, 0.4162,
  17745. 0.1082],
  17746. [ 0.5868, -0.3858, 1.4901, -0.9849, -0.4730, -1.0003, 0.4393,
  17747. 0.1852],
  17748. [ 0.5811, -0.3878, 1.7499, 0.0236, -0.0688, 0.3161, 0.6240,
  17749. 0.2237],
  17750. [ 0.5328, -0.4361, 1.7268, 0.0697, -0.0630, 0.2083, 0.2103,
  17751. 0.0532],
  17752. [ 0.5762, -0.4706, 1.7754, -0.0984, -0.3680, 0.2380, 0.6277,
  17753. 0.1322]]], device='cuda:0')
  17754. loss_train_step before backward: tensor(0.0433, device='cuda:0', grad_fn=<MseLossBackward>)
  17755. loss_train_step after backward: tensor(0.0433, device='cuda:0', grad_fn=<MseLossBackward>)
  17756. loss_train: 2.076877860352397
  17757. step: 54
  17758. running loss: 0.03846070111763698
  17759.  
  17760. Train Steps: 54/90 Loss: 0.0385 torch.Size([8, 600, 800])
  17761. torch.Size([8, 8])
  17762. tensor([[0.6307, 0.4045, 0.8025, 0.5833, 0.3775, 0.4867, 0.6892, 0.5459],
  17763. [0.6159, 0.4085, 0.6900, 0.2283, 0.4088, 0.1950, 0.5123, 0.5397],
  17764. [0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019],
  17765. [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391],
  17766. [0.6200, 0.3999, 0.8653, 0.5207, 0.4100, 0.5125, 0.5975, 0.5103],
  17767. [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364],
  17768. [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207],
  17769. [0.6179, 0.3993, 0.8925, 0.4789, 0.3879, 0.4900, 0.6041, 0.5279]],
  17770. device='cuda:0', dtype=torch.float64)
  17771. predictions are: tensor([[ 0.4562, -0.5427, 1.6782, 0.0694, -0.4969, -0.1446, 0.6702, 0.2226],
  17772. [-0.1811, -0.8887, 1.0253, -1.2883, -0.3301, -1.3747, 0.0681, 0.2783],
  17773. [ 0.8465, -0.3035, 1.8925, -0.1580, -0.5932, -0.4709, 0.5199, 0.1389],
  17774. [ 0.0712, -0.7317, 1.5243, -1.0640, -0.3472, -1.1446, 0.3364, 0.2090],
  17775. [ 0.4380, -0.5539, 1.8446, 0.0722, -0.3768, 0.1330, 0.3437, 0.1530],
  17776. [-0.0180, -0.8051, 1.5171, -1.2229, -0.1305, -1.3399, 0.7107, 0.2390],
  17777. [-0.0261, -0.8366, 1.5963, -1.1512, -0.2253, -1.3229, 0.6682, 0.1688],
  17778. [ 0.6177, -0.4220, 1.8000, -0.2515, -0.5198, -0.0278, 0.3949, 0.1956]],
  17779. device='cuda:0', grad_fn=<AddmmBackward>)
  17780. landmarks are: tensor([[[ 0.6273, -0.4177, 1.4208, 0.4085, -0.5423, -0.0380, 0.8973,
  17781. 0.2356],
  17782. [ 0.5591, -0.3990, 0.9012, -1.2313, -0.3979, -1.3852, 0.0804,
  17783. 0.2071],
  17784. [ 0.6372, -0.4129, 1.8942, -0.0765, -0.6173, -0.4768, 0.6999,
  17785. 0.0325],
  17786. [ 0.5785, -0.4259, 1.4228, -1.0261, -0.4190, -1.2189, 0.4763,
  17787. 0.2043],
  17788. [ 0.5778, -0.4389, 1.7107, 0.1192, -0.3921, 0.0815, 0.4741,
  17789. 0.0711],
  17790. [ 0.6158, -0.3960, 1.4092, -1.2774, -0.2074, -1.1851, 0.8491,
  17791. 0.1917],
  17792. [ 0.6132, -0.4248, 1.5247, -1.1620, -0.2882, -1.3159, 0.6545,
  17793. 0.1193],
  17794. [ 0.5680, -0.4417, 1.8365, -0.0740, -0.4941, -0.0227, 0.5045,
  17795. 0.1525]]], device='cuda:0')
  17796. loss_train_step before backward: tensor(0.0463, device='cuda:0', grad_fn=<MseLossBackward>)
  17797. loss_train_step after backward: tensor(0.0463, device='cuda:0', grad_fn=<MseLossBackward>)
  17798. loss_train: 2.1231579910963774
  17799. step: 55
  17800. running loss: 0.03860287256538868
  17801. Train Steps: 55/90 Loss: 0.0386 torch.Size([8, 600, 800])
  17802. torch.Size([8, 8])
  17803. tensor([[0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118],
  17804. [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717],
  17805. [0.6241, 0.4143, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550],
  17806. [0.6193, 0.4108, 0.7438, 0.2700, 0.3650, 0.3683, 0.6238, 0.5717],
  17807. [0.6228, 0.4004, 0.8750, 0.5250, 0.3825, 0.5233, 0.6362, 0.5000],
  17808. [0.6097, 0.4024, 0.8488, 0.3717, 0.3875, 0.5517, 0.5836, 0.5591],
  17809. [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726],
  17810. [0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933]],
  17811. device='cuda:0', dtype=torch.float64)
  17812. predictions are: tensor([[ 0.3630, -0.5458, 1.6351, 0.1872, -0.3983, -0.4754, 0.3092, 0.3913],
  17813. [ 0.2263, -0.6256, 1.3120, -1.3146, -0.4408, -1.4280, 0.5027, 0.2511],
  17814. [ 0.2667, -0.6203, 1.8972, -0.3692, -0.3622, -0.1678, 0.5342, 0.2368],
  17815. [ 0.2488, -0.6262, 1.3356, -1.1459, -0.5657, -0.8258, 0.5429, 0.2466],
  17816. [ 0.3914, -0.5923, 1.8955, -0.0133, -0.4285, -0.2241, 0.6273, 0.0103],
  17817. [ 0.0806, -0.7299, 1.6935, -0.6597, -0.4001, 0.0085, 0.5942, 0.2190],
  17818. [ 0.4726, -0.4565, 1.3884, -1.1786, -0.5101, -1.2969, 0.1848, -0.0156],
  17819. [ 0.4982, -0.4523, 1.7932, 0.2545, 0.0606, -0.6184, 0.3437, 0.3104]],
  17820. device='cuda:0', grad_fn=<AddmmBackward>)
  17821. landmarks are: tensor([[[ 0.6121, -0.3844, 1.4556, 0.3936, -0.4383, -0.1689, 0.2925,
  17822. 0.5401],
  17823. [ 0.5746, -0.3882, 1.1436, -1.2005, -0.4903, -1.0157, 0.4393,
  17824. 0.3546],
  17825. [ 0.5968, -0.3725, 1.8423, -0.1381, -0.4037, 0.1852, 0.6009,
  17826. 0.2776],
  17827. [ 0.5747, -0.3886, 1.1494, -1.0388, -0.6000, -0.5846, 0.5952,
  17828. 0.3546],
  17829. [ 0.5908, -0.4366, 1.7557, 0.1390, -0.5192, 0.1313, 0.6529,
  17830. 0.0236],
  17831. [ 0.5301, -0.4273, 1.6344, -0.5692, -0.4961, 0.2622, 0.4098,
  17832. 0.2966],
  17833. [ 0.4997, -0.4446, 1.1610, -0.9772, -0.6058, -1.0311, 0.1404,
  17834. -0.1031],
  17835. [ 0.5959, -0.3579, 1.6055, 0.3623, -0.0573, -0.2074, 0.3122,
  17836. 0.4547]]], device='cuda:0')
  17837. loss_train_step before backward: tensor(0.0382, device='cuda:0', grad_fn=<MseLossBackward>)
  17838. loss_train_step after backward: tensor(0.0382, device='cuda:0', grad_fn=<MseLossBackward>)
  17839. loss_train: 2.161322994157672
  17840. step: 56
  17841. running loss: 0.038595053467101285
  17842. Train Steps: 56/90 Loss: 0.0386 torch.Size([8, 600, 800])
  17843. torch.Size([8, 8])
  17844. tensor([[0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250],
  17845. [0.6189, 0.4033, 0.8650, 0.5267, 0.4487, 0.5150, 0.5925, 0.5050],
  17846. [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118],
  17847. [0.6264, 0.3972, 0.8853, 0.4771, 0.3853, 0.4511, 0.6293, 0.5334],
  17848. [0.6197, 0.4090, 0.7825, 0.2500, 0.4200, 0.2483, 0.5988, 0.5667],
  17849. [0.6308, 0.3990, 0.8688, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133],
  17850. [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426],
  17851. [0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6188, 0.5433]],
  17852. device='cuda:0', dtype=torch.float64)
  17853. predictions are: tensor([[ 0.4092, -0.5138, 1.8144, -0.1999, -0.3258, -0.0545, 0.4307, 0.1656],
  17854. [ 0.4702, -0.4808, 1.7870, -0.0377, -0.1724, -0.1545, 0.3535, 0.1485],
  17855. [ 0.3143, -0.5592, 1.5140, 0.1310, -0.4379, -0.4050, 0.2350, 0.4357],
  17856. [ 0.6361, -0.4048, 1.8149, -0.2947, -0.5310, -0.4568, 0.6891, 0.1461],
  17857. [ 0.0101, -0.7609, 1.3474, -1.3365, -0.3493, -1.3944, 0.4397, 0.2876],
  17858. [ 0.4153, -0.5429, 1.7749, -0.1072, -0.4279, -0.2432, 0.6392, 0.1070],
  17859. [ 0.1730, -0.6428, 1.2906, -1.3643, -0.2571, -1.6480, 0.3724, 0.1490],
  17860. [ 0.5979, -0.4057, 1.8060, -0.3230, -0.5050, -0.1868, 0.4154, 0.1172]],
  17861. device='cuda:0', grad_fn=<AddmmBackward>)
  17862. landmarks are: tensor([[[ 0.5991, -0.3803, 1.8018, -0.0534, -0.3460, 0.1852, 0.5374,
  17863. 0.1390],
  17864. [ 0.5726, -0.4230, 1.7095, 0.1467, -0.2132, 0.0928, 0.4508,
  17865. 0.0467],
  17866. [ 0.6121, -0.3844, 1.4556, 0.3936, -0.4383, -0.1689, 0.2925,
  17867. 0.5401],
  17868. [ 0.6075, -0.4514, 1.8032, -0.0822, -0.5061, -0.2023, 0.6208,
  17869. 0.1779],
  17870. [ 0.5764, -0.3969, 1.3284, -1.1312, -0.3460, -1.1389, 0.4797,
  17871. 0.3315],
  17872. [ 0.6275, -0.4430, 1.7268, 0.1082, -0.4614, 0.0159, 0.6644,
  17873. 0.0851],
  17874. [ 0.5784, -0.4153, 1.2972, -1.2541, -0.2265, -1.4553, 0.4741,
  17875. 0.2203],
  17876. [ 0.5773, -0.4105, 1.8192, -0.1304, -0.5307, 0.0467, 0.5721,
  17877. 0.2237]]], device='cuda:0')
  17878. loss_train_step before backward: tensor(0.0283, device='cuda:0', grad_fn=<MseLossBackward>)
  17879. loss_train_step after backward: tensor(0.0283, device='cuda:0', grad_fn=<MseLossBackward>)
  17880. loss_train: 2.189621262252331
  17881. step: 57
  17882. running loss: 0.03841440810969001
  17883. Train Steps: 57/90 Loss: 0.0384 torch.Size([8, 600, 800])
  17884. torch.Size([8, 8])
  17885. tensor([[0.6207, 0.4110, 0.8738, 0.5000, 0.4800, 0.5633, 0.6300, 0.5433],
  17886. [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390],
  17887. [0.6229, 0.4198, 0.7662, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783],
  17888. [0.6189, 0.4049, 0.8888, 0.4417, 0.4213, 0.5200, 0.5988, 0.5633],
  17889. [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5363, 0.5550],
  17890. [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717],
  17891. [0.6339, 0.4123, 0.8638, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436],
  17892. [0.6043, 0.4022, 0.6887, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136]],
  17893. device='cuda:0', dtype=torch.float64)
  17894. predictions are: tensor([[ 0.6948, -0.3456, 1.8998, 0.0423, -0.1258, 0.0501, 0.6125, 0.2276],
  17895. [ 0.3303, -0.5843, 1.8655, 0.0207, -0.4595, -0.1592, 0.3264, 0.0907],
  17896. [ 0.1220, -0.6186, 1.3515, -0.9253, -0.1465, -1.4670, 0.3551, 0.3518],
  17897. [ 0.5260, -0.4394, 1.9115, -0.1421, -0.3737, 0.0497, 0.5564, 0.1762],
  17898. [ 0.0102, -0.7211, 1.0825, -1.0281, -0.4573, -1.3510, 0.0147, 0.2470],
  17899. [ 0.3063, -0.5432, 1.2584, -1.0544, -0.5468, -1.2192, 0.4297, 0.2727],
  17900. [ 0.4030, -0.5388, 1.8117, 0.1530, -0.4069, 0.0941, 0.9434, 0.1687],
  17901. [ 0.5400, -0.4138, 1.0428, -1.2295, -0.5807, -1.2244, 0.2741, 0.0920]],
  17902. device='cuda:0', grad_fn=<AddmmBackward>)
  17903. landmarks are: tensor([[[ 0.5811, -0.3878, 1.7499, 0.0236, -0.0688, 0.3161, 0.6240,
  17904. 0.2237],
  17905. [ 0.5250, -0.4661, 1.7383, -0.0765, -0.4268, -0.0226, 0.2535,
  17906. 0.2035],
  17907. [ 0.5912, -0.3467, 1.2533, -1.0388, -0.1150, -1.3005, 0.3353,
  17908. 0.3854],
  17909. [ 0.5726, -0.4159, 1.8192, -0.2459, -0.3402, 0.1159, 0.4797,
  17910. 0.3161],
  17911. [ 0.5448, -0.3859, 0.9242, -1.1466, -0.4152, -1.3005, 0.1910,
  17912. 0.2776],
  17913. [ 0.5746, -0.3882, 1.1436, -1.2005, -0.4903, -1.0157, 0.4393,
  17914. 0.3546],
  17915. [ 0.6421, -0.3816, 1.7037, 0.1929, -0.4037, 0.2391, 1.1861,
  17916. 0.2249],
  17917. [ 0.5053, -0.4281, 0.8954, -1.3698, -0.5423, -1.1389, 0.2453,
  17918. 0.0862]]], device='cuda:0')
  17919. loss_train_step before backward: tensor(0.0236, device='cuda:0', grad_fn=<MseLossBackward>)
  17920. loss_train_step after backward: tensor(0.0236, device='cuda:0', grad_fn=<MseLossBackward>)
  17921. loss_train: 2.213260428979993
  17922. step: 58
  17923. running loss: 0.038159662568620564
  17924.  
  17925. Train Steps: 58/90 Loss: 0.0382 torch.Size([8, 600, 800])
  17926. torch.Size([8, 8])
  17927. tensor([[0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383],
  17928. [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116],
  17929. [0.6100, 0.4016, 0.8600, 0.5067, 0.4612, 0.5233, 0.5086, 0.5519],
  17930. [0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6138, 0.5450],
  17931. [0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367],
  17932. [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377],
  17933. [0.6250, 0.4106, 0.8700, 0.3717, 0.3588, 0.4967, 0.6038, 0.5167],
  17934. [0.6084, 0.3981, 0.8588, 0.5233, 0.4600, 0.5367, 0.5680, 0.5006]],
  17935. device='cuda:0', dtype=torch.float64)
  17936. predictions are: tensor([[ 0.7516, -0.2493, 1.6080, -0.1194, -0.3354, -0.4171, 0.2348, 0.2148],
  17937. [ 0.1629, -0.6173, 1.6239, -0.3986, -0.4561, 0.0161, 0.4742, 0.1740],
  17938. [ 0.4031, -0.4658, 1.5838, 0.0052, -0.1537, -0.3175, 0.1387, 0.2543],
  17939. [ 0.5602, -0.4197, 1.6715, -0.3033, -0.3595, -0.1238, 0.5330, 0.2328],
  17940. [ 0.4136, -0.4696, 1.6398, -1.0198, -0.4677, -1.4931, 0.7789, 0.1734],
  17941. [ 0.2517, -0.6238, 1.5922, 0.0767, -0.4577, -0.4736, 0.5105, 0.1944],
  17942. [ 0.7147, -0.2776, 1.5206, -0.6886, -0.5991, -0.2994, 0.6749, 0.2496],
  17943. [ 0.4535, -0.4406, 1.5970, 0.0072, -0.1969, -0.2337, 0.1553, 0.1555]],
  17944. device='cuda:0', grad_fn=<AddmmBackward>)
  17945. landmarks are: tensor([[[ 5.3603e-01, -4.6490e-01, 1.6517e+00, 4.6189e-04, -2.8245e-01,
  17946. -6.8822e-02, 2.3086e-01, 2.0046e-01],
  17947. [ 5.3759e-01, -3.9030e-01, 1.7095e+00, -3.2286e-01, -4.7298e-01,
  17948. 4.7005e-01, 3.8714e-01, 7.7228e-02],
  17949. [ 5.3181e-01, -4.3102e-01, 1.6864e+00, 5.4350e-02, -1.5543e-01,
  17950. 1.3133e-01, 6.3480e-02, 2.6342e-01],
  17951. [ 5.9602e-01, -4.1016e-01, 1.8018e+00, -1.6120e-01, -3.3441e-01,
  17952. 1.1594e-01, 5.4896e-01, 2.3141e-01],
  17953. [ 6.1742e-01, -4.2008e-01, 1.7309e+00, -8.7840e-01, -4.7351e-01,
  17954. -9.5238e-01, 6.2423e-01, 1.9310e-01],
  17955. [ 5.7748e-01, -4.6066e-01, 1.6741e+00, 1.9623e-01, -4.0362e-01,
  17956. -1.2115e-01, 4.5876e-01, 1.9786e-01],
  17957. [ 6.0098e-01, -3.8961e-01, 1.7326e+00, -5.6921e-01, -6.2887e-01,
  17958. 8.1601e-03, 5.0277e-01, 1.0054e-01],
  17959. [ 5.2408e-01, -4.4696e-01, 1.6806e+00, 1.3133e-01, -1.6120e-01,
  17960. 1.9292e-01, 3.3778e-01, 2.6129e-02]]], device='cuda:0')
  17961. loss_train_step before backward: tensor(0.0338, device='cuda:0', grad_fn=<MseLossBackward>)
  17962. loss_train_step after backward: tensor(0.0338, device='cuda:0', grad_fn=<MseLossBackward>)
  17963. loss_train: 2.2470400538295507
  17964. step: 59
  17965. running loss: 0.038085424641178825
  17966. Train Steps: 59/90 Loss: 0.0381 torch.Size([8, 600, 800])
  17967. torch.Size([8, 8])
  17968. tensor([[0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672],
  17969. [0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411],
  17970. [0.6204, 0.4013, 0.8075, 0.2400, 0.4313, 0.2050, 0.5800, 0.5150],
  17971. [0.6346, 0.4086, 0.7938, 0.5500, 0.3962, 0.4867, 0.7343, 0.5702],
  17972. [0.6207, 0.4081, 0.7662, 0.2067, 0.3962, 0.3200, 0.6312, 0.5300],
  17973. [ nan, nan, 0.7648, 0.2722, 0.3962, 0.2183, 0.5060, 0.5422],
  17974. [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114],
  17975. [ nan, nan, 0.8488, 0.2300, 0.5563, 0.2100, 0.7390, 0.5679]],
  17976. device='cuda:0', dtype=torch.float64)
  17977. predictions are: tensor([[ 1.4318, 0.1613, 1.6839, -0.1669, -0.6021, -0.5390, 0.4439, 0.0423],
  17978. [ 1.2904, 0.1413, 1.4333, -0.7264, -0.3609, -1.2067, 0.1633, 0.1794],
  17979. [ 1.0495, -0.0377, 1.2722, -0.9776, -0.3320, -1.2180, 0.2905, 0.1976],
  17980. [ 1.0321, -0.0924, 1.4443, 0.2232, -0.4871, 0.0177, 0.6999, 0.2629],
  17981. [ 0.9761, -0.0933, 1.1736, -1.0616, -0.5285, -0.7053, 0.4697, 0.2521],
  17982. [-1.8286, -1.9014, 1.0395, -0.8848, -0.4785, -1.0044, 0.0752, 0.2423],
  17983. [ 0.9323, -0.1344, 1.6433, 0.1998, -0.2636, 0.3281, 0.3945, 0.1279],
  17984. [-2.3898, -2.3125, 1.5499, -0.9304, 0.1178, -1.0389, 0.9971, 0.3760]],
  17985. device='cuda:0', grad_fn=<AddmmBackward>)
  17986. landmarks are: tensor([[[ 0.6132, -0.4324, 1.8192, -0.0842, -0.6231, -0.6385, 0.5537,
  17987. -0.1278],
  17988. [ 0.5713, -0.4321, 1.4975, -0.8134, -0.3094, -1.3345, 0.3779,
  17989. 0.2134],
  17990. [ 0.5796, -0.4326, 1.4439, -1.1774, -0.2940, -1.3390, 0.3931,
  17991. 0.0928],
  17992. [ 0.6454, -0.3984, 1.3804, 0.2545, -0.4557, -0.0380, 1.1057,
  17993. 0.3478],
  17994. [ 0.5809, -0.4011, 1.2533, -1.3313, -0.4557, -0.8079, 0.6298,
  17995. 0.1621],
  17996. [-2.2859, -2.2859, 1.2469, -1.0288, -0.4557, -1.2774, 0.0511,
  17997. 0.2183],
  17998. [ 0.5450, -0.4730, 1.7420, 0.1372, -0.1919, 0.2614, 0.4976,
  17999. 0.0764],
  18000. [-2.2859, -2.2859, 1.6344, -1.2236, 0.2834, -1.3159, 1.1276,
  18001. 0.3371]]], device='cuda:0')
  18002. loss_train_step before backward: tensor(0.0677, device='cuda:0', grad_fn=<MseLossBackward>)
  18003. loss_train_step after backward: tensor(0.0677, device='cuda:0', grad_fn=<MseLossBackward>)
  18004. loss_train: 2.3147859629243612
  18005. step: 60
  18006. running loss: 0.038579766048739354
  18007. Train Steps: 60/90 Loss: 0.0386 torch.Size([8, 600, 800])
  18008. torch.Size([8, 8])
  18009. tensor([[0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131],
  18010. [0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6362, 0.5200],
  18011. [0.6142, 0.3982, 0.8650, 0.4883, 0.3912, 0.4317, 0.5315, 0.5350],
  18012. [0.6225, 0.4116, 0.8662, 0.3517, 0.3663, 0.3233, 0.5837, 0.5317],
  18013. [0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019],
  18014. [0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689],
  18015. [0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343],
  18016. [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600]],
  18017. device='cuda:0', dtype=torch.float64)
  18018. predictions are: tensor([[ 0.6984, -0.3234, 1.5211, -0.3551, -0.5158, -0.1597, 0.2934, 0.1677],
  18019. [ 0.3330, -0.5850, 1.6220, -0.5140, -0.4023, 0.2297, 0.7771, 0.2031],
  18020. [ 0.2629, -0.6181, 1.7049, 0.1271, -0.3293, -0.2368, 0.1755, 0.1922],
  18021. [ 0.8531, -0.1939, 1.6120, -0.5787, -0.4479, -0.9063, 0.3136, 0.2825],
  18022. [ 0.6014, -0.4231, 1.7773, 0.0603, -0.4681, -0.4661, 0.7059, 0.1404],
  18023. [ 0.3605, -0.4942, 1.2824, -0.5865, -0.5712, -0.4829, 0.1422, 0.3245],
  18024. [ 0.4777, -0.4718, 1.7059, -0.5215, -0.2062, -0.9239, 0.8412, 0.1175],
  18025. [ 0.3245, -0.5201, 1.0536, -1.0747, -0.4880, -0.8538, 0.3860, 0.3297]],
  18026. device='cuda:0', grad_fn=<AddmmBackward>)
  18027. landmarks are: tensor([[[ 0.5581, -0.4586, 1.5586, -0.3747, -0.6792, -0.2391, 0.4455,
  18028. 0.0840],
  18029. [ 0.5841, -0.4199, 1.7961, -0.6693, -0.6231, 0.1082, 0.6529,
  18030. 0.1159],
  18031. [ 0.5513, -0.4467, 1.7095, -0.0303, -0.4788, -0.2921, 0.1692,
  18032. 0.1854],
  18033. [ 0.5893, -0.3847, 1.7152, -0.6616, -0.5942, -0.7925, 0.4104,
  18034. 0.1698],
  18035. [ 0.6372, -0.4129, 1.8942, -0.0765, -0.6173, -0.4768, 0.6999,
  18036. 0.0325],
  18037. [ 0.5327, -0.4381, 1.2880, -0.6308, -0.6866, -0.5307, 0.2658,
  18038. 0.3417],
  18039. [ 0.6125, -0.4369, 1.9173, -0.5384, -0.2594, -0.8386, 0.9741,
  18040. 0.1821],
  18041. [ 0.5466, -0.4080, 1.0668, -1.1764, -0.6289, -0.7617, 0.4855,
  18042. 0.3007]]], device='cuda:0')
  18043. loss_train_step before backward: tensor(0.0159, device='cuda:0', grad_fn=<MseLossBackward>)
  18044. loss_train_step after backward: tensor(0.0159, device='cuda:0', grad_fn=<MseLossBackward>)
  18045. loss_train: 2.330700110644102
  18046. step: 61
  18047. running loss: 0.038208198535149215
  18048. Train Steps: 61/90 Loss: 0.0382 torch.Size([8, 600, 800])
  18049. torch.Size([8, 8])
  18050. tensor([[0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117],
  18051. [0.6275, 0.4048, 0.8488, 0.2883, 0.4463, 0.2033, 0.6321, 0.5155],
  18052. [0.6190, 0.4135, 0.8000, 0.4883, 0.3566, 0.3647, 0.5613, 0.5900],
  18053. [ nan, nan, 0.7240, 0.2722, 0.3900, 0.2567, 0.5168, 0.5933],
  18054. [0.6100, 0.4071, 0.7601, 0.3444, 0.3400, 0.4117, 0.5625, 0.5617],
  18055. [0.6304, 0.4024, 0.8925, 0.4800, 0.3937, 0.4817, 0.7485, 0.5297],
  18056. [0.6162, 0.4134, 0.6700, 0.2467, 0.3962, 0.2533, 0.5737, 0.5467],
  18057. [0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058]],
  18058. device='cuda:0', dtype=torch.float64)
  18059. predictions are: tensor([[ 0.5166, -0.4363, 1.5673, 0.4389, -0.3340, 0.1185, 0.6154, 0.3582],
  18060. [ 0.7376, -0.2867, 1.6665, -0.9273, -0.1881, -1.2132, 0.7366, 0.0424],
  18061. [ 1.0151, -0.0670, 1.6360, 0.0200, -0.5590, -0.5761, 0.2286, 0.3074],
  18062. [-1.9647, -2.0599, 1.0725, -0.9805, -0.3861, -0.9972, 0.1053, 0.3411],
  18063. [ 0.7916, -0.2394, 1.3005, -0.7080, -0.6687, -0.2677, 0.3271, 0.2352],
  18064. [ 0.7570, -0.3201, 1.7928, -0.0304, -0.3660, 0.1177, 0.9711, 0.0881],
  18065. [ 0.8833, -0.1852, 1.0046, -1.0671, -0.4564, -1.1528, 0.3791, 0.2964],
  18066. [ 0.7265, -0.2935, 1.6835, -0.2477, -0.4694, -0.0076, 0.2798, 0.0503]],
  18067. device='cuda:0', grad_fn=<AddmmBackward>)
  18068. landmarks are: tensor([[[ 0.6118, -0.3983, 1.5824, 0.3469, -0.4268, -0.0688, 0.3469,
  18069. 0.5393],
  18070. [ 0.6126, -0.4161, 1.6344, -0.9541, -0.2247, -1.3467, 0.6339,
  18071. 0.0953],
  18072. [ 0.5732, -0.3761, 1.4092, -0.0303, -0.6388, -0.6012, 0.3065,
  18073. 0.4393],
  18074. [-2.2859, -2.2859, 1.0580, -1.0288, -0.4845, -1.1004, 0.1011,
  18075. 0.4543],
  18076. [ 0.5318, -0.4056, 1.2249, -0.6949, -0.7155, -0.3844, 0.3122,
  18077. 0.3084],
  18078. [ 0.6257, -0.4273, 1.8365, -0.0688, -0.4672, -0.0611, 1.1715,
  18079. 0.1608],
  18080. [ 0.5603, -0.3764, 0.8088, -1.1466, -0.4557, -1.1158, 0.3642,
  18081. 0.2391],
  18082. [ 0.5245, -0.4361, 1.5940, -0.2921, -0.5480, -0.0919, 0.2432,
  18083. 0.0502]]], device='cuda:0')
  18084. loss_train_step before backward: tensor(0.0206, device='cuda:0', grad_fn=<MseLossBackward>)
  18085. loss_train_step after backward: tensor(0.0206, device='cuda:0', grad_fn=<MseLossBackward>)
  18086. loss_train: 2.351294633001089
  18087. step: 62
  18088. running loss: 0.03792410698388853
  18089.  
  18090. Train Steps: 62/90 Loss: 0.0379 torch.Size([8, 600, 800])
  18091. torch.Size([8, 8])
  18092. tensor([[0.6079, 0.3964, 0.7420, 0.2958, 0.3563, 0.2917, 0.5351, 0.4980],
  18093. [0.6173, 0.4114, 0.7325, 0.2500, 0.4213, 0.1917, 0.5338, 0.5700],
  18094. [0.6182, 0.4058, 0.8738, 0.4350, 0.3563, 0.3400, 0.5290, 0.5822],
  18095. [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690],
  18096. [0.6135, 0.4115, 0.8838, 0.4667, 0.4288, 0.6050, 0.5778, 0.5097],
  18097. [ nan, nan, 0.7625, 0.2433, 0.3713, 0.2867, 0.5235, 0.5220],
  18098. [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901],
  18099. [0.6250, 0.4054, 0.8770, 0.4723, 0.4662, 0.5367, 0.6162, 0.5433]],
  18100. device='cuda:0', dtype=torch.float64)
  18101. predictions are: tensor([[ 0.9962, -0.0934, 1.1704, -0.8884, -0.5775, -0.9059, 0.1283, 0.0411],
  18102. [ 0.5343, -0.3708, 0.9757, -1.0411, -0.3740, -1.2404, 0.1549, 0.3203],
  18103. [ 0.6687, -0.3089, 1.6206, -0.1213, -0.5958, -0.6372, 0.2967, 0.3001],
  18104. [ 0.9464, -0.1766, 1.5838, 0.2979, -0.3140, 0.3399, 1.1289, 0.2496],
  18105. [ 0.8749, -0.1664, 1.7403, -0.0485, -0.2417, 0.4094, 0.6488, 0.1518],
  18106. [-2.2750, -2.2735, 1.1797, -0.9250, -0.4382, -1.0054, 0.2959, 0.2726],
  18107. [ 0.6862, -0.2847, 1.3516, -0.7464, -0.5798, -1.0829, 0.1398, 0.1386],
  18108. [ 0.9209, -0.1630, 1.7334, -0.0049, -0.1730, 0.2321, 0.8091, 0.1891]],
  18109. device='cuda:0', grad_fn=<AddmmBackward>)
  18110. landmarks are: tensor([[[ 0.5219, -0.4550, 1.1415, -0.9196, -0.6404, -0.9387, 0.1856,
  18111. 0.0141],
  18112. [ 0.5652, -0.3858, 1.0975, -1.1312, -0.3402, -1.4006, 0.1794,
  18113. 0.3469],
  18114. [ 0.5697, -0.4113, 1.7499, -0.2767, -0.6404, -0.7155, 0.1576,
  18115. 0.4032],
  18116. [ 0.6306, -0.4153, 1.5141, 0.2224, -0.3691, 0.2622, 1.0033,
  18117. 0.3425],
  18118. [ 0.5477, -0.3851, 1.7961, -0.1304, -0.3055, 0.5085, 0.3830,
  18119. 0.0682],
  18120. [-2.2859, -2.2859, 1.2360, -1.1620, -0.5711, -0.9618, 0.1322,
  18121. 0.1253],
  18122. [ 0.5515, -0.4129, 1.4785, -1.0080, -0.5192, -1.1004, 0.1034,
  18123. -0.0220],
  18124. [ 0.6009, -0.4135, 1.7651, -0.1043, -0.1323, 0.1929, 0.5605,
  18125. 0.2237]]], device='cuda:0')
  18126. loss_train_step before backward: tensor(0.0234, device='cuda:0', grad_fn=<MseLossBackward>)
  18127. loss_train_step after backward: tensor(0.0234, device='cuda:0', grad_fn=<MseLossBackward>)
  18128. loss_train: 2.3746544364839792
  18129. step: 63
  18130. running loss: 0.03769292756323776
  18131. Train Steps: 63/90 Loss: 0.0377 torch.Size([8, 600, 800])
  18132. torch.Size([8, 8])
  18133. tensor([[0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5988, 0.5667],
  18134. [0.6233, 0.4091, 0.8100, 0.2950, 0.3563, 0.3883, 0.6013, 0.5200],
  18135. [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098],
  18136. [0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166],
  18137. [0.6134, 0.4090, 0.6926, 0.2819, 0.3538, 0.3233, 0.5563, 0.5667],
  18138. [0.6224, 0.4179, 0.8700, 0.5683, 0.4037, 0.4683, 0.5650, 0.5600],
  18139. [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400],
  18140. [0.6144, 0.4032, 0.8563, 0.3283, 0.3525, 0.4200, 0.5775, 0.5583]],
  18141. device='cuda:0', dtype=torch.float64)
  18142. predictions are: tensor([[ 0.7226, -0.3009, 1.7418, -0.4842, -0.5559, -0.6611, 0.4734, 0.2265],
  18143. [ 0.4656, -0.4734, 1.4132, -0.7718, -0.6248, -0.4430, 0.5061, 0.1726],
  18144. [ 0.5246, -0.4583, 1.6505, 0.0021, -0.4564, 0.2503, 0.5228, 0.1319],
  18145. [ 0.3909, -0.5617, 1.6733, 0.2320, -0.2253, 0.2285, 0.9119, 0.1082],
  18146. [ 0.7132, -0.3086, 0.9763, -0.9335, -0.6716, -0.7552, 0.1523, 0.2934],
  18147. [ 0.4919, -0.4394, 1.6127, 0.4938, -0.3679, -0.0914, 0.3918, 0.3192],
  18148. [ 0.6757, -0.3431, 1.2730, -1.1612, -0.1780, -1.3853, 0.4469, 0.1643],
  18149. [ 0.1408, -0.6819, 1.6216, -0.6210, -0.6415, -0.3399, 0.3708, 0.2362]],
  18150. device='cuda:0', grad_fn=<AddmmBackward>)
  18151. landmarks are: tensor([[[ 0.5783, -0.3933, 1.6748, -0.6154, -0.5769, -0.6462, 0.4797,
  18152. 0.3315],
  18153. [ 0.5932, -0.3962, 1.4554, -0.9233, -0.6404, -0.4922, 0.4912,
  18154. 0.1159],
  18155. [ 0.5466, -0.4706, 1.7198, -0.0903, -0.5712, 0.1261, 0.4733,
  18156. 0.0688],
  18157. [ 0.6042, -0.4273, 1.6920, 0.1860, -0.2717, 0.1406, 0.7997,
  18158. 0.1004],
  18159. [ 0.5473, -0.3966, 0.9131, -0.9838, -0.6520, -0.7925, 0.2834,
  18160. 0.3315],
  18161. [ 0.5891, -0.3557, 1.7326, 0.3392, -0.4210, -0.1227, 0.3238,
  18162. 0.3007],
  18163. [ 0.5842, -0.3848, 1.2938, -1.2159, -0.2132, -1.4160, 0.5778,
  18164. 0.2083],
  18165. [ 0.5520, -0.4237, 1.6691, -0.7694, -0.6577, -0.3460, 0.3815,
  18166. 0.2930]]], device='cuda:0')
  18167. loss_train_step before backward: tensor(0.0106, device='cuda:0', grad_fn=<MseLossBackward>)
  18168. loss_train_step after backward: tensor(0.0106, device='cuda:0', grad_fn=<MseLossBackward>)
  18169. loss_train: 2.3852399922907352
  18170. step: 64
  18171. running loss: 0.03726937487954274
  18172. Train Steps: 64/90 Loss: 0.0373 torch.Size([8, 600, 800])
  18173. torch.Size([8, 8])
  18174. tensor([[0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317],
  18175. [0.6250, 0.4054, 0.8770, 0.4723, 0.4662, 0.5367, 0.6162, 0.5433],
  18176. [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6088, 0.5583],
  18177. [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167],
  18178. [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400],
  18179. [0.6125, 0.3974, 0.7725, 0.2517, 0.3538, 0.3317, 0.5887, 0.5500],
  18180. [0.6310, 0.4017, 0.8563, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006],
  18181. [0.6087, 0.3976, 0.8337, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300]],
  18182. device='cuda:0', dtype=torch.float64)
  18183. predictions are: tensor([[-0.2325, -0.9602, 1.2720, -1.0271, -0.3637, -0.9876, 0.4289, 0.2526],
  18184. [ 0.4505, -0.5013, 1.8136, 0.0927, -0.1843, 0.4526, 0.6604, 0.1773],
  18185. [ 0.8687, -0.2430, 1.0114, -1.1226, -0.5207, -0.9863, 0.3914, 0.2113],
  18186. [ 0.6572, -0.3616, 1.5121, -0.5709, -0.6501, -0.3371, 0.4752, 0.4255],
  18187. [ 0.7217, -0.3336, 1.3004, -1.0443, -0.2111, -1.2127, 0.4349, 0.1557],
  18188. [ 0.3431, -0.5483, 1.2653, -0.8805, -0.6536, -0.6157, 0.3194, 0.1887],
  18189. [ 0.3079, -0.6517, 1.7307, 0.5960, -0.5675, 0.1268, 0.7210, 0.0536],
  18190. [ 0.6070, -0.3925, 1.6147, -0.2691, -0.6091, -0.5846, 0.4420, 0.1639]],
  18191. device='cuda:0', grad_fn=<AddmmBackward>)
  18192. landmarks are: tensor([[[ 0.5849, -0.3836, 1.2649, -1.2236, -0.3460, -1.2313, 0.4508,
  18193. 0.1698],
  18194. [ 0.6009, -0.4135, 1.7651, -0.1043, -0.1323, 0.1929, 0.5605,
  18195. 0.2237],
  18196. [ 0.5992, -0.3968, 0.9307, -1.3497, -0.4730, -1.0465, 0.5259,
  18197. 0.2930],
  18198. [ 0.5768, -0.3899, 1.3861, -0.7771, -0.5885, -0.5461, 0.5028,
  18199. 0.5624],
  18200. [ 0.5842, -0.3848, 1.2938, -1.2159, -0.2132, -1.4160, 0.5778,
  18201. 0.2083],
  18202. [ 0.5430, -0.4503, 1.2822, -1.1235, -0.6520, -0.7540, 0.4335,
  18203. 0.2545],
  18204. [ 0.6286, -0.4303, 1.6691, 0.3931, -0.5827, -0.0919, 0.6359,
  18205. 0.0261],
  18206. [ 0.5255, -0.4495, 1.5651, -0.4999, -0.5711, -0.8463, 0.4566,
  18207. 0.1621]]], device='cuda:0')
  18208. loss_train_step before backward: tensor(0.0348, device='cuda:0', grad_fn=<MseLossBackward>)
  18209. loss_train_step after backward: tensor(0.0348, device='cuda:0', grad_fn=<MseLossBackward>)
  18210. loss_train: 2.420039724558592
  18211. step: 65
  18212. running loss: 0.03723138037782449
  18213. Train Steps: 65/90 Loss: 0.0372 torch.Size([8, 600, 800])
  18214. torch.Size([8, 8])
  18215. tensor([[0.6357, 0.4097, 0.9038, 0.3883, 0.4213, 0.2950, 0.6686, 0.5390],
  18216. [0.6254, 0.3993, 0.8988, 0.4767, 0.3987, 0.5517, 0.6955, 0.5285],
  18217. [0.6203, 0.4096, 0.8862, 0.4267, 0.3538, 0.4117, 0.6025, 0.5650],
  18218. [0.6117, 0.4019, 0.8538, 0.4067, 0.3513, 0.3583, 0.5663, 0.5133],
  18219. [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617],
  18220. [0.6273, 0.4143, 0.8750, 0.5700, 0.3987, 0.4717, 0.6013, 0.5467],
  18221. [0.6161, 0.4024, 0.8662, 0.4683, 0.4935, 0.5364, 0.6063, 0.5567],
  18222. [0.6257, 0.4167, 0.8775, 0.3433, 0.3563, 0.4133, 0.6200, 0.5667]],
  18223. device='cuda:0', dtype=torch.float64)
  18224. predictions are: tensor([[ 0.5645, -0.4307, 1.5633, -0.5971, -0.4191, -0.9917, 0.6887, 0.1340],
  18225. [ 0.7224, -0.3567, 1.5991, -0.2279, -0.4591, 0.3066, 0.8598, 0.1349],
  18226. [ 0.2056, -0.6534, 1.6414, -0.3374, -0.5785, -0.4410, 0.4228, 0.3512],
  18227. [ 0.9953, -0.1275, 1.5085, -0.5903, -0.7189, -0.6876, 0.1175, 0.0700],
  18228. [ 0.3773, -0.5384, 1.5414, -0.1393, -0.3429, 0.0346, 0.3330, 0.2650],
  18229. [ 0.5671, -0.4097, 1.5155, 0.2661, -0.4947, -0.1401, 0.4437, 0.1082],
  18230. [ 0.5217, -0.4431, 1.5807, -0.2839, -0.1187, 0.0570, 0.4562, 0.2333],
  18231. [ 0.3458, -0.5701, 1.5426, -0.6712, -0.6604, -0.4480, 0.5075, 0.3324]],
  18232. device='cuda:0', grad_fn=<AddmmBackward>)
  18233. landmarks are: tensor([[[ 0.6504, -0.3936, 1.8885, -0.4922, -0.3402, -0.9233, 0.8022,
  18234. 0.2035],
  18235. [ 0.6026, -0.4417, 1.8654, -0.0842, -0.4441, 0.2622, 0.9265,
  18236. 0.1554],
  18237. [ 0.5790, -0.3940, 1.8076, -0.3152, -0.6520, -0.3844, 0.4970,
  18238. 0.3238],
  18239. [ 0.5393, -0.4294, 1.6575, -0.4075, -0.6635, -0.6308, 0.3296,
  18240. 0.0851],
  18241. [ 0.5637, -0.4143, 1.7519, -0.0787, -0.3055, -0.0149, 0.3758,
  18242. 0.3084],
  18243. [ 0.6115, -0.3724, 1.7557, 0.3469, -0.4441, -0.1073, 0.4912,
  18244. 0.2391],
  18245. [ 0.5598, -0.4273, 1.7152, -0.1227, -0.0065, 0.1917, 0.5143,
  18246. 0.2853],
  18247. [ 0.6040, -0.3614, 1.7672, -0.7001, -0.6404, -0.3768, 0.5778,
  18248. 0.3315]]], device='cuda:0')
  18249. loss_train_step before backward: tensor(0.0216, device='cuda:0', grad_fn=<MseLossBackward>)
  18250. loss_train_step after backward: tensor(0.0216, device='cuda:0', grad_fn=<MseLossBackward>)
  18251. loss_train: 2.4416198935359716
  18252. step: 66
  18253. running loss: 0.03699424081115109
  18254.  
  18255. Train Steps: 66/90 Loss: 0.0370 torch.Size([8, 600, 800])
  18256. torch.Size([8, 8])
  18257. tensor([[0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6038, 0.4833],
  18258. [0.6213, 0.4001, 0.7712, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183],
  18259. [0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6262, 0.5167],
  18260. [0.6095, 0.3970, 0.8688, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940],
  18261. [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268],
  18262. [ nan, nan, 0.7725, 0.2611, 0.3675, 0.2733, 0.5413, 0.5167],
  18263. [0.6129, 0.4063, 0.8738, 0.5250, 0.4313, 0.4733, 0.5230, 0.5874],
  18264. [0.6261, 0.4066, 0.8325, 0.2150, 0.4763, 0.2667, 0.7002, 0.5633]],
  18265. device='cuda:0', dtype=torch.float64)
  18266. predictions are: tensor([[ 0.5820, -0.4485, 1.6251, -0.0173, -0.6775, 0.0058, 0.4772, 0.1054],
  18267. [ 0.8377, -0.2557, 1.0665, -1.3392, -0.4191, -1.2162, 0.4193, 0.1196],
  18268. [ 0.7974, -0.2742, 1.7813, -0.2587, -0.6334, -0.5775, 0.6602, 0.2203],
  18269. [ 0.5028, -0.4743, 1.6421, -0.0655, -0.2851, 0.0431, 0.2469, 0.1449],
  18270. [ 0.6925, -0.3438, 1.5984, -0.0479, -0.4077, 0.1826, 0.6601, 0.2109],
  18271. [-1.8189, -2.0004, 1.1048, -1.0873, -0.5757, -0.9374, 0.1628, 0.3239],
  18272. [ 0.8944, -0.1868, 1.5583, 0.3516, -0.4822, 0.1158, 0.2916, 0.2641],
  18273. [ 0.9293, -0.2237, 1.5071, -1.1302, -0.2633, -1.0934, 1.0246, 0.1812]],
  18274. device='cuda:0', grad_fn=<AddmmBackward>)
  18275. landmarks are: tensor([[[ 0.5726, -0.4795, 1.7788, -0.0149, -0.5711, -0.0688, 0.5028,
  18276. -0.0534],
  18277. [ 0.5838, -0.4379, 1.2764, -1.3082, -0.2824, -1.4545, 0.4162,
  18278. 0.1082],
  18279. [ 0.6094, -0.3947, 1.8885, -0.2998, -0.5769, -0.6770, 0.6067,
  18280. 0.1005],
  18281. [ 0.5292, -0.4520, 1.7268, -0.0842, -0.0413, -0.0324, 0.1116,
  18282. -0.0039],
  18283. [ 0.5828, -0.4682, 1.7031, -0.0497, -0.2458, 0.0818, 0.6381,
  18284. 0.1474],
  18285. [-2.2859, -2.2859, 1.2820, -1.0801, -0.5885, -1.0234, 0.2141,
  18286. 0.1005],
  18287. [ 0.5453, -0.4091, 1.7499, 0.1390, -0.2940, -0.0996, 0.1300,
  18288. 0.4272],
  18289. [ 0.6059, -0.4080, 1.5594, -1.2928, -0.0861, -1.0542, 0.9485,
  18290. 0.3157]]], device='cuda:0')
  18291. loss_train_step before backward: tensor(0.0246, device='cuda:0', grad_fn=<MseLossBackward>)
  18292. loss_train_step after backward: tensor(0.0246, device='cuda:0', grad_fn=<MseLossBackward>)
  18293. loss_train: 2.466248968616128
  18294. step: 67
  18295. running loss: 0.03680968609874818
  18296. Train Steps: 67/90 Loss: 0.0368 torch.Size([8, 600, 800])
  18297. torch.Size([8, 8])
  18298. tensor([[0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320],
  18299. [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650],
  18300. [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938],
  18301. [0.6125, 0.3974, 0.7725, 0.2517, 0.3538, 0.3317, 0.5887, 0.5500],
  18302. [0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183],
  18303. [0.6307, 0.4045, 0.8025, 0.5833, 0.3775, 0.4867, 0.6892, 0.5459],
  18304. [0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750],
  18305. [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5413, 0.5433]],
  18306. device='cuda:0', dtype=torch.float64)
  18307. predictions are: tensor([[ 0.5870, -0.4610, 1.3650, -1.2479, -0.2693, -0.9599, 0.7566, 0.1759],
  18308. [ 0.4017, -0.5527, 1.6589, -0.4986, -0.6020, -0.2526, 0.3103, 0.3728],
  18309. [ 0.7367, -0.3825, 1.8921, 0.1706, -0.5680, -0.3303, 0.6146, -0.0270],
  18310. [ 0.3524, -0.5784, 1.2265, -1.1055, -0.5940, -0.6406, 0.4431, 0.2381],
  18311. [ 0.4678, -0.5461, 1.7167, -0.0406, -0.5576, -0.2412, 0.6385, 0.1505],
  18312. [ 0.4932, -0.5308, 1.5870, 0.1907, -0.4714, 0.0821, 0.8611, 0.1655],
  18313. [ 0.6406, -0.3934, 1.2931, -0.9672, -0.4253, -0.9065, 0.1885, 0.2612],
  18314. [ 0.3304, -0.5758, 1.4735, -0.6033, -0.6111, -0.6446, 0.1847, 0.2778]],
  18315. device='cuda:0', grad_fn=<AddmmBackward>)
  18316. landmarks are: tensor([[[ 0.6141, -0.4153, 1.4208, -1.2697, -0.2940, -1.0234, 0.8644,
  18317. 0.1715],
  18318. [ 0.5757, -0.3917, 1.7095, -0.4768, -0.6346, -0.4229, 0.3931,
  18319. 0.3238],
  18320. [ 0.6273, -0.4105, 1.8480, 0.1082, -0.5538, -0.5076, 0.6414,
  18321. -0.0049],
  18322. [ 0.5430, -0.4503, 1.2822, -1.1235, -0.6520, -0.7540, 0.4335,
  18323. 0.2545],
  18324. [ 0.5944, -0.4543, 1.8018, 0.0082, -0.6058, -0.4306, 0.4162,
  18325. 0.1082],
  18326. [ 0.6273, -0.4177, 1.4208, 0.4085, -0.5423, -0.0380, 0.8973,
  18327. 0.2356],
  18328. [ 0.5695, -0.3927, 1.3226, -0.9002, -0.4672, -1.1928, 0.1737,
  18329. 0.3700],
  18330. [ 0.5894, -0.3479, 1.7730, -0.6847, -0.5538, -0.7155, 0.2141,
  18331. 0.2237]]], device='cuda:0')
  18332. loss_train_step before backward: tensor(0.0130, device='cuda:0', grad_fn=<MseLossBackward>)
  18333. loss_train_step after backward: tensor(0.0130, device='cuda:0', grad_fn=<MseLossBackward>)
  18334. loss_train: 2.479281820356846
  18335. step: 68
  18336. running loss: 0.036460026769953614
  18337. Train Steps: 68/90 Loss: 0.0365 torch.Size([8, 600, 800])
  18338. torch.Size([8, 8])
  18339. tensor([[0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400],
  18340. [0.6182, 0.4058, 0.8738, 0.4350, 0.3563, 0.3400, 0.5290, 0.5822],
  18341. [0.6203, 0.4078, 0.8800, 0.5083, 0.3900, 0.5000, 0.6100, 0.5583],
  18342. [0.6183, 0.4076, 0.8838, 0.4517, 0.3813, 0.4483, 0.5775, 0.5633],
  18343. [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483],
  18344. [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333],
  18345. [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567],
  18346. [0.6128, 0.4022, 0.8738, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064]],
  18347. device='cuda:0', dtype=torch.float64)
  18348. predictions are: tensor([[ 0.4208, -0.5286, 1.5660, -0.6811, -0.6707, -0.0938, 0.6795, 0.2433],
  18349. [ 0.4844, -0.5045, 1.7132, -0.2291, -0.6871, -0.6187, 0.3231, 0.2310],
  18350. [ 0.5048, -0.5052, 1.8143, 0.1818, -0.5250, 0.1356, 0.5827, 0.1163],
  18351. [ 0.4867, -0.5049, 1.7367, -0.1605, -0.5433, -0.1209, 0.4980, 0.3073],
  18352. [ 0.8457, -0.3071, 1.1189, -1.3358, -0.4476, -1.1324, 0.6848, 0.1563],
  18353. [ 0.5788, -0.4468, 1.1256, -1.3333, -0.3039, -1.3215, 0.4436, 0.2026],
  18354. [ 0.4527, -0.5231, 1.6824, -0.2289, -0.6501, -0.3667, 0.5120, 0.2052],
  18355. [ 0.2446, -0.6634, 1.8283, 0.1008, -0.0463, 0.0247, 0.2871, 0.1155]],
  18356. device='cuda:0', grad_fn=<AddmmBackward>)
  18357. landmarks are: tensor([[[ 0.5631, -0.4008, 1.6113, -0.7309, -0.6693, -0.1304, 0.5836,
  18358. 0.2083],
  18359. [ 0.5697, -0.4113, 1.7499, -0.2767, -0.6404, -0.7155, 0.1576,
  18360. 0.4032],
  18361. [ 0.5794, -0.4023, 1.7788, 0.0620, -0.4845, 0.0236, 0.5316,
  18362. 0.2930],
  18363. [ 0.5700, -0.4032, 1.7961, -0.1997, -0.5249, -0.2151, 0.3815,
  18364. 0.3161],
  18365. [ 0.5784, -0.4085, 1.0859, -1.3929, -0.4037, -1.1158, 0.5605,
  18366. 0.2468],
  18367. [ 0.5853, -0.3920, 1.1090, -1.3313, -0.2882, -1.3390, 0.4624,
  18368. 0.1775],
  18369. [ 0.5432, -0.4336, 1.7095, -0.1766, -0.5942, -0.4845, 0.3007,
  18370. 0.2853],
  18371. [ 0.5446, -0.4280, 1.7499, 0.0543, 0.0156, 0.1301, 0.1918,
  18372. 0.0532]]], device='cuda:0')
  18373. loss_train_step before backward: tensor(0.0102, device='cuda:0', grad_fn=<MseLossBackward>)
  18374. loss_train_step after backward: tensor(0.0102, device='cuda:0', grad_fn=<MseLossBackward>)
  18375. loss_train: 2.4894718173891306
  18376. step: 69
  18377. running loss: 0.03607930170129175
  18378. Train Steps: 69/90 Loss: 0.0361 torch.Size([8, 600, 800])
  18379. torch.Size([8, 8])
  18380. tensor([[0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585],
  18381. [0.6200, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391],
  18382. [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167],
  18383. [0.6043, 0.4022, 0.6887, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136],
  18384. [0.6143, 0.4034, 0.8800, 0.4833, 0.4512, 0.5367, 0.5289, 0.5097],
  18385. [0.6188, 0.4099, 0.7400, 0.2433, 0.3962, 0.2750, 0.6162, 0.5467],
  18386. [0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767],
  18387. [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550]],
  18388. device='cuda:0', dtype=torch.float64)
  18389. predictions are: tensor([[ 0.4360, -0.4805, 1.6811, -0.2378, -0.3917, 0.2554, 0.4442, 0.2100],
  18390. [ 0.7784, -0.3134, 1.6368, -0.8815, -0.3902, -1.1743, 0.6127, 0.2010],
  18391. [ 0.4541, -0.5240, 1.7817, 0.1039, -0.5630, -0.0397, 0.6083, 0.1318],
  18392. [ 0.6691, -0.3933, 0.9687, -1.3212, -0.6521, -1.1426, 0.4844, 0.1541],
  18393. [ 0.1000, -0.7177, 1.7854, -0.0132, -0.3303, -0.0130, 0.2209, 0.1332],
  18394. [ 0.6156, -0.4212, 1.1427, -1.2524, -0.5990, -1.0458, 0.6910, 0.2183],
  18395. [ 0.3454, -0.6151, 1.7236, 0.0457, -0.5889, -0.2072, 0.4643, 0.2432],
  18396. [ 0.4638, -0.4857, 1.7544, -0.1624, -0.1677, -0.1337, 0.4243, 0.2240]],
  18397. device='cuda:0', grad_fn=<AddmmBackward>)
  18398. landmarks are: tensor([[[ 0.5115, -0.4332, 1.7557, -0.3152, -0.2536, 0.3931, 0.4139,
  18399. 0.2936],
  18400. [ 0.5776, -0.3987, 1.6005, -0.9121, -0.3844, -1.2358, 0.4247,
  18401. 0.2043],
  18402. [ 0.5828, -0.4066, 1.7557, 0.0774, -0.3691, -0.0226, 0.4277,
  18403. 0.1005],
  18404. [ 0.5053, -0.4281, 0.8954, -1.3698, -0.5423, -1.1389, 0.2453,
  18405. 0.0862],
  18406. [ 0.5515, -0.4225, 1.7788, -0.0534, -0.2016, 0.1929, 0.1568,
  18407. 0.0682],
  18408. [ 0.5725, -0.3926, 1.1321, -1.1620, -0.4557, -1.0157, 0.5605,
  18409. 0.2391],
  18410. [ 0.5610, -0.4381, 1.7730, 0.0390, -0.4326, -0.1458, 0.1794,
  18411. 0.3777],
  18412. [ 0.5351, -0.4321, 1.7326, -0.1381, 0.0463, 0.1049, 0.4162,
  18413. 0.2776]]], device='cuda:0')
  18414. loss_train_step before backward: tensor(0.0172, device='cuda:0', grad_fn=<MseLossBackward>)
  18415. loss_train_step after backward: tensor(0.0172, device='cuda:0', grad_fn=<MseLossBackward>)
  18416. loss_train: 2.506653221324086
  18417. step: 70
  18418. running loss: 0.03580933173320123
  18419.  
  18420. Train Steps: 70/90 Loss: 0.0358 torch.Size([8, 600, 800])
  18421. torch.Size([8, 8])
  18422. tensor([[0.6260, 0.4153, 0.9000, 0.4533, 0.4025, 0.2633, 0.6223, 0.4967],
  18423. [0.6198, 0.4115, 0.7762, 0.2717, 0.3713, 0.3200, 0.5837, 0.5683],
  18424. [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500],
  18425. [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334],
  18426. [0.6212, 0.4033, 0.8938, 0.4167, 0.3813, 0.4267, 0.5613, 0.5583],
  18427. [0.6156, 0.4125, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084],
  18428. [0.6250, 0.3993, 0.9138, 0.4333, 0.3763, 0.5217, 0.6995, 0.5320],
  18429. [0.6265, 0.4071, 0.8875, 0.3367, 0.3975, 0.3350, 0.6312, 0.5250]],
  18430. device='cuda:0', dtype=torch.float64)
  18431. predictions are: tensor([[ 0.8442, -0.2878, 1.8822, -0.1673, -0.3492, -1.0421, 0.6640, 0.0426],
  18432. [ 0.7231, -0.3500, 1.3377, -0.9462, -0.5609, -0.9033, 0.3942, 0.3295],
  18433. [ 0.1892, -0.6626, 1.2548, -0.9844, -0.6395, -0.4658, 0.4720, 0.2314],
  18434. [ 0.4308, -0.5277, 1.1750, -1.1513, -0.5298, -0.9072, 0.5386, 0.2045],
  18435. [ 0.3504, -0.5871, 1.8031, -0.1725, -0.5336, -0.3180, 0.3269, 0.2640],
  18436. [ 0.2974, -0.5743, 1.7718, 0.0869, -0.1664, 0.3221, 0.2445, 0.2308],
  18437. [ 0.6129, -0.4227, 1.6943, -0.1902, -0.5263, 0.1815, 0.7341, 0.1576],
  18438. [ 0.4973, -0.4933, 1.8406, -0.6789, -0.4059, -0.7973, 0.6918, 0.1629]],
  18439. device='cuda:0', grad_fn=<AddmmBackward>)
  18440. landmarks are: tensor([[[ 0.6055, -0.3676, 1.8711, -0.1920, -0.4268, -1.0696, 0.5887,
  18441. 0.0081],
  18442. [ 0.5768, -0.3852, 1.2995, -1.0311, -0.5711, -0.8079, 0.4104,
  18443. 0.3392],
  18444. [ 0.5763, -0.4147, 1.3226, -1.0619, -0.6635, -0.4152, 0.5374,
  18445. 0.2545],
  18446. [ 0.5730, -0.4562, 1.2195, -1.2440, -0.5497, -0.7711, 0.5704,
  18447. 0.1779],
  18448. [ 0.5832, -0.4231, 1.8423, -0.3614, -0.5249, -0.3152, 0.3065,
  18449. 0.2930],
  18450. [ 0.5573, -0.3808, 1.8018, -0.0534, -0.1843, 0.4008, 0.3769,
  18451. 0.0622],
  18452. [ 0.6010, -0.4417, 1.9346, -0.2844, -0.5480, 0.1236, 0.9448,
  18453. 0.1715],
  18454. [ 0.6078, -0.4057, 1.8134, -0.7309, -0.4499, -0.7386, 0.6298,
  18455. 0.1390]]], device='cuda:0')
  18456. loss_train_step before backward: tensor(0.0133, device='cuda:0', grad_fn=<MseLossBackward>)
  18457. loss_train_step after backward: tensor(0.0133, device='cuda:0', grad_fn=<MseLossBackward>)
  18458. loss_train: 2.5199545985087752
  18459. step: 71
  18460. running loss: 0.03549231828885599
  18461. Train Steps: 71/90 Loss: 0.0355 torch.Size([8, 600, 800])
  18462. torch.Size([8, 8])
  18463. tensor([[0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611],
  18464. [0.6117, 0.4019, 0.8538, 0.4067, 0.3513, 0.3583, 0.5663, 0.5133],
  18465. [0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533],
  18466. [0.6151, 0.4125, 0.8738, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483],
  18467. [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114],
  18468. [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533],
  18469. [0.6161, 0.4024, 0.8838, 0.4583, 0.3688, 0.3733, 0.5311, 0.5344],
  18470. [0.6268, 0.4052, 0.8175, 0.2250, 0.4688, 0.1917, 0.6375, 0.5267]],
  18471. device='cuda:0', dtype=torch.float64)
  18472. predictions are: tensor([[ 0.5248, -0.4641, 1.7781, -0.0666, -0.2967, -0.0768, 0.3448, 0.2732],
  18473. [ 0.6372, -0.3929, 1.7426, -0.5425, -0.6774, -0.6999, 0.3333, 0.1048],
  18474. [ 0.3841, -0.5738, 1.0240, -1.3658, -0.5510, -0.8458, 0.6108, 0.2919],
  18475. [ 0.4925, -0.4407, 1.6825, -0.1932, -0.6649, -0.5765, 0.1744, 0.2428],
  18476. [ 0.3824, -0.5599, 1.7717, 0.1097, -0.2361, 0.2542, 0.5080, 0.1504],
  18477. [ 0.6134, -0.4053, 1.7990, -0.2906, -0.3894, 0.1740, 0.5379, 0.2180],
  18478. [ 0.6294, -0.4055, 1.8176, -0.1581, -0.6121, -0.5588, 0.3933, 0.0921],
  18479. [ 0.5168, -0.5085, 1.4544, -1.3228, -0.1340, -1.4168, 0.8669, 0.1577]],
  18480. device='cuda:0', grad_fn=<AddmmBackward>)
  18481. landmarks are: tensor([[[ 0.5491, -0.4290, 1.7788, -0.1073, -0.2651, -0.1073, 0.2555,
  18482. 0.3057],
  18483. [ 0.5393, -0.4294, 1.6575, -0.4075, -0.6635, -0.6308, 0.3296,
  18484. 0.0851],
  18485. [ 0.5878, -0.4052, 1.0229, -1.2855, -0.5596, -0.8232, 0.5316,
  18486. 0.2699],
  18487. [ 0.5551, -0.3807, 1.7499, -0.2459, -0.6346, -0.5384, 0.0871,
  18488. 0.2468],
  18489. [ 0.5450, -0.4730, 1.7420, 0.1372, -0.1919, 0.2614, 0.4976,
  18490. 0.0764],
  18491. [ 0.5499, -0.4225, 1.8018, -0.2921, -0.3055, 0.0543, 0.4046,
  18492. 0.2699],
  18493. [ 0.5598, -0.4273, 1.7961, -0.1689, -0.5827, -0.5615, 0.1671,
  18494. 0.1824],
  18495. [ 0.6092, -0.4143, 1.4901, -1.2467, -0.1208, -1.4006, 0.6587,
  18496. 0.1467]]], device='cuda:0')
  18497. loss_train_step before backward: tensor(0.0061, device='cuda:0', grad_fn=<MseLossBackward>)
  18498. loss_train_step after backward: tensor(0.0061, device='cuda:0', grad_fn=<MseLossBackward>)
  18499. loss_train: 2.5260266615077853
  18500. step: 72
  18501. running loss: 0.03508370363205257
  18502. Train Steps: 72/90 Loss: 0.0351 torch.Size([8, 600, 800])
  18503. torch.Size([8, 8])
  18504. tensor([[0.6268, 0.4061, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300],
  18505. [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114],
  18506. [0.6224, 0.4179, 0.8700, 0.5683, 0.4037, 0.4683, 0.5650, 0.5600],
  18507. [0.6218, 0.4137, 0.7263, 0.2233, 0.4075, 0.2650, 0.6212, 0.5783],
  18508. [0.6278, 0.4253, 0.8875, 0.5017, 0.4113, 0.2750, 0.5413, 0.6196],
  18509. [0.6168, 0.4029, 0.8523, 0.3417, 0.3588, 0.5000, 0.6125, 0.5400],
  18510. [0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494],
  18511. [0.6040, 0.4002, 0.7338, 0.2267, 0.3975, 0.2100, 0.5231, 0.4778]],
  18512. device='cuda:0', dtype=torch.float64)
  18513. predictions are: tensor([[ 0.6334, -0.4061, 1.6728, -1.1904, -0.2485, -1.1383, 0.6056, 0.1502],
  18514. [ 0.3768, -0.5504, 1.8360, 0.0306, -0.2621, 0.2948, 0.3617, 0.1141],
  18515. [ 0.7082, -0.3276, 1.7454, 0.2261, -0.4818, -0.0875, 0.3203, 0.2773],
  18516. [ 0.4360, -0.5399, 1.2143, -1.1695, -0.4310, -1.0023, 0.5435, 0.3108],
  18517. [ 0.7105, -0.3180, 1.7618, -0.0368, -0.3727, -0.7612, 0.3573, 0.3785],
  18518. [ 0.3095, -0.5736, 1.6237, -0.7393, -0.6324, -0.0113, 0.5981, 0.2196],
  18519. [ 0.5194, -0.4909, 1.8626, 0.1112, -0.5627, -0.3060, 0.6890, 0.1156],
  18520. [ 0.3677, -0.5692, 1.1701, -1.4718, -0.4682, -1.2223, 0.2472, 0.0670]],
  18521. device='cuda:0', grad_fn=<AddmmBackward>)
  18522. landmarks are: tensor([[[ 0.6093, -0.4104, 1.5709, -1.1620, -0.1727, -1.2313, 0.6471,
  18523. 0.1621],
  18524. [ 0.5450, -0.4730, 1.7420, 0.1372, -0.1919, 0.2614, 0.4976,
  18525. 0.0764],
  18526. [ 0.5891, -0.3557, 1.7326, 0.3392, -0.4210, -0.1227, 0.3238,
  18527. 0.3007],
  18528. [ 0.5863, -0.3749, 1.0686, -1.2543, -0.4037, -1.0619, 0.5836,
  18529. 0.3854],
  18530. [ 0.6139, -0.3216, 1.8134, 0.0313, -0.3864, -1.0157, 0.2144,
  18531. 0.5762],
  18532. [ 0.5631, -0.4249, 1.6509, -0.7078, -0.6289, 0.0236, 0.5432,
  18533. 0.2083],
  18534. [ 0.6224, -0.4032, 1.5940, 0.2930, -0.5769, -0.2690, 0.8900,
  18535. 0.2516],
  18536. [ 0.5037, -0.4375, 1.1032, -1.2390, -0.4499, -1.3159, 0.1301,
  18537. -0.0791]]], device='cuda:0')
  18538. loss_train_step before backward: tensor(0.0123, device='cuda:0', grad_fn=<MseLossBackward>)
  18539. loss_train_step after backward: tensor(0.0123, device='cuda:0', grad_fn=<MseLossBackward>)
  18540. loss_train: 2.5383365219458938
  18541. step: 73
  18542. running loss: 0.03477173317734101
  18543. Train Steps: 73/90 Loss: 0.0348 torch.Size([8, 600, 800])
  18544. torch.Size([8, 8])
  18545. tensor([[0.6163, 0.4114, 0.7650, 0.2017, 0.3763, 0.2867, 0.5631, 0.5071],
  18546. [ nan, nan, 0.8488, 0.2300, 0.5563, 0.2100, 0.7390, 0.5679],
  18547. [0.6263, 0.4065, 0.9038, 0.4317, 0.3588, 0.4550, 0.6325, 0.5250],
  18548. [0.6200, 0.3993, 0.8519, 0.4923, 0.3962, 0.4717, 0.6013, 0.5433],
  18549. [0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5837, 0.5583],
  18550. [0.6144, 0.4032, 0.8563, 0.3283, 0.3525, 0.4200, 0.5775, 0.5583],
  18551. [0.6200, 0.4118, 0.8287, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799],
  18552. [0.6142, 0.3982, 0.8650, 0.4883, 0.3912, 0.4317, 0.5315, 0.5350]],
  18553. device='cuda:0', dtype=torch.float64)
  18554. predictions are: tensor([[ 5.4418e-01, -4.4780e-01, 1.1559e+00, -1.3561e+00, -5.1064e-01,
  18555. -1.0292e+00, 4.7806e-01, 1.2951e-01],
  18556. [-1.2765e+00, -1.6316e+00, 1.6362e+00, -1.2697e+00, 1.1559e-01,
  18557. -9.7170e-01, 9.1181e-01, 3.6843e-01],
  18558. [ 8.5290e-01, -2.6447e-01, 1.9745e+00, -1.9718e-01, -5.8324e-01,
  18559. -2.4832e-01, 6.2102e-01, 1.3964e-01],
  18560. [ 6.6774e-01, -3.8083e-01, 1.7173e+00, 6.7244e-04, -4.7820e-01,
  18561. -1.5158e-01, 5.3323e-01, 1.8132e-01],
  18562. [ 6.6050e-01, -3.5493e-01, 1.7779e+00, -1.7826e-01, -1.4851e-01,
  18563. 1.2158e-01, 2.9447e-01, 1.5800e-01],
  18564. [ 1.0210e+00, -1.2668e-01, 1.6734e+00, -6.9517e-01, -6.8150e-01,
  18565. -4.4491e-01, 4.3289e-01, 2.1880e-01],
  18566. [ 6.1925e-01, -3.6422e-01, 1.4968e+00, -5.0518e-01, -4.9594e-01,
  18567. -1.0194e+00, 1.9179e-01, 3.0572e-01],
  18568. [ 7.5316e-01, -3.0276e-01, 1.7571e+00, -1.9632e-03, -4.9168e-01,
  18569. -3.3726e-01, 1.9005e-01, 1.0943e-01]], device='cuda:0',
  18570. grad_fn=<AddmmBackward>)
  18571. landmarks are: tensor([[[ 0.5606, -0.3856, 1.2476, -1.3544, -0.5480, -0.9618, 0.3152,
  18572. 0.0562],
  18573. [-2.2859, -2.2859, 1.6344, -1.2236, 0.2834, -1.3159, 1.1276,
  18574. 0.3371],
  18575. [ 0.6070, -0.4085, 1.8885, -0.2921, -0.6289, -0.1843, 0.6356,
  18576. 0.1390],
  18577. [ 0.5777, -0.4415, 1.6491, -0.0122, -0.4557, -0.1073, 0.4912,
  18578. 0.2237],
  18579. [ 0.5223, -0.4336, 1.7557, -0.2074, -0.0573, 0.3084, 0.4104,
  18580. 0.2930],
  18581. [ 0.5520, -0.4237, 1.6691, -0.7694, -0.6577, -0.3460, 0.3815,
  18582. 0.2930],
  18583. [ 0.5779, -0.3840, 1.5420, -0.4306, -0.5423, -0.9772, 0.2041,
  18584. 0.3928],
  18585. [ 0.5513, -0.4467, 1.7095, -0.0303, -0.4788, -0.2921, 0.1692,
  18586. 0.1854]]], device='cuda:0')
  18587. loss_train_step before backward: tensor(0.0367, device='cuda:0', grad_fn=<MseLossBackward>)
  18588. loss_train_step after backward: tensor(0.0367, device='cuda:0', grad_fn=<MseLossBackward>)
  18589. loss_train: 2.5750327026471496
  18590. step: 74
  18591. running loss: 0.03479773922496148
  18592.  
  18593. Train Steps: 74/90 Loss: 0.0348 torch.Size([8, 600, 800])
  18594. torch.Size([8, 8])
  18595. tensor([[0.6127, 0.4084, 0.8700, 0.4467, 0.3987, 0.4317, 0.5013, 0.5471],
  18596. [0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124],
  18597. [0.6164, 0.4076, 0.8838, 0.4117, 0.3713, 0.5550, 0.6238, 0.5350],
  18598. [0.6064, 0.4019, 0.8650, 0.4517, 0.4037, 0.5367, 0.5703, 0.5609],
  18599. [0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938],
  18600. [0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367],
  18601. [0.6346, 0.4092, 0.7712, 0.5917, 0.4037, 0.4767, 0.7343, 0.5725],
  18602. [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5413, 0.5433]],
  18603. device='cuda:0', dtype=torch.float64)
  18604. predictions are: tensor([[ 0.5096, -0.4283, 1.7634, -0.3190, -0.4290, -0.4513, 0.0348, 0.2677],
  18605. [ 0.3356, -0.5704, 1.5252, -1.1793, -0.3212, -1.1210, 0.4800, 0.1549],
  18606. [ 0.5342, -0.4191, 1.7516, -0.4826, -0.4391, 0.1963, 0.6230, 0.2563],
  18607. [ 0.5286, -0.4179, 1.7052, -0.2810, -0.3513, 0.1736, 0.3214, 0.2292],
  18608. [ 0.6839, -0.3655, 1.7865, 0.1565, -0.4345, -0.2359, 0.5560, 0.0668],
  18609. [ 0.5197, -0.4576, 1.6927, -0.9587, -0.3475, -1.0610, 0.6220, 0.1999],
  18610. [ 0.7338, -0.3127, 1.4894, -0.0715, -0.4081, -0.2553, 0.6845, 0.3432],
  18611. [ 0.4452, -0.4472, 1.5220, -0.7479, -0.5569, -0.9219, 0.1381, 0.2604]],
  18612. device='cuda:0', grad_fn=<AddmmBackward>)
  18613. landmarks are: tensor([[[ 0.5441, -0.3997, 1.7326, -0.2228, -0.4441, -0.2921, 0.0296,
  18614. 0.2409],
  18615. [ 0.5790, -0.4079, 1.5929, -1.0630, -0.4729, -1.0725, 0.4137,
  18616. 0.0807],
  18617. [ 0.5614, -0.4032, 1.7961, -0.3844, -0.5711, 0.2776, 0.5952,
  18618. 0.1852],
  18619. [ 0.5151, -0.4296, 1.7095, -0.1997, -0.4210, 0.1929, 0.3484,
  18620. 0.3047],
  18621. [ 0.6395, -0.4213, 1.7037, 0.3623, -0.4326, -0.1073, 0.6560,
  18622. -0.0049],
  18623. [ 0.6174, -0.4201, 1.7309, -0.8784, -0.4735, -0.9524, 0.6242,
  18624. 0.1931],
  18625. [ 0.6454, -0.3960, 1.2764, 0.4470, -0.4210, -0.0842, 1.1057,
  18626. 0.3585],
  18627. [ 0.5894, -0.3479, 1.7730, -0.6847, -0.5538, -0.7155, 0.2141,
  18628. 0.2237]]], device='cuda:0')
  18629. loss_train_step before backward: tensor(0.0165, device='cuda:0', grad_fn=<MseLossBackward>)
  18630. loss_train_step after backward: tensor(0.0165, device='cuda:0', grad_fn=<MseLossBackward>)
  18631. loss_train: 2.5915102371945977
  18632. step: 75
  18633. running loss: 0.0345534698292613
  18634. Train Steps: 75/90 Loss: 0.0346 torch.Size([8, 600, 800])
  18635. torch.Size([8, 8])
  18636. tensor([[0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220],
  18637. [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609],
  18638. [0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297],
  18639. [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617],
  18640. [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332],
  18641. [0.6139, 0.4019, 0.7137, 0.2150, 0.4375, 0.1533, 0.5293, 0.5006],
  18642. [0.6263, 0.4233, 0.7924, 0.4626, 0.3788, 0.2883, 0.5573, 0.6047],
  18643. [0.6259, 0.4156, 0.8812, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960]],
  18644. device='cuda:0', dtype=torch.float64)
  18645. predictions are: tensor([[ 0.2447, -0.6079, 1.0104, -1.2852, -0.3396, -1.1869, 0.1303, 0.1883],
  18646. [-0.0586, -0.8399, 1.7326, -1.1093, 0.0978, -0.9643, 0.9543, 0.2626],
  18647. [ 0.6468, -0.3746, 1.8626, 0.2272, -0.6019, -0.0139, 0.5560, 0.1825],
  18648. [ 0.6586, -0.3544, 1.8994, -0.1186, -0.5412, -0.0036, 0.1801, 0.2295],
  18649. [ 0.6833, -0.3836, 1.8911, -0.0879, -0.6091, -0.1609, 0.6603, 0.1590],
  18650. [ 0.1767, -0.6590, 1.0074, -1.3677, -0.3177, -1.2696, 0.1919, 0.1461],
  18651. [ 0.6589, -0.3171, 1.5171, -0.1376, -0.4783, -0.5075, 0.2700, 0.4525],
  18652. [ 0.7460, -0.2777, 1.8169, -0.8115, -0.1521, -0.7355, 0.5067, 0.1502]],
  18653. device='cuda:0', grad_fn=<AddmmBackward>)
  18654. landmarks are: tensor([[[ 0.5566, -0.4160, 0.9935, -1.3313, -0.2824, -1.5161, 0.2144,
  18655. 0.1253],
  18656. [ 0.6372, -0.3575, 1.6633, -1.2082, 0.2199, -1.2467, 1.1313,
  18657. 0.3050],
  18658. [ 0.6504, -0.3647, 1.7730, 0.2930, -0.6058, -0.2382, 0.7109,
  18659. 0.1608],
  18660. [ 0.5488, -0.4221, 1.8018, -0.2459, -0.4845, -0.3075, 0.2309,
  18661. 0.3087],
  18662. [ 0.6520, -0.4056, 1.9173, -0.0765, -0.5596, -0.4537, 0.7949,
  18663. 0.1768],
  18664. [ 0.5496, -0.4295, 1.0109, -1.2928, -0.2651, -1.5777, 0.1589,
  18665. 0.0261],
  18666. [ 0.6069, -0.3309, 1.3742, -0.1493, -0.5365, -0.9541, 0.2884,
  18667. 0.5071],
  18668. [ 0.6052, -0.3663, 1.7845, -0.8156, -0.0804, -1.4237, 0.5866,
  18669. 0.0051]]], device='cuda:0')
  18670. loss_train_step before backward: tensor(0.0396, device='cuda:0', grad_fn=<MseLossBackward>)
  18671. loss_train_step after backward: tensor(0.0396, device='cuda:0', grad_fn=<MseLossBackward>)
  18672. loss_train: 2.6311407880857587
  18673. step: 76
  18674. running loss: 0.03462027352744419
  18675. Train Steps: 76/90 Loss: 0.0346 torch.Size([8, 600, 800])
  18676. torch.Size([8, 8])
  18677. tensor([[0.6022, 0.3994, 0.8025, 0.3350, 0.3350, 0.4400, 0.5565, 0.5025],
  18678. [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350],
  18679. [ nan, nan, 0.7425, 0.2117, 0.3937, 0.2433, 0.5438, 0.5567],
  18680. [0.6321, 0.4048, 0.8738, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927],
  18681. [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333],
  18682. [0.6346, 0.4144, 0.9088, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899],
  18683. [0.6245, 0.4100, 0.7762, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417],
  18684. [0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407]],
  18685. device='cuda:0', dtype=torch.float64)
  18686. predictions are: tensor([[ 0.5096, -0.4109, 1.4637, -0.8476, -0.5801, -0.3472, 0.3674, 0.1769],
  18687. [ 0.8256, -0.1998, 1.7899, -0.3321, -0.4646, -0.4428, 0.3457, 0.2799],
  18688. [-1.3500, -1.6118, 1.2457, -1.3102, -0.3704, -1.1211, 0.2283, 0.2938],
  18689. [ 0.8079, -0.2665, 1.8021, 0.1917, -0.4374, -0.2552, 0.4940, 0.0330],
  18690. [ 0.8357, -0.2062, 1.8751, -0.1593, -0.4265, 0.0671, 0.5050, 0.2274],
  18691. [ 0.6870, -0.3474, 1.9166, -0.1231, -0.2887, -0.2234, 0.7814, 0.2500],
  18692. [ 0.5397, -0.4013, 1.2299, -1.1997, 0.0241, -1.3590, 0.4714, 0.2773],
  18693. [ 0.7879, -0.2330, 1.6780, -0.3961, -0.5048, -0.8428, 0.1642, 0.2339]],
  18694. device='cuda:0', grad_fn=<AddmmBackward>)
  18695. landmarks are: tensor([[[ 0.4956, -0.4413, 1.4208, -0.7386, -0.7386, -0.2536, 0.2843,
  18696. 0.0351],
  18697. [ 0.6072, -0.3239, 1.8423, -0.3537, -0.4961, -0.3921, 0.2083,
  18698. 0.1852],
  18699. [-2.2859, -2.2859, 1.1436, -1.3082, -0.4672, -1.1620, 0.2256,
  18700. 0.2853],
  18701. [ 0.6335, -0.4162, 1.7499, 0.3084, -0.4961, -0.2459, 0.6524,
  18702. -0.0102],
  18703. [ 0.6003, -0.3747, 1.8885, -0.1997, -0.5018, -0.0149, 0.5490,
  18704. 0.1775],
  18705. [ 0.6454, -0.3719, 1.9115, -0.1304, -0.5076, -0.2844, 1.0033,
  18706. 0.4386],
  18707. [ 0.5985, -0.3921, 1.2995, -1.0927, 0.0062, -1.5854, 0.4277,
  18708. 0.2160],
  18709. [ 0.5470, -0.3952, 1.6377, -0.4253, -0.6289, -0.8079, 0.0249,
  18710. 0.2116]]], device='cuda:0')
  18711. loss_train_step before backward: tensor(0.0323, device='cuda:0', grad_fn=<MseLossBackward>)
  18712. loss_train_step after backward: tensor(0.0323, device='cuda:0', grad_fn=<MseLossBackward>)
  18713. loss_train: 2.6634310306981206
  18714. step: 77
  18715. running loss: 0.03459001338568988
  18716. Train Steps: 77/90 Loss: 0.0346 torch.Size([8, 600, 800])
  18717. torch.Size([8, 8])
  18718. tensor([[0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5963, 0.6217],
  18719. [0.6185, 0.4079, 0.8838, 0.4617, 0.4838, 0.5650, 0.6175, 0.5850],
  18720. [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617],
  18721. [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383],
  18722. [0.6255, 0.4017, 0.8688, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901],
  18723. [0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742],
  18724. [0.6160, 0.4093, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808],
  18725. [0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967]],
  18726. device='cuda:0', dtype=torch.float64)
  18727. predictions are: tensor([[ 0.4709, -0.4453, 1.0841, -1.2571, -0.3121, -1.3903, 0.4219, 0.4682],
  18728. [ 0.5359, -0.4140, 1.8498, -0.2117, -0.0671, 0.0072, 0.5131, 0.3493],
  18729. [ 0.7026, -0.3289, 1.8133, -0.0557, -0.2773, -0.1168, 0.3260, 0.2713],
  18730. [ 0.3245, -0.5751, 1.8032, -0.2144, -0.1927, 0.1364, 0.4431, 0.2329],
  18731. [ 0.6127, -0.4122, 1.7185, -0.8684, -0.6133, -0.9153, 0.6240, 0.0393],
  18732. [ 0.4521, -0.5100, 1.7841, -0.8017, -0.4028, -1.2877, 0.5973, 0.0130],
  18733. [ 0.4815, -0.4352, 1.6452, -0.2379, -0.5157, -0.3379, 0.2251, 0.3758],
  18734. [ 0.7428, -0.3001, 1.7455, -0.2779, -0.5834, -0.0526, 0.3788, 0.1368]],
  18735. device='cuda:0', grad_fn=<AddmmBackward>)
  18736. landmarks are: tensor([[[ 0.5687, -0.4088, 1.0397, -1.1466, -0.3113, -1.1928, 0.4681,
  18737. 0.5855],
  18738. [ 0.5707, -0.4017, 1.7961, -0.1535, -0.0515, 0.3238, 0.5663,
  18739. 0.4162],
  18740. [ 0.5637, -0.4143, 1.7519, -0.0787, -0.3055, -0.0149, 0.3758,
  18741. 0.3084],
  18742. [ 0.5664, -0.4321, 1.8249, -0.2074, -0.2651, 0.4162, 0.5663,
  18743. 0.2006],
  18744. [ 0.6031, -0.4307, 1.7268, -0.8002, -0.6058, -0.6462, 0.6442,
  18745. -0.0220],
  18746. [ 0.6134, -0.4218, 1.7268, -0.6154, -0.4730, -1.0850, 0.5463,
  18747. -0.0957],
  18748. [ 0.5592, -0.3956, 1.5543, -0.2456, -0.5885, -0.1689, 0.1392,
  18749. 0.3968],
  18750. [ 0.4974, -0.4482, 1.6633, -0.3306, -0.6173, 0.1313, 0.2925,
  18751. 0.0081]]], device='cuda:0')
  18752. loss_train_step before backward: tensor(0.0132, device='cuda:0', grad_fn=<MseLossBackward>)
  18753. loss_train_step after backward: tensor(0.0132, device='cuda:0', grad_fn=<MseLossBackward>)
  18754. loss_train: 2.6766006434336305
  18755. step: 78
  18756. running loss: 0.03431539286453372
  18757.  
  18758. Train Steps: 78/90 Loss: 0.0343 torch.Size([8, 600, 800])
  18759. torch.Size([8, 8])
  18760. tensor([[0.6200, 0.4059, 0.8700, 0.4900, 0.4163, 0.5000, 0.6162, 0.5467],
  18761. [0.6197, 0.4090, 0.7825, 0.2500, 0.4200, 0.2483, 0.5988, 0.5667],
  18762. [0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967],
  18763. [0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019],
  18764. [0.6339, 0.4123, 0.8638, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436],
  18765. [0.6321, 0.4048, 0.8738, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927],
  18766. [0.6261, 0.4029, 0.8720, 0.3364, 0.3665, 0.3753, 0.6531, 0.5183],
  18767. [0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719]],
  18768. device='cuda:0', dtype=torch.float64)
  18769. predictions are: tensor([[ 0.5273, -0.4399, 1.8495, -0.1630, -0.2124, -0.1018, 0.4078, 0.2698],
  18770. [ 0.5313, -0.4043, 1.3250, -1.3841, -0.2848, -1.3463, 0.4527, 0.3200],
  18771. [ 0.6812, -0.3249, 1.7316, -0.3940, -0.4961, -0.0325, 0.3028, 0.1664],
  18772. [ 0.5938, -0.4122, 1.8738, -0.2503, -0.4744, -0.6442, 0.5637, 0.1794],
  18773. [ 0.7607, -0.2876, 1.7610, -0.0193, -0.2895, 0.1308, 0.7129, 0.3571],
  18774. [ 0.5551, -0.4227, 1.8155, 0.0922, -0.4076, -0.4884, 0.3522, 0.0836],
  18775. [ 0.4960, -0.4693, 1.6879, -0.8681, -0.4898, -0.7002, 0.5810, 0.1831],
  18776. [ 0.3679, -0.5129, 1.5779, -0.5601, -0.5443, -0.5261, 0.1086, 0.2616]],
  18777. device='cuda:0', grad_fn=<AddmmBackward>)
  18778. landmarks are: tensor([[[ 0.5776, -0.4109, 1.7326, -0.0226, -0.3633, 0.0236, 0.5605,
  18779. 0.2391],
  18780. [ 0.5764, -0.3969, 1.3284, -1.1312, -0.3460, -1.1389, 0.4797,
  18781. 0.3315],
  18782. [ 0.4974, -0.4482, 1.6633, -0.3306, -0.6173, 0.1313, 0.2925,
  18783. 0.0081],
  18784. [ 0.6372, -0.4129, 1.8942, -0.0765, -0.6173, -0.4768, 0.6999,
  18785. 0.0325],
  18786. [ 0.6421, -0.3816, 1.7037, 0.1929, -0.4037, 0.2391, 1.1861,
  18787. 0.2249],
  18788. [ 0.6335, -0.4162, 1.7499, 0.3084, -0.4961, -0.2459, 0.6524,
  18789. -0.0102],
  18790. [ 0.6059, -0.4249, 1.7420, -0.7321, -0.5931, -0.5523, 0.7307,
  18791. 0.1080],
  18792. [ 0.5367, -0.4294, 1.5709, -0.4999, -0.6693, -0.3075, 0.2455,
  18793. 0.3559]]], device='cuda:0')
  18794. loss_train_step before backward: tensor(0.0195, device='cuda:0', grad_fn=<MseLossBackward>)
  18795. loss_train_step after backward: tensor(0.0195, device='cuda:0', grad_fn=<MseLossBackward>)
  18796. loss_train: 2.6960538187995553
  18797. step: 79
  18798. running loss: 0.0341272635291083
  18799. Train Steps: 79/90 Loss: 0.0341 torch.Size([8, 600, 800])
  18800. torch.Size([8, 8])
  18801. tensor([[0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810],
  18802. [0.6203, 0.4096, 0.8862, 0.4267, 0.3538, 0.4117, 0.6025, 0.5650],
  18803. [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235],
  18804. [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500],
  18805. [0.6364, 0.4154, 0.8938, 0.3717, 0.4500, 0.2583, 0.6448, 0.5285],
  18806. [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459],
  18807. [0.6113, 0.4006, 0.8700, 0.5350, 0.3638, 0.3767, 0.5097, 0.4882],
  18808. [0.6198, 0.4164, 0.8700, 0.5067, 0.4625, 0.5650, 0.5464, 0.5197]],
  18809. device='cuda:0', dtype=torch.float64)
  18810. predictions are: tensor([[ 0.6406, -0.3394, 1.5624, -0.9658, -0.4175, -1.0238, 0.4895, 0.0445],
  18811. [ 0.8191, -0.2232, 1.7964, -0.2992, -0.5615, -0.4212, 0.4417, 0.3493],
  18812. [ 0.4210, -0.5028, 1.7112, -0.0776, -0.3531, 0.2189, 0.5235, 0.2024],
  18813. [ 0.3600, -0.5230, 1.3145, -1.0232, -0.6596, -0.3986, 0.5122, 0.2381],
  18814. [ 0.4282, -0.4474, 1.7761, -0.7220, -0.1502, -1.1927, 0.5608, 0.2853],
  18815. [ 0.2045, -0.6083, 1.6021, -1.0291, -0.0299, -1.2599, 0.7801, 0.2161],
  18816. [ 0.6250, -0.3410, 1.7121, 0.1525, -0.6350, -0.5269, 0.2029, 0.1414],
  18817. [ 0.5669, -0.3527, 1.7078, 0.0535, -0.1782, 0.1812, 0.2248, 0.2197]],
  18818. device='cuda:0', grad_fn=<AddmmBackward>)
  18819. landmarks are: tensor([[[ 0.6064, -0.3990, 1.6113, -0.8309, -0.4268, -1.0696, 0.6421,
  18820. -0.0640],
  18821. [ 0.5790, -0.3940, 1.8076, -0.3152, -0.6520, -0.3844, 0.4970,
  18822. 0.3238],
  18823. [ 0.5762, -0.4706, 1.7754, -0.0984, -0.3680, 0.2380, 0.6277,
  18824. 0.1322],
  18825. [ 0.5763, -0.4147, 1.3226, -1.0619, -0.6635, -0.4152, 0.5374,
  18826. 0.2545],
  18827. [ 0.6537, -0.3671, 1.8423, -0.5692, -0.2074, -1.0927, 0.6926,
  18828. 0.1554],
  18829. [ 0.6520, -0.3912, 1.7095, -0.9079, -0.0284, -1.3621, 0.8096,
  18830. 0.2356],
  18831. [ 0.5379, -0.4358, 1.7326, 0.1852, -0.6058, -0.5461, 0.0684,
  18832. -0.0310],
  18833. [ 0.5770, -0.3624, 1.7326, 0.0543, -0.1497, 0.3238, 0.2378,
  18834. 0.1146]]], device='cuda:0')
  18835. loss_train_step before backward: tensor(0.0123, device='cuda:0', grad_fn=<MseLossBackward>)
  18836. loss_train_step after backward: tensor(0.0123, device='cuda:0', grad_fn=<MseLossBackward>)
  18837. loss_train: 2.7083711810410023
  18838. step: 80
  18839. running loss: 0.03385463976301253
  18840. Train Steps: 80/90 Loss: 0.0339 torch.Size([8, 600, 800])
  18841. torch.Size([8, 8])
  18842. tensor([[0.6064, 0.4019, 0.8650, 0.4517, 0.4037, 0.5367, 0.5703, 0.5609],
  18843. [0.6284, 0.4093, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167],
  18844. [0.6162, 0.4134, 0.6700, 0.2467, 0.3962, 0.2533, 0.5737, 0.5467],
  18845. [0.6113, 0.4088, 0.6859, 0.2208, 0.4363, 0.1700, 0.5188, 0.5533],
  18846. [ nan, nan, 0.9088, 0.3783, 0.4562, 0.2617, 0.6741, 0.5575],
  18847. [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750],
  18848. [0.6124, 0.4075, 0.7696, 0.4153, 0.3475, 0.3767, 0.5157, 0.5427],
  18849. [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209]],
  18850. device='cuda:0', dtype=torch.float64)
  18851. predictions are: tensor([[ 0.7237, -0.3149, 1.8418, -0.0973, -0.3929, 0.4974, 0.4936, 0.1480],
  18852. [ 0.5533, -0.4386, 2.0288, -0.0327, -0.5607, -0.3373, 0.6171, 0.0117],
  18853. [ 0.6416, -0.3521, 1.0816, -1.1016, -0.4769, -1.0356, 0.4465, 0.2484],
  18854. [-0.0895, -0.8211, 1.0411, -1.2723, -0.3167, -1.3586, 0.1415, 0.1784],
  18855. [ 0.0629, -0.7368, 2.1032, -0.5214, -0.1178, -1.0612, 0.8426, 0.1643],
  18856. [ 0.6451, -0.3647, 1.2101, -1.0310, -0.4915, -0.8103, 0.5724, 0.2705],
  18857. [ 0.6988, -0.3193, 1.5759, -0.4323, -0.6370, -0.3351, 0.2304, 0.1933],
  18858. [ 0.6208, -0.3282, 1.7503, -0.0942, -0.2058, -0.9515, 0.4224, 0.3945]],
  18859. device='cuda:0', grad_fn=<AddmmBackward>)
  18860. landmarks are: tensor([[[ 0.5151, -0.4296, 1.7095, -0.1997, -0.4210, 0.1929, 0.3484,
  18861. 0.3047],
  18862. [ 0.6164, -0.3956, 1.8249, -0.1150, -0.6000, -0.5076, 0.5836,
  18863. 0.1005],
  18864. [ 0.5603, -0.3764, 0.8088, -1.1466, -0.4557, -1.1158, 0.3642,
  18865. 0.2391],
  18866. [ 0.5377, -0.3978, 0.8824, -1.2663, -0.2709, -1.5007, 0.1102,
  18867. 0.2699],
  18868. [-2.2859, -2.2859, 1.9115, -0.5384, -0.1785, -1.0773, 0.8278,
  18869. 0.2890],
  18870. [ 0.5805, -0.3818, 1.0282, -1.1774, -0.4903, -0.9310, 0.5894,
  18871. 0.3700],
  18872. [ 0.5427, -0.4035, 1.2688, -0.3675, -0.6808, -0.5461, 0.0959,
  18873. 0.2206],
  18874. [ 0.6174, -0.3118, 1.6402, -0.2074, -0.1958, -1.0927, 0.2267,
  18875. 0.5822]]], device='cuda:0')
  18876. loss_train_step before backward: tensor(0.1471, device='cuda:0', grad_fn=<MseLossBackward>)
  18877. loss_train_step after backward: tensor(0.1471, device='cuda:0', grad_fn=<MseLossBackward>)
  18878. loss_train: 2.855488333851099
  18879. step: 81
  18880. running loss: 0.03525294239322344
  18881. Train Steps: 81/90 Loss: 0.0353 torch.Size([8, 600, 800])
  18882. torch.Size([8, 8])
  18883. tensor([[ nan, nan, 0.6688, 0.2513, 0.4113, 0.2117, 0.5193, 0.5933],
  18884. [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383],
  18885. [0.6202, 0.4064, 0.7879, 0.2179, 0.4567, 0.1725, 0.5955, 0.5478],
  18886. [0.6268, 0.4029, 0.8500, 0.2683, 0.3937, 0.3500, 0.6860, 0.5297],
  18887. [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690],
  18888. [0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283],
  18889. [0.6124, 0.4069, 0.8314, 0.5001, 0.3738, 0.4650, 0.5167, 0.5402],
  18890. [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301]],
  18891. device='cuda:0', dtype=torch.float64)
  18892. predictions are: tensor([[-1.1345, -1.4592, 0.8842, -1.3519, -0.4274, -1.4106, 0.0831, 0.2845],
  18893. [ 0.6571, -0.3376, 1.7894, 0.0479, -0.3040, -0.0971, 0.2572, 0.1928],
  18894. [ 0.6914, -0.2991, 1.3371, -1.1903, -0.2251, -1.5271, 0.3750, 0.1489],
  18895. [ 0.8150, -0.2614, 1.6584, -0.8945, -0.5536, -0.6788, 0.8310, 0.1009],
  18896. [ 0.7828, -0.2719, 1.6696, 0.2097, -0.3416, 0.3041, 0.8104, 0.3228],
  18897. [ 0.4134, -0.5005, 1.9845, -0.5028, -0.5451, -0.6079, 0.6033, 0.1341],
  18898. [ 0.7405, -0.2666, 1.6761, 0.0836, -0.4790, -0.1078, 0.1919, 0.2615],
  18899. [ 0.7103, -0.3291, 1.7885, 0.0517, -0.3564, -0.0208, 0.6138, 0.1718]],
  18900. device='cuda:0', grad_fn=<AddmmBackward>)
  18901. landmarks are: tensor([[[-2.2859e+00, -2.2859e+00, 8.0331e-01, -1.1250e+00, -3.8637e-01,
  18902. -1.3082e+00, 1.1262e-01, 4.5430e-01],
  18903. [ 5.3603e-01, -4.6490e-01, 1.6517e+00, 4.6189e-04, -2.8245e-01,
  18904. -6.8822e-02, 2.3086e-01, 2.0046e-01],
  18905. [ 5.7870e-01, -4.0862e-01, 1.3535e+00, -1.2794e+00, -1.7640e-01,
  18906. -1.4891e+00, 4.6447e-01, 2.4425e-01],
  18907. [ 6.0919e-01, -4.2490e-01, 1.6402e+00, -1.0465e+00, -4.6721e-01,
  18908. -6.6928e-01, 8.8267e-01, 1.6077e-01],
  18909. [ 6.3060e-01, -4.1527e-01, 1.5141e+00, 2.2241e-01, -3.6905e-01,
  18910. 2.6220e-01, 1.0033e+00, 3.4245e-01],
  18911. [ 6.0733e-01, -4.0577e-01, 1.8885e+00, -4.9992e-01, -5.9423e-01,
  18912. -4.7683e-01, 6.4134e-01, 1.5443e-01],
  18913. [ 5.4249e-01, -4.0670e-01, 1.5543e+00, 2.4057e-02, -5.5958e-01,
  18914. -1.3811e-01, 1.0049e-01, 2.0932e-01],
  18915. [ 6.0095e-01, -4.5619e-01, 1.7198e+00, -9.0441e-03, -3.4644e-01,
  18916. 1.0758e-02, 6.2944e-01, 1.6266e-01]]], device='cuda:0')
  18917. loss_train_step before backward: tensor(0.0413, device='cuda:0', grad_fn=<MseLossBackward>)
  18918. loss_train_step after backward: tensor(0.0413, device='cuda:0', grad_fn=<MseLossBackward>)
  18919. loss_train: 2.8967771902680397
  18920. step: 82
  18921. running loss: 0.03532655110082975
  18922.  
  18923. Train Steps: 82/90 Loss: 0.0353 torch.Size([8, 600, 800])
  18924. torch.Size([8, 8])
  18925. tensor([[0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092],
  18926. [0.6145, 0.4007, 0.8775, 0.4533, 0.4562, 0.5533, 0.6088, 0.5533],
  18927. [0.6182, 0.3998, 0.8793, 0.4191, 0.3552, 0.4285, 0.6038, 0.5312],
  18928. [0.6230, 0.4152, 0.7588, 0.2283, 0.4012, 0.2883, 0.6200, 0.5767],
  18929. [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5988, 0.5667],
  18930. [0.6109, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567],
  18931. [0.6350, 0.4043, 0.8738, 0.5650, 0.3850, 0.4750, 0.6401, 0.4950],
  18932. [ nan, nan, 0.6469, 0.1943, 0.4025, 0.2000, 0.5125, 0.5533]],
  18933. device='cuda:0', dtype=torch.float64)
  18934. predictions are: tensor([[ 0.7146, -0.3538, 1.8599, 0.1712, -0.4480, 0.1083, 0.5213, 0.0950],
  18935. [ 0.6914, -0.3237, 1.8138, -0.0676, -0.1364, 0.2315, 0.5159, 0.2180],
  18936. [ 0.7692, -0.2724, 1.7680, -0.3090, -0.5663, -0.4063, 0.5210, 0.2294],
  18937. [ 0.6687, -0.3124, 1.3652, -0.8803, -0.4324, -1.0893, 0.5859, 0.3320],
  18938. [ 0.5855, -0.3619, 1.7856, -0.5942, -0.4860, -0.7473, 0.4492, 0.2532],
  18939. [ 0.6554, -0.3291, 1.1327, -1.2360, -0.4624, -1.1226, 0.3796, 0.2043],
  18940. [ 0.8600, -0.2718, 1.8370, 0.4075, -0.4948, -0.0732, 0.7471, 0.0575],
  18941. [-1.5150, -1.7410, 0.9122, -1.3620, -0.3742, -1.4723, 0.1188, 0.2358]],
  18942. device='cuda:0', grad_fn=<AddmmBackward>)
  18943. landmarks are: tensor([[[ 5.7829e-01, -4.3634e-01, 1.7724e+00, 4.3211e-02, -5.1531e-01,
  18944. 8.7136e-02, 4.8401e-01, 6.6263e-02],
  18945. [ 5.5260e-01, -4.3510e-01, 1.7672e+00, -1.9199e-01, -1.7852e-01,
  18946. 2.6990e-01, 5.2587e-01, 2.6990e-01],
  18947. [ 5.6966e-01, -4.3934e-01, 1.7754e+00, -3.5028e-01, -6.4527e-01,
  18948. -3.0670e-01, 5.0278e-01, 1.6774e-01],
  18949. [ 5.9151e-01, -3.6821e-01, 1.2187e+00, -1.2313e+00, -4.3256e-01,
  18950. -9.5412e-01, 5.7783e-01, 3.7768e-01],
  18951. [ 5.7829e-01, -3.9330e-01, 1.6748e+00, -6.1540e-01, -5.7691e-01,
  18952. -6.4619e-01, 4.7968e-01, 3.3149e-01],
  18953. [ 5.3557e-01, -4.2171e-01, 1.0339e+00, -1.4776e+00, -5.0762e-01,
  18954. -1.1081e+00, 4.2194e-01, 2.8530e-01],
  18955. [ 6.4706e-01, -4.1832e-01, 1.7499e+00, 3.2379e-01, -5.0762e-01,
  18956. -9.1917e-02, 6.7064e-01, 4.6189e-04],
  18957. [-2.2859e+00, -2.2859e+00, 7.0230e-01, -1.3883e+00, -4.2679e-01,
  18958. -1.3621e+00, 8.1293e-02, 2.6990e-01]]], device='cuda:0')
  18959. loss_train_step before backward: tensor(0.0244, device='cuda:0', grad_fn=<MseLossBackward>)
  18960. loss_train_step after backward: tensor(0.0244, device='cuda:0', grad_fn=<MseLossBackward>)
  18961. loss_train: 2.921171437948942
  18962. step: 83
  18963. running loss: 0.03519483660179448
  18964. Train Steps: 83/90 Loss: 0.0352 torch.Size([8, 600, 800])
  18965. torch.Size([8, 8])
  18966. tensor([[0.6260, 0.4120, 0.8013, 0.2350, 0.4888, 0.1533, 0.6281, 0.4895],
  18967. [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100],
  18968. [0.6336, 0.4154, 0.8900, 0.2767, 0.4988, 0.2867, 0.7422, 0.5540],
  18969. [0.6219, 0.4114, 0.8175, 0.2817, 0.3925, 0.2783, 0.5900, 0.5350],
  18970. [0.6185, 0.4042, 0.7700, 0.2250, 0.4062, 0.2117, 0.5763, 0.5150],
  18971. [0.6245, 0.4115, 0.8700, 0.4883, 0.4625, 0.5517, 0.6100, 0.5217],
  18972. [0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317],
  18973. [0.6224, 0.4061, 0.8988, 0.4300, 0.3838, 0.4750, 0.6112, 0.5483]],
  18974. device='cuda:0', dtype=torch.float64)
  18975. predictions are: tensor([[ 0.3414, -0.5336, 1.3901, -1.0815, -0.1304, -1.4158, 0.6134, 0.1596],
  18976. [ 0.5834, -0.4411, 1.6709, 0.4052, -0.5984, 0.0808, 0.5935, 0.1417],
  18977. [ 0.3512, -0.5651, 1.7264, -0.7843, -0.0953, -1.0059, 0.9068, 0.1935],
  18978. [ 0.4498, -0.4207, 1.3572, -0.9040, -0.5169, -0.8613, 0.3319, 0.2744],
  18979. [ 0.3137, -0.5197, 1.1709, -1.0104, -0.5328, -1.2113, 0.2474, 0.1146],
  18980. [ 0.5019, -0.4251, 1.6912, 0.0891, -0.2694, 0.3256, 0.5324, 0.2705],
  18981. [ 0.3252, -0.5454, 1.1839, -1.0608, -0.4705, -1.1051, 0.3983, 0.2607],
  18982. [ 0.4924, -0.4734, 1.8682, -0.0788, -0.6548, 0.1146, 0.5234, 0.1737]],
  18983. device='cuda:0', grad_fn=<AddmmBackward>)
  18984. landmarks are: tensor([[[ 0.6053, -0.3830, 1.4150, -1.2005, -0.0284, -1.5777, 0.6154,
  18985. -0.0250],
  18986. [ 0.6125, -0.4273, 1.6402, 0.2006, -0.4788, -0.0303, 0.6182,
  18987. 0.0697],
  18988. [ 0.6405, -0.3671, 1.8249, -1.0080, 0.0178, -0.9618, 1.1422,
  18989. 0.2730],
  18990. [ 0.5868, -0.3858, 1.4901, -0.9849, -0.4730, -1.0003, 0.4393,
  18991. 0.1852],
  18992. [ 0.5707, -0.4189, 1.2707, -1.2467, -0.4095, -1.3082, 0.3758,
  18993. 0.0928],
  18994. [ 0.5987, -0.3852, 1.7326, -0.0303, -0.1497, 0.2622, 0.5316,
  18995. 0.1236],
  18996. [ 0.5849, -0.3836, 1.2649, -1.2236, -0.3460, -1.2313, 0.4508,
  18997. 0.1698],
  18998. [ 0.5888, -0.4101, 1.8654, -0.2998, -0.5134, -0.0919, 0.5374,
  18999. 0.2468]]], device='cuda:0')
  19000. loss_train_step before backward: tensor(0.0179, device='cuda:0', grad_fn=<MseLossBackward>)
  19001. loss_train_step after backward: tensor(0.0179, device='cuda:0', grad_fn=<MseLossBackward>)
  19002. loss_train: 2.9391067922115326
  19003. step: 84
  19004. running loss: 0.034989366573946815
  19005. Train Steps: 84/90 Loss: 0.0350 torch.Size([8, 600, 800])
  19006. torch.Size([8, 8])
  19007. tensor([[0.6212, 0.4033, 0.8938, 0.4167, 0.3813, 0.4267, 0.5613, 0.5583],
  19008. [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583],
  19009. [0.6357, 0.4097, 0.9038, 0.3883, 0.4213, 0.2950, 0.6686, 0.5390],
  19010. [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483],
  19011. [0.6212, 0.4159, 0.8675, 0.5783, 0.4088, 0.4317, 0.5613, 0.5917],
  19012. [0.6250, 0.3993, 0.9138, 0.4333, 0.3763, 0.5217, 0.6995, 0.5320],
  19013. [0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204],
  19014. [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285]],
  19015. device='cuda:0', dtype=torch.float64)
  19016. predictions are: tensor([[ 0.3572, -0.5497, 1.8215, -0.4146, -0.5355, -0.3716, 0.3348, 0.2300],
  19017. [ 0.5134, -0.4433, 1.6347, -0.5314, -0.6411, -0.5109, 0.4675, 0.2653],
  19018. [ 0.3472, -0.5754, 1.7767, -0.6031, -0.3520, -1.0399, 0.7266, 0.1543],
  19019. [ 0.5444, -0.4306, 1.1022, -1.4533, -0.3904, -1.3121, 0.5611, 0.1790],
  19020. [ 0.3579, -0.5079, 1.5207, 0.2874, -0.3740, -0.3332, 0.2718, 0.4807],
  19021. [ 0.6658, -0.3873, 1.7824, -0.3300, -0.4989, 0.1572, 0.8910, 0.1273],
  19022. [ 0.5774, -0.4470, 1.5978, 0.1259, -0.3104, 0.0424, 0.7856, 0.1582],
  19023. [ 0.4544, -0.4687, 1.3953, -0.5208, -0.5736, -0.3778, 0.0729, 0.1174]],
  19024. device='cuda:0', grad_fn=<AddmmBackward>)
  19025. landmarks are: tensor([[[ 0.5832, -0.4231, 1.8423, -0.3614, -0.5249, -0.3152, 0.3065,
  19026. 0.2930],
  19027. [ 0.5761, -0.3966, 1.6171, -0.4845, -0.6346, -0.4691, 0.4739,
  19028. 0.2930],
  19029. [ 0.6504, -0.3936, 1.8885, -0.4922, -0.3402, -0.9233, 0.8022,
  19030. 0.2035],
  19031. [ 0.5784, -0.4085, 1.0859, -1.3929, -0.4037, -1.1158, 0.5605,
  19032. 0.2468],
  19033. [ 0.5836, -0.3649, 1.7210, 0.3854, -0.3979, -0.2921, 0.3065,
  19034. 0.4470],
  19035. [ 0.6010, -0.4417, 1.9346, -0.2844, -0.5480, 0.1236, 0.9448,
  19036. 0.1715],
  19037. [ 0.6042, -0.4273, 1.7198, 0.2184, -0.3478, 0.1149, 0.8062,
  19038. 0.1176],
  19039. [ 0.5253, -0.4296, 1.3084, -0.3098, -0.6115, -0.2767, 0.0928,
  19040. 0.1552]]], device='cuda:0')
  19041. loss_train_step before backward: tensor(0.0091, device='cuda:0', grad_fn=<MseLossBackward>)
  19042. loss_train_step after backward: tensor(0.0091, device='cuda:0', grad_fn=<MseLossBackward>)
  19043. loss_train: 2.948201633989811
  19044. step: 85
  19045. running loss: 0.03468472510576248
  19046. Train Steps: 85/90 Loss: 0.0347 torch.Size([8, 600, 800])
  19047. torch.Size([8, 8])
  19048. tensor([[0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6138, 0.5450],
  19049. [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5637, 0.5633],
  19050. [0.6122, 0.4006, 0.8850, 0.4217, 0.4088, 0.5517, 0.6063, 0.5517],
  19051. [0.6234, 0.4179, 0.7825, 0.3450, 0.3813, 0.2867, 0.5675, 0.5617],
  19052. [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167],
  19053. [0.6192, 0.4128, 0.8513, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633],
  19054. [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750],
  19055. [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100]],
  19056. device='cuda:0', dtype=torch.float64)
  19057. predictions are: tensor([[ 0.4660, -0.5156, 1.7767, -0.2449, -0.3343, 0.1230, 0.6427, 0.1392],
  19058. [ 0.3637, -0.5195, 1.6345, -0.0874, -0.6553, -0.6107, 0.3418, 0.3022],
  19059. [ 0.4526, -0.5204, 1.7592, -0.2958, -0.4045, 0.2497, 0.6422, 0.1193],
  19060. [ 0.0573, -0.7052, 1.4025, -0.9575, -0.4882, -1.0913, 0.3891, 0.2999],
  19061. [ 0.4061, -0.4735, 1.1699, -0.8911, -0.0988, -1.4151, 0.2780, 0.3841],
  19062. [ 0.5552, -0.4399, 1.6502, 0.2949, -0.3972, 0.1041, 0.5352, 0.2250],
  19063. [ 0.5542, -0.4346, 1.0822, -1.1684, -0.5267, -1.0174, 0.6169, 0.2570],
  19064. [ 0.7242, -0.3797, 1.7948, -0.4427, -0.6509, -0.5013, 0.7786, 0.0073]],
  19065. device='cuda:0', grad_fn=<AddmmBackward>)
  19066. landmarks are: tensor([[[ 0.5960, -0.4102, 1.8018, -0.1612, -0.3344, 0.1159, 0.5490,
  19067. 0.2314],
  19068. [ 0.5902, -0.3493, 1.7961, -0.0072, -0.5942, -0.5615, 0.3180,
  19069. 0.3161],
  19070. [ 0.5417, -0.4355, 1.8018, -0.3383, -0.3979, 0.2622, 0.5143,
  19071. 0.2622],
  19072. [ 0.5935, -0.3558, 1.3284, -0.6924, -0.5249, -0.9618, 0.3353,
  19073. 0.3084],
  19074. [ 0.6161, -0.3075, 1.1678, -0.6500, 0.0813, -1.4006, 0.2545,
  19075. 0.5624],
  19076. [ 0.5742, -0.3792, 1.6460, 0.3084, -0.3460, 0.1467, 0.4162,
  19077. 0.3161],
  19078. [ 0.5805, -0.3818, 1.0282, -1.1774, -0.4903, -0.9310, 0.5894,
  19079. 0.3700],
  19080. [ 0.6220, -0.4436, 1.8711, -0.3691, -0.6173, -0.4999, 0.6702,
  19081. 0.0697]]], device='cuda:0')
  19082. loss_train_step before backward: tensor(0.0161, device='cuda:0', grad_fn=<MseLossBackward>)
  19083. loss_train_step after backward: tensor(0.0161, device='cuda:0', grad_fn=<MseLossBackward>)
  19084. loss_train: 2.964329108595848
  19085. step: 86
  19086. running loss: 0.034468943123207536
  19087.  
  19088. Train Steps: 86/90 Loss: 0.0345 torch.Size([8, 600, 800])
  19089. torch.Size([8, 8])
  19090. tensor([[0.6282, 0.4034, 0.7830, 0.2080, 0.4532, 0.2080, 0.6404, 0.5323],
  19091. [0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153],
  19092. [0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350],
  19093. [0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719],
  19094. [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967],
  19095. [0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585],
  19096. [0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250],
  19097. [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717]],
  19098. device='cuda:0', dtype=torch.float64)
  19099. predictions are: tensor([[ 0.5998, -0.3945, 1.2492, -1.0363, -0.3124, -1.2346, 0.6505, 0.1615],
  19100. [ 0.4399, -0.5304, 1.5987, 0.2052, -0.2732, 0.2001, 0.5628, 0.1825],
  19101. [ 0.4002, -0.4918, 1.2865, -1.0020, -0.2786, -1.2601, 0.4286, 0.2335],
  19102. [ 0.5183, -0.4353, 1.5123, -0.2794, -0.6870, -0.2431, 0.3356, 0.1993],
  19103. [ 0.1278, -0.6566, 1.4739, -1.0337, -0.1406, -1.3200, 0.6354, 0.1578],
  19104. [ 0.4852, -0.4609, 1.6576, -0.2034, -0.3438, 0.3765, 0.6157, 0.1554],
  19105. [-0.0316, -0.7990, 1.7789, -0.2751, -0.6377, -0.7240, 0.6715, 0.2198],
  19106. [ 0.5022, -0.4296, 1.0507, -1.0560, -0.6059, -0.9220, 0.4741, 0.3350]],
  19107. device='cuda:0', grad_fn=<AddmmBackward>)
  19108. landmarks are: tensor([[[ 6.1577e-01, -4.2249e-01, 1.3307e+00, -1.3253e+00, -1.9244e-01,
  19109. -1.3252e+00, 6.7213e-01, 1.7271e-01],
  19110. [ 5.5484e-01, -4.6823e-01, 1.7309e+00, 9.6578e-02, -1.3942e-01,
  19111. 1.7573e-01, 5.0451e-01, 9.4188e-02],
  19112. [ 5.7679e-01, -4.0308e-01, 1.3838e+00, -1.1527e+00, -2.1876e-01,
  19113. -1.4216e+00, 4.3790e-01, 1.8502e-01],
  19114. [ 5.3672e-01, -4.2941e-01, 1.5709e+00, -4.9992e-01, -6.6928e-01,
  19115. -3.0747e-01, 2.4546e-01, 3.5585e-01],
  19116. [ 6.0479e-01, -3.7229e-01, 1.6517e+00, -1.0773e+00, 4.6189e-04,
  19117. -1.5161e+00, 5.8660e-01, 8.0947e-03],
  19118. [ 5.1155e-01, -4.3318e-01, 1.7557e+00, -3.1517e-01, -2.5358e-01,
  19119. 3.9307e-01, 4.1387e-01, 2.9364e-01],
  19120. [ 6.1114e-01, -3.8276e-01, 1.8885e+00, -3.8445e-01, -5.6536e-01,
  19121. -8.0785e-01, 5.6628e-01, 1.3903e-01],
  19122. [ 5.7460e-01, -3.8822e-01, 1.1436e+00, -1.2005e+00, -4.9030e-01,
  19123. -1.0157e+00, 4.3926e-01, 3.5458e-01]]], device='cuda:0')
  19124. loss_train_step before backward: tensor(0.0250, device='cuda:0', grad_fn=<MseLossBackward>)
  19125. loss_train_step after backward: tensor(0.0250, device='cuda:0', grad_fn=<MseLossBackward>)
  19126. loss_train: 2.9893352556973696
  19127. step: 87
  19128. running loss: 0.03436017535284333
  19129. Train Steps: 87/90 Loss: 0.0344 torch.Size([8, 600, 800])
  19130. torch.Size([8, 8])
  19131. tensor([[0.6339, 0.4118, 0.7988, 0.5800, 0.3912, 0.4583, 0.7343, 0.5760],
  19132. [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773],
  19133. [0.6148, 0.3996, 0.8488, 0.3867, 0.3488, 0.4067, 0.5863, 0.5000],
  19134. [ nan, nan, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552],
  19135. [0.6222, 0.3957, 0.8838, 0.5017, 0.3937, 0.4600, 0.5900, 0.5017],
  19136. [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390],
  19137. [0.6346, 0.4086, 0.7938, 0.5500, 0.3962, 0.4867, 0.7343, 0.5702],
  19138. [0.6259, 0.4156, 0.8812, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960]],
  19139. device='cuda:0', dtype=torch.float64)
  19140. predictions are: tensor([[ 0.6628, -0.3783, 1.3547, 0.0378, -0.5396, -0.2219, 0.7363, 0.3387],
  19141. [ 0.9989, -0.0989, 1.0561, -1.0009, -0.3688, -1.1186, 0.2621, 0.3592],
  19142. [ 0.8307, -0.2395, 1.6155, -0.5370, -0.7173, -0.2707, 0.4068, 0.0687],
  19143. [-2.1859, -2.2229, 1.5529, -1.2666, 0.0458, -1.3234, 0.9324, 0.2525],
  19144. [ 0.4735, -0.4862, 1.6272, -0.1025, -0.5379, -0.1717, 0.3618, 0.0620],
  19145. [ 0.6799, -0.3520, 1.6324, -0.1560, -0.4966, 0.0338, 0.2962, 0.1715],
  19146. [ 0.6859, -0.3553, 1.3792, 0.0354, -0.5202, -0.0513, 0.8590, 0.3345],
  19147. [ 0.6793, -0.2859, 1.6432, -0.9845, -0.1492, -1.2886, 0.5025, 0.1337]],
  19148. device='cuda:0', grad_fn=<AddmmBackward>)
  19149. landmarks are: tensor([[[ 0.6421, -0.3840, 1.4035, 0.3931, -0.4788, -0.1689, 1.1057,
  19150. 0.3745],
  19151. [ 0.5841, -0.3574, 1.0859, -0.9541, -0.2824, -1.2851, 0.3460,
  19152. 0.3808],
  19153. [ 0.5538, -0.4400, 1.6344, -0.4999, -0.6751, -0.4075, 0.4219,
  19154. 0.0236],
  19155. [-2.2859, -2.2859, 1.7557, -1.1466, 0.0871, -1.0773, 1.1239,
  19156. 0.2783],
  19157. [ 0.5879, -0.4583, 1.7961, 0.0313, -0.4672, -0.1612, 0.4393,
  19158. 0.0313],
  19159. [ 0.5250, -0.4661, 1.7383, -0.0765, -0.4268, -0.0226, 0.2535,
  19160. 0.2035],
  19161. [ 0.6454, -0.3984, 1.3804, 0.2545, -0.4557, -0.0380, 1.1057,
  19162. 0.3478],
  19163. [ 0.6052, -0.3663, 1.7845, -0.8156, -0.0804, -1.4237, 0.5866,
  19164. 0.0051]]], device='cuda:0')
  19165. loss_train_step before backward: tensor(0.0192, device='cuda:0', grad_fn=<MseLossBackward>)
  19166. loss_train_step after backward: tensor(0.0192, device='cuda:0', grad_fn=<MseLossBackward>)
  19167. loss_train: 3.0085563641041517
  19168. step: 88
  19169. running loss: 0.034188140501183545
  19170. Train Steps: 88/90 Loss: 0.0342 torch.Size([8, 600, 800])
  19171. torch.Size([8, 8])
  19172. tensor([[ nan, nan, 0.7240, 0.2722, 0.3900, 0.2567, 0.5168, 0.5933],
  19173. [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150],
  19174. [0.6364, 0.4165, 0.9088, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297],
  19175. [0.6286, 0.4060, 0.9188, 0.4333, 0.3675, 0.4167, 0.7034, 0.5528],
  19176. [0.6075, 0.4007, 0.8275, 0.4917, 0.4050, 0.5100, 0.5167, 0.5280],
  19177. [0.6208, 0.4082, 0.8538, 0.3067, 0.3588, 0.3717, 0.6112, 0.5517],
  19178. [0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431],
  19179. [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309]],
  19180. device='cuda:0', dtype=torch.float64)
  19181. predictions are: tensor([[-1.8385, -1.9641, 0.9082, -1.2111, -0.4355, -1.1670, 0.1173, 0.2900],
  19182. [ 0.6356, -0.3790, 1.6797, -0.1061, -0.5099, -0.3461, 0.6313, 0.1793],
  19183. [ 0.5199, -0.3973, 1.6923, -0.4086, -0.3310, -0.9304, 0.6163, 0.2632],
  19184. [ 0.7638, -0.3175, 1.7645, -0.4099, -0.4495, -0.4600, 0.9688, 0.2919],
  19185. [ 0.7587, -0.2697, 1.4616, -0.1549, -0.3250, 0.0384, 0.2837, 0.1989],
  19186. [ 0.7750, -0.2476, 1.4770, -0.9351, -0.5789, -0.6013, 0.5758, 0.2528],
  19187. [ 0.6622, -0.3352, 1.5473, -0.1471, -0.1829, -0.0524, 0.2201, 0.1321],
  19188. [ 0.6658, -0.3254, 1.6760, -0.0658, -0.3696, -0.5965, 0.7424, 0.2032]],
  19189. device='cuda:0', grad_fn=<AddmmBackward>)
  19190. landmarks are: tensor([[[-2.2859, -2.2859, 1.0580, -1.0288, -0.4845, -1.1004, 0.1011,
  19191. 0.4543],
  19192. [ 0.6096, -0.4046, 1.8249, -0.0072, -0.6115, -0.3537, 0.6182,
  19193. 0.0928],
  19194. [ 0.6537, -0.3623, 1.9115, -0.2690, -0.4037, -0.8309, 0.6926,
  19195. 0.1608],
  19196. [ 0.6174, -0.4105, 1.9577, -0.2844, -0.5885, -0.3614, 0.9631,
  19197. 0.2676],
  19198. [ 0.5200, -0.4353, 1.5363, -0.0149, -0.4152, 0.0697, 0.1005,
  19199. 0.1530],
  19200. [ 0.5814, -0.4003, 1.6575, -0.8694, -0.6289, -0.5692, 0.5374,
  19201. 0.2622],
  19202. [ 0.5351, -0.4153, 1.7326, -0.0457, -0.2214, -0.0466, 0.0434,
  19203. 0.2228],
  19204. [ 0.6520, -0.3623, 1.8885, 0.0313, -0.5538, -0.5384, 0.6926,
  19205. 0.1661]]], device='cuda:0')
  19206. loss_train_step before backward: tensor(0.0178, device='cuda:0', grad_fn=<MseLossBackward>)
  19207. loss_train_step after backward: tensor(0.0178, device='cuda:0', grad_fn=<MseLossBackward>)
  19208. loss_train: 3.0263278856873512
  19209. step: 89
  19210. running loss: 0.034003684108846646
  19211. Train Steps: 89/90 Loss: 0.0340 torch.Size([8, 600, 800])
  19212. torch.Size([8, 8])
  19213. tensor([[0.6112, 0.4029, 0.8638, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567],
  19214. [0.6264, 0.3972, 0.8853, 0.4771, 0.3853, 0.4511, 0.6293, 0.5334],
  19215. [0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742],
  19216. [0.6245, 0.4100, 0.7762, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417],
  19217. [0.6228, 0.4119, 0.7938, 0.2233, 0.4674, 0.1773, 0.6188, 0.5433],
  19218. [0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433],
  19219. [0.6170, 0.4102, 0.7468, 0.3695, 0.3463, 0.3767, 0.5238, 0.5823],
  19220. [0.6361, 0.4076, 0.8862, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297]],
  19221. device='cuda:0', dtype=torch.float64)
  19222. predictions are: tensor([[ 0.2155, -0.6617, 1.7206, -0.2581, -0.0849, -0.0335, 0.5475, 0.2886],
  19223. [ 0.3063, -0.6256, 1.7132, -0.2126, -0.6088, -0.1912, 0.7480, 0.2040],
  19224. [ 0.4653, -0.5070, 1.7394, 0.1501, -0.6849, -0.4636, 0.5853, 0.0483],
  19225. [ 0.5087, -0.4337, 1.2280, -1.1637, -0.1142, -1.4992, 0.4559, 0.2600],
  19226. [ 0.6276, -0.3673, 1.3516, -1.2932, -0.1980, -1.4377, 0.6106, 0.2283],
  19227. [ 0.3026, -0.6008, 1.6122, -0.0452, -0.0811, -0.0767, 0.2343, 0.2728],
  19228. [ 0.5610, -0.3835, 1.2338, -0.8067, -0.7232, -0.4564, 0.3050, 0.3639],
  19229. [ 0.5379, -0.4936, 1.6835, 0.0435, -0.6293, -0.0777, 0.8585, 0.2135]],
  19230. device='cuda:0', grad_fn=<AddmmBackward>)
  19231. landmarks are: tensor([[[ 0.5371, -0.4252, 1.7037, -0.0688, -0.0342, 0.0620, 0.3758,
  19232. 0.2853],
  19233. [ 0.6075, -0.4514, 1.8032, -0.0822, -0.5061, -0.2023, 0.6208,
  19234. 0.1779],
  19235. [ 0.6113, -0.4057, 1.7499, 0.3007, -0.5885, -0.5384, 0.4513,
  19236. -0.0957],
  19237. [ 0.5985, -0.3921, 1.2995, -1.0927, 0.0062, -1.5854, 0.4277,
  19238. 0.2160],
  19239. [ 0.5908, -0.3832, 1.3804, -1.2543, -0.1270, -1.4671, 0.5721,
  19240. 0.2237],
  19241. [ 0.5295, -0.4373, 1.6553, 0.1011, 0.0380, 0.0671, 0.0813,
  19242. 0.2237],
  19243. [ 0.5639, -0.3911, 1.1634, -0.5794, -0.6866, -0.5461, 0.1334,
  19244. 0.4036],
  19245. [ 0.6520, -0.4032, 1.8076, 0.1852, -0.5711, -0.1381, 0.7876,
  19246. 0.1608]]], device='cuda:0')
  19247. loss_train_step before backward: tensor(0.0141, device='cuda:0', grad_fn=<MseLossBackward>)
  19248. loss_train_step after backward: tensor(0.0141, device='cuda:0', grad_fn=<MseLossBackward>)
  19249. loss_train: 3.0404702592641115
  19250. step: 90
  19251. running loss: 0.03378300288071235
  19252.  
  19253. Valid Steps: 10/10 Loss: nan 7.2476
  19254. --------------------------------------------------
  19255. Epoch: 5 Train Loss: 0.0338 Valid Loss: nan
  19256. --------------------------------------------------
  19257. size of train loader is: 90
  19258. torch.Size([8, 600, 800])
  19259. torch.Size([8, 8])
  19260. tensor([[0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220],
  19261. [0.6364, 0.4165, 0.9088, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297],
  19262. [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329],
  19263. [ nan, nan, 0.7192, 0.2346, 0.4037, 0.2050, 0.5138, 0.5650],
  19264. [0.6115, 0.4081, 0.6725, 0.2433, 0.4088, 0.1933, 0.5167, 0.5544],
  19265. [0.6293, 0.4024, 0.8750, 0.5000, 0.4012, 0.5733, 0.7121, 0.5633],
  19266. [0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166],
  19267. [0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6138, 0.5450]],
  19268. device='cuda:0', dtype=torch.float64)
  19269. predictions are: tensor([[ 0.7542, -0.2940, 1.0693, -1.0823, -0.2846, -1.4095, 0.2587, 0.1636],
  19270. [ 0.4289, -0.4749, 1.8449, -0.2291, -0.3635, -0.8582, 0.6296, 0.2635],
  19271. [ 0.7719, -0.2729, 1.6264, -0.6635, -0.6348, -0.6002, 0.4818, 0.1923],
  19272. [-2.1929, -2.2416, 1.0535, -1.1674, -0.3955, -1.2469, 0.2622, 0.2562],
  19273. [ 0.8213, -0.2376, 0.9933, -1.0158, -0.3735, -1.2595, 0.0756, 0.2700],
  19274. [ 0.7109, -0.3502, 1.7879, 0.0807, -0.3919, 0.4202, 0.9466, 0.2984],
  19275. [ 0.7260, -0.3469, 1.6753, 0.1680, -0.3162, 0.1486, 0.9122, 0.1825],
  19276. [ 0.6985, -0.3393, 1.7840, -0.1376, -0.3363, 0.1751, 0.5855, 0.2307]],
  19277. device='cuda:0', grad_fn=<AddmmBackward>)
  19278. landmarks are: tensor([[[ 0.5566, -0.4160, 0.9935, -1.3313, -0.2824, -1.5161, 0.2144,
  19279. 0.1253],
  19280. [ 0.6537, -0.3623, 1.9115, -0.2690, -0.4037, -0.8309, 0.6926,
  19281. 0.1608],
  19282. [ 0.5680, -0.4393, 1.5920, -0.6672, -0.6453, -0.5457, 0.5149,
  19283. 0.1753],
  19284. [-2.2859, -2.2859, 1.0361, -1.2021, -0.4210, -1.3390, 0.0871,
  19285. 0.3238],
  19286. [ 0.5385, -0.4010, 0.8203, -1.1620, -0.3979, -1.3929, 0.1005,
  19287. 0.2747],
  19288. [ 0.6207, -0.4273, 1.7557, 0.0236, -0.4326, 0.3623, 1.0033,
  19289. 0.3157],
  19290. [ 0.6042, -0.4273, 1.6920, 0.1860, -0.2717, 0.1406, 0.7997,
  19291. 0.1004],
  19292. [ 0.5960, -0.4102, 1.8018, -0.1612, -0.3344, 0.1159, 0.5490,
  19293. 0.2314]]], device='cuda:0')
  19294. loss_train_step before backward: tensor(0.0097, device='cuda:0', grad_fn=<MseLossBackward>)
  19295. loss_train_step after backward: tensor(0.0097, device='cuda:0', grad_fn=<MseLossBackward>)
  19296. loss_train: 0.009727518074214458
  19297. step: 1
  19298. running loss: 0.009727518074214458
  19299. Train Steps: 1/90 Loss: 0.0097 torch.Size([8, 600, 800])
  19300. torch.Size([8, 8])
  19301. tensor([[0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118],
  19302. [0.6185, 0.4067, 0.8838, 0.4450, 0.4037, 0.4733, 0.5213, 0.5142],
  19303. [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817],
  19304. [0.6199, 0.4093, 0.7913, 0.2533, 0.4288, 0.2467, 0.5975, 0.5700],
  19305. [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6088, 0.5583],
  19306. [0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343],
  19307. [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5413, 0.5433],
  19308. [0.6289, 0.4032, 0.8419, 0.5446, 0.4075, 0.5017, 0.6312, 0.5117]],
  19309. device='cuda:0', dtype=torch.float64)
  19310. predictions are: tensor([[ 0.6004, -0.3672, 1.2156, -0.6897, -0.3006, -1.1231, 0.3223, 0.4997],
  19311. [ 0.3149, -0.5709, 1.8053, -0.1810, -0.4348, -0.0823, 0.1140, 0.1113],
  19312. [ 0.3440, -0.6193, 1.6411, 0.4132, -0.2512, 0.0696, 0.2662, 0.1777],
  19313. [ 0.6088, -0.4192, 1.4295, -1.1601, -0.3286, -1.0138, 0.7049, 0.2644],
  19314. [ 0.5523, -0.4556, 1.0715, -1.3078, -0.4542, -1.0982, 0.6409, 0.2777],
  19315. [ 0.4007, -0.5953, 1.8373, -0.6258, -0.2968, -0.8168, 1.0265, 0.0947],
  19316. [ 0.1804, -0.6489, 1.6171, -0.5607, -0.6342, -0.7571, 0.2685, 0.2146],
  19317. [ 0.4799, -0.5429, 1.6780, 0.1912, -0.4269, 0.0378, 0.8408, 0.1385]],
  19318. device='cuda:0', grad_fn=<AddmmBackward>)
  19319. landmarks are: tensor([[[ 0.6077, -0.3226, 0.9993, -0.6462, -0.2651, -1.3082, 0.2946,
  19320. 0.5401],
  19321. [ 0.5708, -0.4075, 1.7961, -0.2305, -0.4210, -0.0996, 0.1219,
  19322. 0.0893],
  19323. [ 0.5193, -0.4699, 1.5767, 0.4008, -0.2420, 0.0774, 0.1178,
  19324. -0.0610],
  19325. [ 0.5771, -0.3955, 1.3688, -1.1158, -0.3055, -1.1466, 0.4739,
  19326. 0.3469],
  19327. [ 0.5992, -0.3968, 0.9307, -1.3497, -0.4730, -1.0465, 0.5259,
  19328. 0.2930],
  19329. [ 0.6125, -0.4369, 1.9173, -0.5384, -0.2594, -0.8386, 0.9741,
  19330. 0.1821],
  19331. [ 0.5894, -0.3479, 1.7730, -0.6847, -0.5538, -0.7155, 0.2141,
  19332. 0.2237],
  19333. [ 0.6189, -0.4238, 1.6026, 0.2295, -0.4037, 0.0313, 0.6298,
  19334. 0.0774]]], device='cuda:0')
  19335. loss_train_step before backward: tensor(0.0148, device='cuda:0', grad_fn=<MseLossBackward>)
  19336. loss_train_step after backward: tensor(0.0148, device='cuda:0', grad_fn=<MseLossBackward>)
  19337. loss_train: 0.024488049559295177
  19338. step: 2
  19339. running loss: 0.012244024779647589
  19340. Train Steps: 2/90 Loss: 0.0122 torch.Size([8, 600, 800])
  19341. torch.Size([8, 8])
  19342. tensor([[0.6250, 0.4054, 0.8770, 0.4723, 0.4662, 0.5367, 0.6162, 0.5433],
  19343. [0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689],
  19344. [0.6201, 0.4064, 0.8688, 0.5050, 0.4225, 0.5100, 0.6138, 0.5500],
  19345. [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366],
  19346. [0.6310, 0.4017, 0.8563, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006],
  19347. [ nan, nan, 0.6412, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700],
  19348. [0.6200, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148],
  19349. [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967]],
  19350. device='cuda:0', dtype=torch.float64)
  19351. predictions are: tensor([[ 0.7454, -0.3024, 1.7860, -0.3361, -0.0949, 0.0958, 0.5557, 0.2262],
  19352. [ 0.5253, -0.4278, 1.4045, -0.8717, -0.5915, -0.6161, 0.2637, 0.3010],
  19353. [ 0.6854, -0.3350, 1.7206, -0.0855, -0.2740, -0.0146, 0.3494, 0.2215],
  19354. [ 0.7256, -0.3493, 1.4856, 0.0067, -0.3950, -0.1420, 0.8548, 0.2708],
  19355. [ 0.8537, -0.2916, 1.6853, 0.2443, -0.5348, -0.2137, 0.6321, 0.1528],
  19356. [-1.9638, -2.1107, 1.0300, -1.3369, -0.1977, -1.5700, 0.3303, 0.3528],
  19357. [ 0.7093, -0.3309, 1.8327, -0.2651, -0.5227, -0.4226, 0.3786, 0.1303],
  19358. [ 0.8808, -0.2086, 1.8253, -0.0877, -0.4148, -0.8260, 0.6492, 0.1294]],
  19359. device='cuda:0', grad_fn=<AddmmBackward>)
  19360. landmarks are: tensor([[[ 0.6009, -0.4135, 1.7651, -0.1043, -0.1323, 0.1929, 0.5605,
  19361. 0.2237],
  19362. [ 0.5327, -0.4381, 1.2880, -0.6308, -0.6866, -0.5307, 0.2658,
  19363. 0.3417],
  19364. [ 0.5784, -0.4088, 1.7268, 0.0467, -0.3344, 0.0697, 0.5490,
  19365. 0.2545],
  19366. [ 0.6289, -0.4345, 1.3794, 0.3679, -0.4845, 0.0390, 0.9265,
  19367. 0.1928],
  19368. [ 0.6286, -0.4303, 1.6691, 0.3931, -0.5827, -0.0919, 0.6359,
  19369. 0.0261],
  19370. [-2.2859, -2.2859, 0.6760, -1.4083, -0.3286, -1.4160, 0.2487,
  19371. 0.3469],
  19372. [ 0.5781, -0.4205, 1.8159, -0.0691, -0.6116, -0.3069, 0.4236,
  19373. 0.0919],
  19374. [ 0.6054, -0.3767, 1.7788, 0.0774, -0.5711, -0.7694, 0.5887,
  19375. 0.0081]]], device='cuda:0')
  19376. loss_train_step before backward: tensor(0.0186, device='cuda:0', grad_fn=<MseLossBackward>)
  19377. loss_train_step after backward: tensor(0.0186, device='cuda:0', grad_fn=<MseLossBackward>)
  19378. loss_train: 0.043093412183225155
  19379. step: 3
  19380. running loss: 0.014364470727741718
  19381. Train Steps: 3/90 Loss: 0.0144 torch.Size([8, 600, 800])
  19382. torch.Size([8, 8])
  19383. tensor([[0.6286, 0.4078, 0.8063, 0.2267, 0.4788, 0.1533, 0.5953, 0.4913],
  19384. [0.6204, 0.4013, 0.8075, 0.2400, 0.4313, 0.2050, 0.5800, 0.5150],
  19385. [0.6221, 0.4107, 0.7788, 0.3033, 0.3950, 0.2817, 0.6075, 0.5517],
  19386. [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5413, 0.5433],
  19387. [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683],
  19388. [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083],
  19389. [0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222],
  19390. [0.6268, 0.4029, 0.8500, 0.2683, 0.3937, 0.3500, 0.6860, 0.5297]],
  19391. device='cuda:0', dtype=torch.float64)
  19392. predictions are: tensor([[ 0.0854, -0.7300, 1.3343, -1.0758, -0.0698, -1.3501, 0.5538, 0.1221],
  19393. [ 0.4797, -0.4857, 1.3555, -1.0026, -0.2330, -1.1817, 0.5848, 0.1294],
  19394. [ 0.4533, -0.4955, 1.4321, -0.8290, -0.3719, -0.8532, 0.5331, 0.2373],
  19395. [ 0.3314, -0.5476, 1.5791, -0.4780, -0.6127, -0.5765, 0.2331, 0.1541],
  19396. [ 0.3811, -0.5397, 1.6226, 0.3160, -0.5185, -0.2334, 0.2644, 0.2589],
  19397. [ 0.2556, -0.6680, 1.6682, 0.1286, -0.5395, 0.0395, 0.4531, 0.3512],
  19398. [ 0.4959, -0.4210, 1.4818, -0.3671, -0.0721, -0.9667, 0.3686, 0.3975],
  19399. [ 0.6450, -0.4446, 1.6500, -0.9368, -0.5058, -0.4594, 1.1056, 0.0772]],
  19400. device='cuda:0', grad_fn=<AddmmBackward>)
  19401. landmarks are: tensor([[[ 0.6177, -0.4022, 1.4381, -1.2390, -0.0746, -1.5777, 0.4639,
  19402. -0.0168],
  19403. [ 0.5796, -0.4326, 1.4439, -1.1774, -0.2940, -1.3390, 0.3931,
  19404. 0.0928],
  19405. [ 0.5875, -0.3888, 1.3111, -0.8848, -0.4614, -0.9849, 0.5201,
  19406. 0.2622],
  19407. [ 0.5894, -0.3479, 1.7730, -0.6847, -0.5538, -0.7155, 0.2141,
  19408. 0.2237],
  19409. [ 0.5777, -0.3915, 1.7961, 0.1698, -0.5192, -0.5307, 0.2141,
  19410. 0.3392],
  19411. [ 0.5713, -0.4105, 1.7557, 0.0467, -0.6520, -0.2690, 0.3988,
  19412. 0.5239],
  19413. [ 0.6237, -0.2983, 1.3919, -0.4691, -0.0457, -1.2313, 0.2453,
  19414. 0.5882],
  19415. [ 0.6092, -0.4249, 1.6402, -1.0465, -0.4672, -0.6693, 0.8827,
  19416. 0.1608]]], device='cuda:0')
  19417. loss_train_step before backward: tensor(0.0271, device='cuda:0', grad_fn=<MseLossBackward>)
  19418. loss_train_step after backward: tensor(0.0271, device='cuda:0', grad_fn=<MseLossBackward>)
  19419. loss_train: 0.07015109900385141
  19420. step: 4
  19421. running loss: 0.017537774750962853
  19422.  
  19423. Train Steps: 4/90 Loss: 0.0175 torch.Size([8, 600, 800])
  19424. torch.Size([8, 8])
  19425. tensor([[0.6083, 0.3957, 0.8638, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980],
  19426. [0.6148, 0.4076, 0.8666, 0.4820, 0.4138, 0.5067, 0.5250, 0.5767],
  19427. [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466],
  19428. [0.6189, 0.4029, 0.8375, 0.5767, 0.4745, 0.4829, 0.5551, 0.5598],
  19429. [ nan, nan, 0.8525, 0.2217, 0.5413, 0.2367, 0.7367, 0.5482],
  19430. [0.6085, 0.4008, 0.8588, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283],
  19431. [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609],
  19432. [0.6200, 0.4098, 0.8237, 0.2917, 0.4012, 0.2967, 0.6000, 0.5683]],
  19433. device='cuda:0', dtype=torch.float64)
  19434. predictions are: tensor([[ 0.6732, -0.3474, 1.6743, -0.0082, -0.4961, 0.0134, 0.2572, -0.0216],
  19435. [ 0.8126, -0.2493, 1.6378, 0.0115, -0.5362, 0.0172, 0.2641, 0.2438],
  19436. [ 0.7811, -0.2574, 1.5581, 0.0977, -0.4353, 0.0952, 0.1921, 0.1597],
  19437. [ 0.7708, -0.2707, 1.4945, 0.3949, -0.2718, -0.1890, 0.3506, 0.2391],
  19438. [-2.1231, -2.2121, 1.5529, -1.2340, 0.0463, -1.3679, 1.1119, 0.3009],
  19439. [ 0.6323, -0.3641, 1.6140, 0.1118, -0.1846, -0.1604, 0.1566, 0.1369],
  19440. [ 0.3799, -0.5494, 1.6646, -1.1212, 0.0340, -1.5158, 1.2984, 0.2480],
  19441. [ 0.9152, -0.1758, 1.4520, -1.0302, -0.5784, -0.9599, 0.4555, 0.2445]],
  19442. device='cuda:0', grad_fn=<AddmmBackward>)
  19443. landmarks are: tensor([[[ 5.2379e-01, -4.5797e-01, 1.7037e+00, 4.6189e-04, -2.7090e-01,
  19444. 6.2048e-02, 1.8356e-01, 1.4106e-02],
  19445. [ 5.5393e-01, -4.0308e-01, 1.7168e+00, -5.9398e-02, -3.7483e-01,
  19446. 5.4350e-02, 1.3903e-01, 3.7768e-01],
  19447. [ 5.1784e-01, -4.4796e-01, 1.6748e+00, 6.2048e-02, -2.7667e-01,
  19448. 2.0831e-01, 1.0666e-01, 2.3862e-01],
  19449. [ 5.7258e-01, -4.2487e-01, 1.5824e+00, 3.7768e-01, -9.4206e-02,
  19450. -5.5582e-02, 2.7815e-01, 2.9966e-01],
  19451. [-2.2859e+00, -2.2859e+00, 1.6517e+00, -1.2620e+00, 2.1409e-01,
  19452. -1.1928e+00, 1.1166e+00, 2.4627e-01],
  19453. [ 5.2448e-01, -4.3472e-01, 1.6806e+00, 1.1594e-01, 4.6468e-03,
  19454. 1.2940e-02, 1.0439e-01, 1.5443e-01],
  19455. [ 6.3718e-01, -3.5749e-01, 1.6633e+00, -1.2082e+00, 2.1986e-01,
  19456. -1.2467e+00, 1.1313e+00, 3.0505e-01],
  19457. [ 5.7783e-01, -3.9299e-01, 1.5189e+00, -9.3872e-01, -4.3256e-01,
  19458. -9.1563e-01, 4.8545e-01, 3.3918e-01]]], device='cuda:0')
  19459. loss_train_step before backward: tensor(0.0189, device='cuda:0', grad_fn=<MseLossBackward>)
  19460. loss_train_step after backward: tensor(0.0189, device='cuda:0', grad_fn=<MseLossBackward>)
  19461. loss_train: 0.08904801961034536
  19462. step: 5
  19463. running loss: 0.017809603922069074
  19464. Train Steps: 5/90 Loss: 0.0178 torch.Size([8, 600, 800])
  19465. torch.Size([8, 8])
  19466. tensor([[0.6163, 0.4114, 0.7650, 0.2017, 0.3763, 0.2867, 0.5631, 0.5071],
  19467. [0.6091, 0.3997, 0.8314, 0.4334, 0.3788, 0.4550, 0.5213, 0.5656],
  19468. [0.6128, 0.4084, 0.8738, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397],
  19469. [0.6353, 0.4128, 0.9138, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991],
  19470. [0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367],
  19471. [0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901],
  19472. [0.6148, 0.4076, 0.8666, 0.4820, 0.4138, 0.5067, 0.5250, 0.5767],
  19473. [0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131]],
  19474. device='cuda:0', dtype=torch.float64)
  19475. predictions are: tensor([[ 0.4679, -0.5249, 1.1432, -1.3886, -0.4626, -1.0470, 0.5615, 0.1004],
  19476. [ 0.6455, -0.3653, 1.5256, -0.2370, -0.4582, -0.1594, 0.2167, 0.2687],
  19477. [ 0.3983, -0.5051, 1.6557, 0.0228, -0.5898, -0.5597, 0.1420, 0.1631],
  19478. [-0.1206, -0.9222, 1.9298, -0.4981, 0.0381, -0.9257, 1.4428, 0.4271],
  19479. [ 0.5584, -0.3998, 1.7356, -0.2781, -0.2670, -1.1656, 0.4453, 0.2577],
  19480. [ 0.4390, -0.5099, 1.8071, -0.1181, -0.3375, -0.1904, 0.1749, 0.0565],
  19481. [ 0.6322, -0.3949, 1.6822, 0.0830, -0.2574, 0.1128, 0.3545, 0.3131],
  19482. [ 0.5678, -0.4407, 1.5844, -0.4062, -0.5713, -0.2275, 0.3795, 0.1258]],
  19483. device='cuda:0', grad_fn=<AddmmBackward>)
  19484. landmarks are: tensor([[[ 0.5606, -0.3856, 1.2476, -1.3544, -0.5480, -0.9618, 0.3152,
  19485. 0.0562],
  19486. [ 0.5275, -0.4396, 1.5543, -0.2841, -0.5365, -0.1843, 0.1221,
  19487. 0.3265],
  19488. [ 0.5443, -0.3994, 1.7499, -0.1227, -0.6173, -0.5769, 0.0049,
  19489. 0.2071],
  19490. [ 0.6487, -0.3792, 1.9346, -0.6539, -0.1208, -0.7848, 1.0143,
  19491. 0.4814],
  19492. [ 0.5900, -0.3932, 1.8307, -0.3921, -0.4268, -1.1851, 0.3758,
  19493. 0.1929],
  19494. [ 0.5663, -0.3955, 1.7788, -0.2382, -0.4037, -0.2690, 0.0828,
  19495. -0.0220],
  19496. [ 0.5539, -0.4031, 1.7168, -0.0594, -0.3748, 0.0543, 0.1390,
  19497. 0.3777],
  19498. [ 0.5581, -0.4586, 1.5586, -0.3747, -0.6792, -0.2391, 0.4455,
  19499. 0.0840]]], device='cuda:0')
  19500. loss_train_step before backward: tensor(0.0255, device='cuda:0', grad_fn=<MseLossBackward>)
  19501. loss_train_step after backward: tensor(0.0255, device='cuda:0', grad_fn=<MseLossBackward>)
  19502. loss_train: 0.11450012680143118
  19503. step: 6
  19504. running loss: 0.019083354466905195
  19505. Train Steps: 6/90 Loss: 0.0191 torch.Size([8, 600, 800])
  19506. torch.Size([8, 8])
  19507. tensor([[0.6200, 0.4059, 0.8700, 0.4900, 0.4163, 0.5000, 0.6162, 0.5467],
  19508. [0.6265, 0.4088, 0.8025, 0.1850, 0.4163, 0.2500, 0.6290, 0.4947],
  19509. [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290],
  19510. [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611],
  19511. [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817],
  19512. [ nan, nan, 0.7648, 0.2722, 0.3962, 0.2183, 0.5060, 0.5422],
  19513. [0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407],
  19514. [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103]],
  19515. device='cuda:0', dtype=torch.float64)
  19516. predictions are: tensor([[ 0.7570, -0.3084, 1.7813, 0.1175, -0.2729, 0.0054, 0.5814, 0.2380],
  19517. [ 0.6667, -0.3722, 1.4920, -1.2591, -0.1936, -1.2800, 0.9240, 0.1477],
  19518. [ 0.4875, -0.4911, 1.6052, -0.7537, -0.5829, -0.5058, 0.7371, 0.1907],
  19519. [ 0.8956, -0.2000, 1.7388, 0.0653, -0.1816, -0.1037, 0.2166, 0.2962],
  19520. [ 0.8230, -0.2782, 1.6046, 0.5730, -0.1689, -0.0237, 0.2071, 0.1582],
  19521. [-2.3906, -2.3562, 1.1195, -0.9826, -0.3667, -1.1408, 0.1601, 0.2370],
  19522. [ 0.6998, -0.2829, 1.6248, -0.1946, -0.5209, -0.8853, 0.0696, 0.2143],
  19523. [ 0.8486, -0.2305, 1.5716, -0.7333, -0.5215, -0.5948, 0.5211, 0.1786]],
  19524. device='cuda:0', grad_fn=<AddmmBackward>)
  19525. landmarks are: tensor([[[ 5.7760e-01, -4.1093e-01, 1.7326e+00, -2.2633e-02, -3.6328e-01,
  19526. 2.3557e-02, 5.6051e-01, 2.3911e-01],
  19527. [ 6.0785e-01, -3.9761e-01, 1.4208e+00, -1.4314e+00, -3.6328e-01,
  19528. -1.1312e+00, 6.1950e-01, -9.2270e-04],
  19529. [ 5.6966e-01, -4.5379e-01, 1.5308e+00, -8.7027e-01, -6.5720e-01,
  19530. -3.6388e-01, 5.7392e-01, 1.5759e-01],
  19531. [ 5.4908e-01, -4.2902e-01, 1.7788e+00, -1.0731e-01, -2.6513e-01,
  19532. -1.0731e-01, 2.5553e-01, 3.0567e-01],
  19533. [ 5.1928e-01, -4.6990e-01, 1.5767e+00, 4.0077e-01, -2.4203e-01,
  19534. 7.7444e-02, 1.1776e-01, -6.1038e-02],
  19535. [-2.2859e+00, -2.2859e+00, 1.2469e+00, -1.0288e+00, -4.5566e-01,
  19536. -1.2774e+00, 5.1142e-02, 2.1834e-01],
  19537. [ 5.4700e-01, -3.9515e-01, 1.6377e+00, -4.2531e-01, -6.2887e-01,
  19538. -8.0785e-01, 2.4925e-02, 2.1157e-01],
  19539. [ 5.8863e-01, -3.7837e-01, 1.4554e+00, -9.0793e-01, -6.5774e-01,
  19540. -4.8453e-01, 3.4395e-01, 7.1216e-02]]], device='cuda:0')
  19541. loss_train_step before backward: tensor(0.0186, device='cuda:0', grad_fn=<MseLossBackward>)
  19542. loss_train_step after backward: tensor(0.0186, device='cuda:0', grad_fn=<MseLossBackward>)
  19543. loss_train: 0.1331191835924983
  19544. step: 7
  19545. running loss: 0.019017026227499758
  19546. Train Steps: 7/90 Loss: 0.0190 torch.Size([8, 600, 800])
  19547. torch.Size([8, 8])
  19548. tensor([[0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6063, 0.6083],
  19549. [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383],
  19550. [0.6246, 0.4028, 0.8738, 0.4867, 0.4088, 0.5667, 0.6362, 0.5200],
  19551. [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834],
  19552. [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717],
  19553. [0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461],
  19554. [0.6204, 0.4055, 0.8438, 0.5733, 0.4574, 0.4801, 0.5487, 0.5617],
  19555. [0.6078, 0.4033, 0.8019, 0.3055, 0.3450, 0.4200, 0.6025, 0.5550]],
  19556. device='cuda:0', dtype=torch.float64)
  19557. predictions are: tensor([[ 0.7288, -0.3058, 1.7726, -0.3509, -0.5335, -0.3225, 0.3383, 0.4083],
  19558. [ 0.2176, -0.6818, 1.7063, -0.6477, -0.6027, -0.5390, 0.5127, 0.2334],
  19559. [ 0.6948, -0.3953, 1.9897, 0.2174, -0.3004, 0.3106, 0.7040, 0.1262],
  19560. [ 0.3446, -0.6356, 1.5774, -1.2607, 0.0750, -1.6404, 0.7996, 0.0959],
  19561. [ 0.3937, -0.5598, 0.9217, -0.9847, -0.4205, -1.2133, 0.1011, 0.2899],
  19562. [ 0.3524, -0.5586, 1.6655, -0.3946, -0.6244, -0.4767, 0.1380, 0.1607],
  19563. [ 0.5475, -0.4451, 1.7190, 0.6524, -0.0622, -0.1217, 0.2448, 0.2306],
  19564. [ 0.3456, -0.5822, 1.4855, -0.7606, -0.6151, -0.3893, 0.4202, 0.2709]],
  19565. device='cuda:0', grad_fn=<AddmmBackward>)
  19566. landmarks are: tensor([[[ 0.5774, -0.3868, 1.6286, -0.5692, -0.6462, -0.2767, 0.5143,
  19567. 0.5239],
  19568. [ 0.5779, -0.4275, 1.5894, -0.8362, -0.6577, -0.5153, 0.5605,
  19569. 0.2006],
  19570. [ 0.5990, -0.4256, 1.7499, -0.0380, -0.3979, 0.3315, 0.6529,
  19571. 0.1159],
  19572. [ 0.6126, -0.4057, 1.4439, -1.3159, -0.1150, -1.5777, 0.5537,
  19573. -0.0530],
  19574. [ 0.5633, -0.3883, 0.8648, -1.0288, -0.5192, -1.1081, 0.2776,
  19575. 0.3546],
  19576. [ 0.5124, -0.4483, 1.5420, -0.6231, -0.7155, -0.4152, 0.2679,
  19577. 0.2365],
  19578. [ 0.5795, -0.4129, 1.6113, 0.3623, -0.1733, -0.0684, 0.2487,
  19579. 0.3084],
  19580. [ 0.5215, -0.4232, 1.4182, -0.8747, -0.6924, -0.3460, 0.4970,
  19581. 0.2776]]], device='cuda:0')
  19582. loss_train_step before backward: tensor(0.0201, device='cuda:0', grad_fn=<MseLossBackward>)
  19583. loss_train_step after backward: tensor(0.0201, device='cuda:0', grad_fn=<MseLossBackward>)
  19584. loss_train: 0.1532347435131669
  19585. step: 8
  19586. running loss: 0.019154342939145863
  19587.  
  19588. Train Steps: 8/90 Loss: 0.0192 torch.Size([8, 600, 800])
  19589. torch.Size([8, 8])
  19590. tensor([[0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290],
  19591. [0.6222, 0.3937, 0.8350, 0.5617, 0.4138, 0.4600, 0.5800, 0.5233],
  19592. [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869],
  19593. [0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733],
  19594. [0.6212, 0.4033, 0.8938, 0.4167, 0.3813, 0.4267, 0.5613, 0.5583],
  19595. [0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117],
  19596. [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517],
  19597. [0.6053, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168]],
  19598. device='cuda:0', dtype=torch.float64)
  19599. predictions are: tensor([[ 0.2550, -0.6382, 1.3415, -1.3801, -0.2225, -1.3262, 0.6423, 0.1599],
  19600. [ 0.5471, -0.4761, 1.7444, 0.2648, -0.4119, -0.1947, 0.4265, 0.1811],
  19601. [ 0.6230, -0.3890, 1.8253, 0.3196, -0.4646, -0.2568, 0.0844, 0.0392],
  19602. [ 0.8241, -0.2153, 1.8669, -0.0251, -0.5363, -0.3031, 0.3125, 0.3552],
  19603. [ 0.9951, -0.1611, 1.9537, -0.3856, -0.5628, -0.2982, 0.3243, 0.1922],
  19604. [ 0.4725, -0.4909, 1.6059, 0.3220, -0.4420, -0.0804, 0.3932, 0.4512],
  19605. [ 0.4065, -0.5371, 1.8867, -0.1379, -0.1220, 0.2303, 0.4275, 0.2397],
  19606. [-0.3573, -1.0397, 0.9568, -1.4847, -0.3793, -1.4299, 0.2651, 0.1650]],
  19607. device='cuda:0', grad_fn=<AddmmBackward>)
  19608. landmarks are: tensor([[[ 0.6125, -0.4273, 1.2807, -1.3253, -0.2574, -1.2542, 0.6864,
  19609. 0.1575],
  19610. [ 0.5880, -0.4676, 1.5709, 0.3084, -0.3748, -0.1612, 0.3931,
  19611. 0.1313],
  19612. [ 0.5320, -0.4488, 1.6633, 0.3315, -0.5018, -0.2459, 0.0828,
  19613. -0.0370],
  19614. [ 0.6121, -0.3138, 1.8423, 0.0082, -0.6462, -0.3075, 0.3469,
  19615. 0.3623],
  19616. [ 0.5832, -0.4231, 1.8423, -0.3614, -0.5249, -0.3152, 0.3065,
  19617. 0.2930],
  19618. [ 0.6118, -0.3983, 1.5824, 0.3469, -0.4268, -0.0688, 0.3469,
  19619. 0.5393],
  19620. [ 0.5897, -0.3804, 1.7788, -0.0226, -0.2016, 0.3007, 0.6125,
  19621. 0.2622],
  19622. [ 0.5102, -0.4223, 0.8999, -1.4011, -0.4383, -1.3082, 0.2267,
  19623. 0.1013]]], device='cuda:0')
  19624. loss_train_step before backward: tensor(0.0303, device='cuda:0', grad_fn=<MseLossBackward>)
  19625. loss_train_step after backward: tensor(0.0303, device='cuda:0', grad_fn=<MseLossBackward>)
  19626. loss_train: 0.18358023930341005
  19627. step: 9
  19628. running loss: 0.020397804367045563
  19629. Train Steps: 9/90 Loss: 0.0204 torch.Size([8, 600, 800])
  19630. torch.Size([8, 8])
  19631. tensor([[0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617],
  19632. [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116],
  19633. [0.6224, 0.4097, 0.7438, 0.2267, 0.3850, 0.2850, 0.5988, 0.5250],
  19634. [0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272],
  19635. [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609],
  19636. [0.6179, 0.4118, 0.7278, 0.4237, 0.3588, 0.3400, 0.5675, 0.5917],
  19637. [0.6164, 0.4119, 0.7913, 0.2650, 0.3538, 0.3500, 0.5614, 0.5038],
  19638. [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167]],
  19639. device='cuda:0', dtype=torch.float64)
  19640. predictions are: tensor([[ 4.3023e-01, -5.1482e-01, 1.7756e+00, 1.1360e-03, -3.6645e-01,
  19641. -2.6895e-01, 8.9147e-02, 2.8362e-01],
  19642. [ 2.5134e-01, -6.3133e-01, 1.8180e+00, -1.0324e-01, -3.5429e-01,
  19643. 4.1739e-01, 4.8015e-01, 1.5867e-01],
  19644. [ 3.5384e-01, -5.8044e-01, 1.1428e+00, -1.0631e+00, -3.7179e-01,
  19645. -9.1526e-01, 4.6287e-01, 2.6820e-01],
  19646. [ 4.9021e-01, -4.6988e-01, 1.6632e+00, -4.7789e-01, -3.7840e-01,
  19647. -1.0013e+00, 1.3930e-01, 1.6179e-01],
  19648. [ 7.3768e-01, -3.5570e-01, 1.7729e+00, -6.4928e-01, -2.7519e-01,
  19649. -8.5562e-01, 8.7062e-01, 2.3698e-01],
  19650. [ 6.4180e-01, -3.7256e-01, 1.3457e+00, -3.2032e-01, -4.9100e-01,
  19651. -7.8489e-01, 2.0598e-01, 4.0648e-01],
  19652. [ 4.5137e-02, -7.7108e-01, 1.3785e+00, -9.7295e-01, -5.5496e-01,
  19653. -6.3843e-01, 3.7514e-01, 1.5874e-01],
  19654. [ 5.1822e-01, -4.5746e-01, 1.6131e+00, -6.3268e-01, -4.9638e-01,
  19655. -8.6872e-01, 2.3403e-01, 9.5013e-02]], device='cuda:0',
  19656. grad_fn=<AddmmBackward>)
  19657. landmarks are: tensor([[[ 0.5488, -0.4221, 1.8018, -0.2459, -0.4845, -0.3075, 0.2309,
  19658. 0.3087],
  19659. [ 0.5376, -0.3903, 1.7095, -0.3229, -0.4730, 0.4701, 0.3871,
  19660. 0.0772],
  19661. [ 0.5891, -0.3937, 1.1494, -1.2390, -0.5076, -0.9695, 0.4797,
  19662. 0.1390],
  19663. [ 0.5711, -0.4015, 1.6979, -0.6770, -0.5365, -1.0619, 0.1712,
  19664. 0.1494],
  19665. [ 0.6026, -0.4032, 1.7326, -0.7771, -0.3691, -0.8617, 0.9704,
  19666. 0.3050],
  19667. [ 0.5680, -0.3840, 1.0756, -0.3290, -0.6289, -0.7155, 0.3353,
  19668. 0.4470],
  19669. [ 0.5614, -0.3834, 1.3688, -1.0619, -0.6520, -0.6693, 0.3069,
  19670. 0.0412],
  19671. [ 0.5483, -0.4249, 1.5305, -0.7386, -0.6115, -0.8694, 0.3353,
  19672. 0.1005]]], device='cuda:0')
  19673. loss_train_step before backward: tensor(0.0199, device='cuda:0', grad_fn=<MseLossBackward>)
  19674. loss_train_step after backward: tensor(0.0199, device='cuda:0', grad_fn=<MseLossBackward>)
  19675. loss_train: 0.20348771009594202
  19676. step: 10
  19677. running loss: 0.020348771009594202
  19678. Train Steps: 10/90 Loss: 0.0203 torch.Size([8, 600, 800])
  19679. torch.Size([8, 8])
  19680. tensor([[0.6091, 0.3997, 0.8314, 0.4334, 0.3788, 0.4550, 0.5213, 0.5656],
  19681. [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773],
  19682. [0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750],
  19683. [0.6183, 0.4076, 0.8838, 0.4517, 0.3813, 0.4483, 0.5775, 0.5633],
  19684. [0.6162, 0.4134, 0.6700, 0.2467, 0.3962, 0.2533, 0.5737, 0.5467],
  19685. [0.6204, 0.4055, 0.8438, 0.5733, 0.4574, 0.4801, 0.5487, 0.5617],
  19686. [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261],
  19687. [0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835]],
  19688. device='cuda:0', dtype=torch.float64)
  19689. predictions are: tensor([[ 0.3729, -0.5449, 1.6055, -0.3498, -0.5716, -0.1778, 0.1901, 0.2834],
  19690. [ 0.4369, -0.5189, 1.7649, -0.1136, -0.0909, -0.0224, 0.1736, 0.2540],
  19691. [ 0.2000, -0.6297, 1.3461, -1.0150, -0.4848, -1.0933, 0.1059, 0.2289],
  19692. [ 0.4935, -0.4793, 1.7645, -0.2782, -0.5397, -0.1706, 0.3770, 0.3191],
  19693. [ 0.6598, -0.3805, 0.9234, -1.1801, -0.5330, -1.1614, 0.4945, 0.2989],
  19694. [ 0.4239, -0.5186, 1.6509, 0.3934, -0.2377, -0.0880, 0.3082, 0.2429],
  19695. [ 0.3168, -0.5442, 1.8375, -0.1244, -0.4559, -0.0795, 0.3196, 0.0846],
  19696. [ 0.8351, -0.2616, 1.8822, -0.5596, -0.4198, -1.2483, 0.5455, 0.0289]],
  19697. device='cuda:0', grad_fn=<AddmmBackward>)
  19698. landmarks are: tensor([[[ 0.5275, -0.4396, 1.5543, -0.2841, -0.5365, -0.1843, 0.1221,
  19699. 0.3265],
  19700. [ 0.5428, -0.4244, 1.7095, -0.0380, 0.0232, 0.0725, 0.0866,
  19701. 0.3806],
  19702. [ 0.5695, -0.3927, 1.3226, -0.9002, -0.4672, -1.1928, 0.1737,
  19703. 0.3700],
  19704. [ 0.5700, -0.4032, 1.7961, -0.1997, -0.5249, -0.2151, 0.3815,
  19705. 0.3161],
  19706. [ 0.5603, -0.3764, 0.8088, -1.1466, -0.4557, -1.1158, 0.3642,
  19707. 0.2391],
  19708. [ 0.5795, -0.4129, 1.6113, 0.3623, -0.1733, -0.0684, 0.2487,
  19709. 0.3084],
  19710. [ 0.5746, -0.3623, 1.7961, -0.1150, -0.3691, -0.0380, 0.2208,
  19711. 0.1439],
  19712. [ 0.6117, -0.3877, 1.8249, -0.5230, -0.3517, -1.2851, 0.4200,
  19713. -0.0529]]], device='cuda:0')
  19714. loss_train_step before backward: tensor(0.0111, device='cuda:0', grad_fn=<MseLossBackward>)
  19715. loss_train_step after backward: tensor(0.0111, device='cuda:0', grad_fn=<MseLossBackward>)
  19716. loss_train: 0.2146181659772992
  19717. step: 11
  19718. running loss: 0.019510742361572655
  19719. Train Steps: 11/90 Loss: 0.0195 torch.Size([8, 600, 800])
  19720. torch.Size([8, 8])
  19721. tensor([[0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6038, 0.4833],
  19722. [0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343],
  19723. [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575],
  19724. [0.6142, 0.4127, 0.7575, 0.3067, 0.3438, 0.4383, 0.5778, 0.5207],
  19725. [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600],
  19726. [0.6276, 0.4235, 0.8888, 0.5333, 0.3800, 0.3117, 0.5427, 0.6164],
  19727. [0.6102, 0.4005, 0.8688, 0.5100, 0.4813, 0.5400, 0.5404, 0.5064],
  19728. [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517]],
  19729. device='cuda:0', dtype=torch.float64)
  19730. predictions are: tensor([[ 0.4516, -0.5583, 1.8792, 0.0522, -0.5844, -0.0748, 0.2217, 0.0709],
  19731. [ 0.4405, -0.5722, 1.8521, -0.5826, -0.3625, -0.8696, 0.6606, 0.0918],
  19732. [ 0.5997, -0.4428, 1.7636, -0.7638, -0.4621, -0.3558, 0.7902, 0.2362],
  19733. [ 0.2645, -0.6072, 1.2612, -0.9835, -0.6613, -0.3469, 0.1612, 0.2814],
  19734. [ 0.4704, -0.4822, 1.2027, -1.1104, -0.6415, -0.7891, 0.2351, 0.2756],
  19735. [ 0.5423, -0.4085, 1.7679, 0.4169, -0.5631, -0.7001, 0.2094, 0.4311],
  19736. [ 0.4537, -0.5225, 1.8894, 0.1367, -0.0578, 0.1081, 0.0480, 0.0833],
  19737. [ 0.7690, -0.3127, 1.0238, -1.1468, -0.4148, -1.2598, 0.2532, 0.3526]],
  19738. device='cuda:0', grad_fn=<AddmmBackward>)
  19739. landmarks are: tensor([[[ 0.5726, -0.4795, 1.7788, -0.0149, -0.5711, -0.0688, 0.5028,
  19740. -0.0534],
  19741. [ 0.6125, -0.4369, 1.9173, -0.5384, -0.2594, -0.8386, 0.9741,
  19742. 0.1821],
  19743. [ 0.6158, -0.4249, 1.8654, -0.9002, -0.3229, -0.3537, 0.9667,
  19744. 0.2890],
  19745. [ 0.5509, -0.3798, 1.2129, -0.8694, -0.6982, -0.2613, 0.3830,
  19746. 0.1193],
  19747. [ 0.5466, -0.4080, 1.0668, -1.1764, -0.6289, -0.7617, 0.4855,
  19748. 0.3007],
  19749. [ 0.6131, -0.3299, 1.8192, 0.1775, -0.5307, -0.8463, 0.2206,
  19750. 0.5612],
  19751. [ 0.5328, -0.4361, 1.7268, 0.0697, -0.0630, 0.2083, 0.2103,
  19752. 0.0532],
  19753. [ 0.5713, -0.3671, 0.8665, -1.0696, -0.3691, -1.2236, 0.3527,
  19754. 0.2622]]], device='cuda:0')
  19755. loss_train_step before backward: tensor(0.0157, device='cuda:0', grad_fn=<MseLossBackward>)
  19756. loss_train_step after backward: tensor(0.0157, device='cuda:0', grad_fn=<MseLossBackward>)
  19757. loss_train: 0.23029645066708326
  19758. step: 12
  19759. running loss: 0.019191370888923604
  19760.  
  19761. Train Steps: 12/90 Loss: 0.0192 torch.Size([8, 600, 800])
  19762. torch.Size([8, 8])
  19763. tensor([[0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261],
  19764. [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267],
  19765. [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268],
  19766. [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5413, 0.5433],
  19767. [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617],
  19768. [0.6275, 0.4081, 0.8063, 0.2017, 0.4825, 0.1583, 0.6156, 0.4869],
  19769. [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285],
  19770. [0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461]],
  19771. device='cuda:0', dtype=torch.float64)
  19772. predictions are: tensor([[ 0.4027, -0.4660, 1.7366, -0.1386, -0.4564, -0.1423, 0.1591, 0.1444],
  19773. [ 0.6957, -0.3605, 1.5943, 0.3531, -0.5002, -0.2424, 0.2633, 0.1841],
  19774. [ 0.5300, -0.4568, 1.6870, -0.1113, -0.3307, -0.0143, 0.4898, 0.2449],
  19775. [ 0.6326, -0.3350, 1.5015, -0.6746, -0.6986, -0.8277, 0.0879, 0.2304],
  19776. [ 0.4498, -0.4974, 1.7203, -0.1162, -0.3503, -0.0235, 0.2616, 0.2893],
  19777. [ 0.7554, -0.3196, 1.3230, -1.4331, -0.1000, -1.5792, 0.5366, 0.1289],
  19778. [ 0.3027, -0.6293, 1.6558, -0.7107, -0.6150, -0.2197, 0.6912, 0.2634],
  19779. [ 0.4142, -0.4875, 1.4956, -0.6237, -0.7294, -0.4820, 0.0756, 0.2566]],
  19780. device='cuda:0', grad_fn=<AddmmBackward>)
  19781. landmarks are: tensor([[[ 0.5746, -0.3623, 1.7961, -0.1150, -0.3691, -0.0380, 0.2208,
  19782. 0.1439],
  19783. [ 0.5891, -0.4550, 1.5132, 0.3546, -0.3691, -0.1535, 0.3815,
  19784. 0.1467],
  19785. [ 0.5828, -0.4682, 1.7031, -0.0497, -0.2458, 0.0818, 0.6381,
  19786. 0.1474],
  19787. [ 0.5894, -0.3479, 1.7730, -0.6847, -0.5538, -0.7155, 0.2141,
  19788. 0.2237],
  19789. [ 0.5637, -0.4143, 1.7519, -0.0787, -0.3055, -0.0149, 0.3758,
  19790. 0.3084],
  19791. [ 0.6124, -0.4010, 1.4381, -1.3544, -0.0573, -1.5546, 0.5573,
  19792. -0.0369],
  19793. [ 0.6108, -0.4273, 1.8711, -0.6616, -0.5769, -0.1997, 0.9156,
  19794. 0.1554],
  19795. [ 0.5124, -0.4483, 1.5420, -0.6231, -0.7155, -0.4152, 0.2679,
  19796. 0.2365]]], device='cuda:0')
  19797. loss_train_step before backward: tensor(0.0114, device='cuda:0', grad_fn=<MseLossBackward>)
  19798. loss_train_step after backward: tensor(0.0114, device='cuda:0', grad_fn=<MseLossBackward>)
  19799. loss_train: 0.24174558836966753
  19800. step: 13
  19801. running loss: 0.018595814489974424
  19802. Train Steps: 13/90 Loss: 0.0186 torch.Size([8, 600, 800])
  19803. torch.Size([8, 8])
  19804. tensor([[0.6205, 0.4016, 0.8350, 0.2717, 0.3987, 0.2550, 0.5787, 0.5133],
  19805. [0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626],
  19806. [0.6154, 0.4112, 0.7037, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600],
  19807. [0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317],
  19808. [0.6147, 0.4081, 0.8538, 0.3400, 0.3663, 0.3150, 0.5142, 0.4875],
  19809. [0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124],
  19810. [0.6264, 0.3972, 0.8853, 0.4771, 0.3853, 0.4511, 0.6293, 0.5334],
  19811. [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400]],
  19812. device='cuda:0', dtype=torch.float64)
  19813. predictions are: tensor([[ 0.5512, -0.4012, 1.5784, -0.9252, -0.4582, -0.8089, 0.3276, 0.1340],
  19814. [ 0.4327, -0.5581, 1.7039, 0.5004, -0.5014, 0.0815, 0.4479, 0.0676],
  19815. [ 0.4325, -0.4672, 0.9178, -1.0695, -0.3671, -1.2160, 0.0401, 0.3845],
  19816. [ 0.4380, -0.4994, 1.1840, -1.0801, -0.3987, -0.9658, 0.3837, 0.2877],
  19817. [ 0.4125, -0.4639, 1.6765, -0.6351, -0.6692, -0.6472, 0.0339, 0.1250],
  19818. [ 0.5317, -0.4280, 1.4851, -0.9166, -0.4635, -0.7955, 0.3967, 0.1464],
  19819. [ 0.5392, -0.4849, 1.8711, 0.0163, -0.5591, -0.0264, 0.6641, 0.2175],
  19820. [ 0.4610, -0.4635, 1.2291, -1.0568, -0.2196, -1.1458, 0.3517, 0.3011]],
  19821. device='cuda:0', grad_fn=<AddmmBackward>)
  19822. landmarks are: tensor([[[ 0.5800, -0.4312, 1.5709, -1.0311, -0.4441, -1.1081, 0.3873,
  19823. 0.0851],
  19824. [ 0.6105, -0.4293, 1.5824, 0.5239, -0.4730, -0.0380, 0.5025,
  19825. -0.1492],
  19826. [ 0.5565, -0.3864, 0.9647, -1.2159, -0.3286, -1.4391, 0.1852,
  19827. 0.3007],
  19828. [ 0.5849, -0.3836, 1.2649, -1.2236, -0.3460, -1.2313, 0.4508,
  19829. 0.1698],
  19830. [ 0.5532, -0.4008, 1.6575, -0.7155, -0.5942, -0.8309, 0.0890,
  19831. -0.0340],
  19832. [ 0.5790, -0.4079, 1.5929, -1.0630, -0.4729, -1.0725, 0.4137,
  19833. 0.0807],
  19834. [ 0.6075, -0.4514, 1.8032, -0.0822, -0.5061, -0.2023, 0.6208,
  19835. 0.1779],
  19836. [ 0.5842, -0.3848, 1.2938, -1.2159, -0.2132, -1.4160, 0.5778,
  19837. 0.2083]]], device='cuda:0')
  19838. loss_train_step before backward: tensor(0.0152, device='cuda:0', grad_fn=<MseLossBackward>)
  19839. loss_train_step after backward: tensor(0.0152, device='cuda:0', grad_fn=<MseLossBackward>)
  19840. loss_train: 0.25693511217832565
  19841. step: 14
  19842. running loss: 0.018352508012737547
  19843. Train Steps: 14/90 Loss: 0.0184 torch.Size([8, 600, 800])
  19844. torch.Size([8, 8])
  19845. tensor([[0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123],
  19846. [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517],
  19847. [0.6250, 0.4013, 0.8525, 0.5417, 0.4037, 0.5117, 0.6325, 0.5017],
  19848. [0.6284, 0.4093, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167],
  19849. [0.6161, 0.4024, 0.8662, 0.4683, 0.4935, 0.5364, 0.6063, 0.5567],
  19850. [0.6248, 0.4185, 0.8500, 0.5767, 0.4463, 0.4550, 0.5613, 0.5917],
  19851. [0.6173, 0.4114, 0.7325, 0.2500, 0.4213, 0.1917, 0.5338, 0.5700],
  19852. [0.6255, 0.4017, 0.8688, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901]],
  19853. device='cuda:0', dtype=torch.float64)
  19854. predictions are: tensor([[ 0.5322, -0.4184, 1.0965, -1.2336, -0.5849, -1.0054, 0.0292, 0.1176],
  19855. [ 0.6723, -0.3306, 0.9867, -1.1890, -0.4479, -1.1556, 0.2707, 0.3578],
  19856. [ 0.4028, -0.5790, 1.8348, 0.1465, -0.5165, 0.0932, 0.5661, 0.0781],
  19857. [ 0.6797, -0.3641, 1.9693, -0.1283, -0.6853, -0.4126, 0.5338, 0.0233],
  19858. [ 0.6101, -0.3995, 1.8501, -0.2459, -0.1648, 0.1547, 0.4233, 0.2387],
  19859. [ 0.5019, -0.4424, 1.6876, 0.3924, -0.2844, -0.1806, 0.2704, 0.4552],
  19860. [ 0.2637, -0.5703, 1.0420, -1.2051, -0.4135, -1.2683, 0.0640, 0.3140],
  19861. [ 0.7188, -0.3408, 1.7550, -0.8725, -0.6918, -0.6287, 0.5274, 0.0789]],
  19862. device='cuda:0', grad_fn=<AddmmBackward>)
  19863. landmarks are: tensor([[[ 0.5342, -0.4280, 1.0051, -1.0619, -0.5711, -1.0388, 0.2226,
  19864. 0.0802],
  19865. [ 0.5713, -0.3671, 0.8665, -1.0696, -0.3691, -1.2236, 0.3527,
  19866. 0.2622],
  19867. [ 0.6010, -0.4321, 1.6517, 0.2160, -0.4210, 0.0774, 0.6356,
  19868. 0.0313],
  19869. [ 0.6164, -0.3956, 1.8249, -0.1150, -0.6000, -0.5076, 0.5836,
  19870. 0.1005],
  19871. [ 0.5598, -0.4273, 1.7152, -0.1227, -0.0065, 0.1917, 0.5143,
  19872. 0.2853],
  19873. [ 0.5999, -0.3530, 1.6402, 0.3777, -0.2247, -0.1843, 0.3065,
  19874. 0.4470],
  19875. [ 0.5652, -0.3858, 1.0975, -1.1312, -0.3402, -1.4006, 0.1794,
  19876. 0.3469],
  19877. [ 0.6031, -0.4307, 1.7268, -0.8002, -0.6058, -0.6462, 0.6442,
  19878. -0.0220]]], device='cuda:0')
  19879. loss_train_step before backward: tensor(0.0100, device='cuda:0', grad_fn=<MseLossBackward>)
  19880. loss_train_step after backward: tensor(0.0100, device='cuda:0', grad_fn=<MseLossBackward>)
  19881. loss_train: 0.2669285275042057
  19882. step: 15
  19883. running loss: 0.017795235166947046
  19884. Train Steps: 15/90 Loss: 0.0178 torch.Size([8, 600, 800])
  19885. torch.Size([8, 8])
  19886. tensor([[0.6339, 0.4123, 0.8638, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436],
  19887. [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834],
  19888. [0.6196, 0.4088, 0.8888, 0.4583, 0.4500, 0.5683, 0.6138, 0.5883],
  19889. [0.6164, 0.4076, 0.8838, 0.4117, 0.3713, 0.5550, 0.6238, 0.5350],
  19890. [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390],
  19891. [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650],
  19892. [0.6126, 0.3954, 0.8538, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350],
  19893. [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114]],
  19894. device='cuda:0', dtype=torch.float64)
  19895. predictions are: tensor([[ 7.4917e-01, -2.9123e-01, 1.6643e+00, -7.0021e-02, -5.4613e-01,
  19896. 1.4726e-02, 8.5315e-01, 2.3208e-01],
  19897. [ 8.6743e-01, -1.8532e-01, 1.3006e+00, -1.5328e+00, -2.5562e-01,
  19898. -1.6725e+00, 4.7231e-01, 1.7271e-02],
  19899. [ 5.4636e-01, -3.7996e-01, 1.6489e+00, -4.4702e-01, -2.6623e-01,
  19900. 3.5819e-02, 3.5702e-01, 3.4274e-01],
  19901. [ 4.5332e-01, -4.3862e-01, 1.6072e+00, -5.8392e-01, -5.8043e-01,
  19902. 1.6829e-02, 4.6827e-01, 2.7635e-01],
  19903. [ 5.6296e-01, -3.9525e-01, 1.6733e+00, -2.4120e-01, -5.7210e-01,
  19904. -2.1659e-01, 1.4990e-01, 1.5060e-01],
  19905. [ 5.0192e-01, -4.1994e-01, 1.5758e+00, 1.8359e-01, -5.1492e-01,
  19906. -4.4539e-01, 2.1972e-01, 2.5103e-01],
  19907. [ 5.4839e-01, -4.0392e-01, 1.6198e+00, -1.2432e-01, -5.0563e-01,
  19908. -2.9601e-01, 7.4577e-02, 1.6226e-01],
  19909. [ 5.4278e-01, -4.0608e-01, 1.6447e+00, -8.1759e-02, -3.0073e-01,
  19910. -3.1521e-04, 3.1638e-01, 1.0237e-01]], device='cuda:0',
  19911. grad_fn=<AddmmBackward>)
  19912. landmarks are: tensor([[[ 0.6421, -0.3816, 1.7037, 0.1929, -0.4037, 0.2391, 1.1861,
  19913. 0.2249],
  19914. [ 0.6126, -0.4057, 1.4439, -1.3159, -0.1150, -1.5777, 0.5537,
  19915. -0.0530],
  19916. [ 0.5759, -0.3978, 1.8192, -0.1689, -0.2074, 0.3392, 0.5490,
  19917. 0.4316],
  19918. [ 0.5614, -0.4032, 1.7961, -0.3844, -0.5711, 0.2776, 0.5952,
  19919. 0.1852],
  19920. [ 0.5250, -0.4661, 1.7383, -0.0765, -0.4268, -0.0226, 0.2535,
  19921. 0.2035],
  19922. [ 0.5770, -0.4036, 1.7095, 0.3084, -0.3691, -0.2690, 0.2314,
  19923. 0.3238],
  19924. [ 0.5435, -0.4597, 1.6575, 0.0159, -0.3229, -0.1150, 0.1877,
  19925. 0.1854],
  19926. [ 0.5450, -0.4730, 1.7420, 0.1372, -0.1919, 0.2614, 0.4976,
  19927. 0.0764]]], device='cuda:0')
  19928. loss_train_step before backward: tensor(0.0220, device='cuda:0', grad_fn=<MseLossBackward>)
  19929. loss_train_step after backward: tensor(0.0220, device='cuda:0', grad_fn=<MseLossBackward>)
  19930. loss_train: 0.2889367938041687
  19931. step: 16
  19932. running loss: 0.018058549612760544
  19933.  
  19934. Train Steps: 16/90 Loss: 0.0181 torch.Size([8, 600, 800])
  19935. torch.Size([8, 8])
  19936. tensor([[0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466],
  19937. [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217],
  19938. [0.6111, 0.3995, 0.8788, 0.4567, 0.3813, 0.4833, 0.5450, 0.5700],
  19939. [0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938],
  19940. [0.6201, 0.4017, 0.8871, 0.4621, 0.3517, 0.4675, 0.5999, 0.5106],
  19941. [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749],
  19942. [0.6138, 0.4101, 0.8800, 0.5083, 0.4637, 0.5950, 0.5587, 0.5077],
  19943. [0.6277, 0.4083, 0.8350, 0.2717, 0.4562, 0.1800, 0.5918, 0.4878]],
  19944. device='cuda:0', dtype=torch.float64)
  19945. predictions are: tensor([[ 5.1427e-01, -4.1541e-01, 1.5883e+00, -6.4958e-02, -2.9016e-01,
  19946. 1.5929e-03, 1.6787e-01, 2.4338e-01],
  19947. [ 7.8892e-01, -2.8638e-01, 1.8092e+00, -2.2893e-01, -4.9548e-01,
  19948. 9.6385e-02, 7.4152e-01, 1.4290e-01],
  19949. [ 6.8751e-01, -3.0935e-01, 1.6456e+00, -3.5264e-01, -5.5861e-01,
  19950. -2.3207e-01, 2.7769e-01, 2.3544e-01],
  19951. [ 7.5018e-01, -3.1599e-01, 1.6878e+00, 2.2710e-01, -5.1088e-01,
  19952. -2.6419e-01, 6.2945e-01, 1.2593e-02],
  19953. [ 6.6079e-01, -3.3558e-01, 1.6973e+00, -3.2839e-01, -6.2279e-01,
  19954. -3.0419e-01, 4.1372e-01, 1.6635e-01],
  19955. [ 5.2827e-01, -3.7146e-01, 1.0872e+00, -1.0771e+00, -6.6342e-01,
  19956. -8.6333e-01, 2.2034e-01, 3.3167e-01],
  19957. [ 5.4088e-01, -4.0553e-01, 1.7217e+00, -8.6147e-02, -1.6320e-01,
  19958. 1.2475e-01, 3.7833e-01, 2.1609e-01],
  19959. [ 4.0423e-01, -4.5951e-01, 1.3544e+00, -1.2396e+00, -2.0415e-01,
  19960. -1.3715e+00, 2.6294e-01, 7.7104e-02]], device='cuda:0',
  19961. grad_fn=<AddmmBackward>)
  19962. landmarks are: tensor([[[ 0.5178, -0.4480, 1.6748, 0.0620, -0.2767, 0.2083, 0.1067,
  19963. 0.2386],
  19964. [ 0.6047, -0.4209, 1.7557, -0.0303, -0.4845, 0.2545, 0.6587,
  19965. 0.1236],
  19966. [ 0.5368, -0.4406, 1.7730, -0.1766, -0.5249, -0.0534, 0.2314,
  19967. 0.3469],
  19968. [ 0.6395, -0.4213, 1.7037, 0.3623, -0.4326, -0.1073, 0.6560,
  19969. -0.0049],
  19970. [ 0.5783, -0.4306, 1.8114, -0.1515, -0.6617, -0.1268, 0.4851,
  19971. 0.0727],
  19972. [ 0.5419, -0.4160, 1.1810, -0.8939, -0.6808, -0.7463, 0.2485,
  19973. 0.3694],
  19974. [ 0.5491, -0.3918, 1.7788, 0.0620, -0.1439, 0.4624, 0.2946,
  19975. 0.0592],
  19976. [ 0.6132, -0.4002, 1.5709, -1.0311, -0.1785, -1.4545, 0.4474,
  19977. -0.0328]]], device='cuda:0')
  19978. loss_train_step before backward: tensor(0.0146, device='cuda:0', grad_fn=<MseLossBackward>)
  19979. loss_train_step after backward: tensor(0.0146, device='cuda:0', grad_fn=<MseLossBackward>)
  19980. loss_train: 0.3035747576504946
  19981. step: 17
  19982. running loss: 0.017857338685323212
  19983. Train Steps: 17/90 Loss: 0.0179 torch.Size([8, 600, 800])
  19984. torch.Size([8, 8])
  19985. tensor([[0.6213, 0.4001, 0.7712, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183],
  19986. [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726],
  19987. [0.6200, 0.3978, 0.8900, 0.4550, 0.3775, 0.5200, 0.6150, 0.5367],
  19988. [0.6097, 0.4000, 0.7325, 0.2667, 0.3450, 0.3517, 0.5284, 0.5045],
  19989. [0.6263, 0.4039, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804],
  19990. [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6088, 0.5583],
  19991. [0.6203, 0.4096, 0.8862, 0.4267, 0.3538, 0.4117, 0.6025, 0.5650],
  19992. [0.6128, 0.4022, 0.8738, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064]],
  19993. device='cuda:0', dtype=torch.float64)
  19994. predictions are: tensor([[ 0.7755, -0.2566, 1.2124, -1.1970, -0.2807, -1.3949, 0.4493, 0.0860],
  19995. [ 0.6724, -0.2987, 1.2744, -0.9307, -0.5428, -0.9535, 0.2306, 0.0451],
  19996. [ 0.5606, -0.4360, 1.8056, -0.0158, -0.4819, 0.1948, 0.5600, 0.1729],
  19997. [ 0.4362, -0.4670, 1.2045, -0.9879, -0.5975, -0.6874, 0.2234, 0.1330],
  19998. [ 0.5636, -0.4414, 1.9031, -0.1640, -0.5636, -0.0504, 0.6355, 0.0259],
  19999. [ 0.7105, -0.3035, 1.0117, -1.1499, -0.4594, -1.1200, 0.4641, 0.2386],
  20000. [ 0.5666, -0.4205, 1.8395, -0.2656, -0.5567, -0.3416, 0.5523, 0.3514],
  20001. [ 0.4156, -0.5127, 1.7689, 0.0910, 0.0355, 0.0478, 0.2672, 0.1672]],
  20002. device='cuda:0', grad_fn=<AddmmBackward>)
  20003. landmarks are: tensor([[[ 0.5838, -0.4379, 1.2764, -1.3082, -0.2824, -1.4545, 0.4162,
  20004. 0.1082],
  20005. [ 0.4997, -0.4446, 1.1610, -0.9772, -0.6058, -1.0311, 0.1404,
  20006. -0.1031],
  20007. [ 0.5776, -0.4484, 1.8249, -0.1843, -0.5423, 0.1159, 0.5547,
  20008. 0.1929],
  20009. [ 0.5303, -0.4384, 1.0975, -1.0542, -0.6924, -0.6616, 0.1548,
  20010. 0.0442],
  20011. [ 0.6072, -0.4206, 1.8711, -0.2536, -0.6115, -0.1304, 0.6812,
  20012. -0.0670],
  20013. [ 0.5992, -0.3968, 0.9307, -1.3497, -0.4730, -1.0465, 0.5259,
  20014. 0.2930],
  20015. [ 0.5790, -0.3940, 1.8076, -0.3152, -0.6520, -0.3844, 0.4970,
  20016. 0.3238],
  20017. [ 0.5446, -0.4280, 1.7499, 0.0543, 0.0156, 0.1301, 0.1918,
  20018. 0.0532]]], device='cuda:0')
  20019. loss_train_step before backward: tensor(0.0073, device='cuda:0', grad_fn=<MseLossBackward>)
  20020. loss_train_step after backward: tensor(0.0073, device='cuda:0', grad_fn=<MseLossBackward>)
  20021. loss_train: 0.3108562519773841
  20022. step: 18
  20023. running loss: 0.01726979177652134
  20024. Train Steps: 18/90 Loss: 0.0173 torch.Size([8, 600, 800])
  20025. torch.Size([8, 8])
  20026. tensor([[0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986],
  20027. [0.6293, 0.4097, 0.8800, 0.2517, 0.5262, 0.2600, 0.7430, 0.5378],
  20028. [0.6120, 0.4014, 0.6863, 0.2817, 0.3700, 0.2783, 0.5513, 0.5667],
  20029. [0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166],
  20030. [0.6228, 0.4119, 0.7938, 0.2233, 0.4674, 0.1773, 0.6188, 0.5433],
  20031. [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356],
  20032. [0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407],
  20033. [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183]],
  20034. device='cuda:0', dtype=torch.float64)
  20035. predictions are: tensor([[ 0.6249, -0.3117, 1.7247, -0.4558, -0.2784, -1.0430, 0.3631, 0.0424],
  20036. [ 0.5624, -0.4253, 1.5911, -1.0202, 0.0873, -1.0461, 0.9152, 0.1176],
  20037. [ 0.4162, -0.4453, 0.9554, -0.9243, -0.5591, -0.8039, 0.1167, 0.2402],
  20038. [ 0.5104, -0.4898, 1.7279, -0.0447, -0.4961, 0.3425, 0.7043, 0.1548],
  20039. [ 0.5848, -0.3563, 1.2298, -1.0580, -0.2116, -1.2383, 0.4013, 0.1006],
  20040. [ 0.5874, -0.3731, 1.3149, -0.9523, -0.3525, -1.0013, 0.5720, 0.1029],
  20041. [ 0.3819, -0.4549, 1.5499, -0.3267, -0.6556, -0.6680, -0.0268, 0.1221],
  20042. [ 0.7683, -0.3051, 1.7793, -0.1137, -0.4840, 0.5970, 0.6887, 0.1274]],
  20043. device='cuda:0', grad_fn=<AddmmBackward>)
  20044. landmarks are: tensor([[[ 0.6054, -0.3641, 1.8711, -0.5153, -0.2305, -1.3005, 0.5948,
  20045. 0.0171],
  20046. [ 0.6207, -0.3936, 1.7788, -1.1235, 0.1448, -1.0850, 1.1459,
  20047. 0.1982],
  20048. [ 0.5410, -0.4321, 0.8838, -0.9849, -0.5769, -1.0003, 0.2603,
  20049. 0.3315],
  20050. [ 0.5944, -0.4490, 1.8643, -0.0659, -0.5147, 0.1235, 0.7684,
  20051. 0.1004],
  20052. [ 0.5908, -0.3832, 1.3804, -1.2543, -0.1270, -1.4671, 0.5721,
  20053. 0.2237],
  20054. [ 0.6174, -0.4225, 1.4975, -1.1709, -0.3174, -1.1806, 0.6539,
  20055. 0.1879],
  20056. [ 0.5470, -0.3952, 1.6377, -0.4253, -0.6289, -0.8079, 0.0249,
  20057. 0.2116],
  20058. [ 0.5934, -0.4276, 1.8192, -0.1458, -0.4499, 0.3777, 0.6702,
  20059. 0.1082]]], device='cuda:0')
  20060. loss_train_step before backward: tensor(0.0143, device='cuda:0', grad_fn=<MseLossBackward>)
  20061. loss_train_step after backward: tensor(0.0143, device='cuda:0', grad_fn=<MseLossBackward>)
  20062. loss_train: 0.3251751000061631
  20063. step: 19
  20064. running loss: 0.017114478947692795
  20065. Train Steps: 19/90 Loss: 0.0171 torch.Size([8, 600, 800])
  20066. torch.Size([8, 8])
  20067. tensor([[0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412],
  20068. [0.6266, 0.4101, 0.8350, 0.2333, 0.3950, 0.2950, 0.6264, 0.4921],
  20069. [0.6097, 0.4000, 0.7325, 0.2667, 0.3450, 0.3517, 0.5284, 0.5045],
  20070. [ nan, nan, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633],
  20071. [0.6371, 0.4092, 0.8337, 0.5850, 0.3950, 0.5117, 0.6559, 0.5262],
  20072. [0.6286, 0.4055, 0.9000, 0.4717, 0.3763, 0.4683, 0.7018, 0.5494],
  20073. [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320],
  20074. [0.6260, 0.4120, 0.8013, 0.2350, 0.4888, 0.1533, 0.6281, 0.4895]],
  20075. device='cuda:0', dtype=torch.float64)
  20076. predictions are: tensor([[ 6.3753e-01, -3.0106e-01, 1.7238e+00, -5.3725e-01, -4.3524e-01,
  20077. -8.7413e-01, 2.8037e-01, 1.2785e-01],
  20078. [ 7.4946e-01, -2.6226e-01, 1.4602e+00, -1.1061e+00, -3.8495e-01,
  20079. -8.8866e-01, 5.5216e-01, 3.8942e-02],
  20080. [ 5.6102e-01, -3.5893e-01, 1.1527e+00, -1.0360e+00, -5.2439e-01,
  20081. -6.7386e-01, 1.5924e-01, 1.1717e-01],
  20082. [-1.0101e+00, -1.4053e+00, 1.1850e+00, -1.0470e+00, -4.7518e-01,
  20083. -1.0037e+00, -4.8302e-02, 1.9874e-01],
  20084. [ 8.3262e-01, -2.4918e-01, 1.6203e+00, 3.0693e-01, -3.8922e-01,
  20085. 1.4029e-01, 6.9893e-01, 5.8639e-02],
  20086. [ 9.2227e-01, -1.9825e-01, 1.8672e+00, -6.0081e-02, -4.7998e-01,
  20087. 2.9148e-02, 8.5436e-01, 8.6164e-02],
  20088. [ 7.5961e-01, -2.9459e-01, 1.6385e+00, 1.3106e-01, -3.7348e-01,
  20089. 1.4454e-01, 6.9711e-01, 1.6616e-01],
  20090. [ 7.5710e-01, -2.3514e-01, 1.3542e+00, -1.1528e+00, 1.2851e-03,
  20091. -1.4269e+00, 5.6705e-01, 8.8829e-02]], device='cuda:0',
  20092. grad_fn=<AddmmBackward>)
  20093. landmarks are: tensor([[[ 0.5784, -0.4006, 1.7911, -0.5701, -0.5192, -1.0331, 0.4137,
  20094. 0.2139],
  20095. [ 0.6081, -0.3918, 1.5709, -1.2082, -0.4614, -0.9233, 0.6072,
  20096. -0.0129],
  20097. [ 0.5303, -0.4384, 1.0975, -1.0542, -0.6924, -0.6616, 0.1548,
  20098. 0.0442],
  20099. [-2.2859, -2.2859, 1.2030, -1.0288, -0.4961, -1.1081, 0.0813,
  20100. 0.3161],
  20101. [ 0.6569, -0.3960, 1.5651, 0.4162, -0.4614, 0.0774, 0.7438,
  20102. 0.1447],
  20103. [ 0.6174, -0.4129, 1.8711, -0.1073, -0.5480, -0.1227, 0.9558,
  20104. 0.2516],
  20105. [ 0.6273, -0.4393, 1.6402, 0.1313, -0.5076, 0.0467, 1.1532,
  20106. 0.1715],
  20107. [ 0.6053, -0.3830, 1.4150, -1.2005, -0.0284, -1.5777, 0.6154,
  20108. -0.0250]]], device='cuda:0')
  20109. loss_train_step before backward: tensor(0.0515, device='cuda:0', grad_fn=<MseLossBackward>)
  20110. loss_train_step after backward: tensor(0.0515, device='cuda:0', grad_fn=<MseLossBackward>)
  20111. loss_train: 0.37666080240160227
  20112. step: 20
  20113. running loss: 0.018833040120080115
  20114.  
  20115. Train Steps: 20/90 Loss: 0.0188 torch.Size([8, 600, 800])
  20116. torch.Size([8, 8])
  20117. tensor([[0.6336, 0.4191, 0.8938, 0.5167, 0.3937, 0.3517, 0.7343, 0.5748],
  20118. [0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124],
  20119. [0.6263, 0.4065, 0.9038, 0.4317, 0.3588, 0.4550, 0.6325, 0.5250],
  20120. [0.6268, 0.4029, 0.8500, 0.2683, 0.3937, 0.3500, 0.6860, 0.5297],
  20121. [0.6277, 0.4057, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823],
  20122. [0.6030, 0.3969, 0.7988, 0.3917, 0.3450, 0.3667, 0.5266, 0.4700],
  20123. [ nan, nan, 0.6412, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700],
  20124. [0.6259, 0.4156, 0.8812, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960]],
  20125. device='cuda:0', dtype=torch.float64)
  20126. predictions are: tensor([[ 0.4839, -0.5000, 1.6635, 0.1886, -0.3307, -0.3211, 0.7333, 0.2023],
  20127. [ 0.6194, -0.3507, 1.4114, -0.9203, -0.3818, -0.8013, 0.4622, 0.0413],
  20128. [ 0.6624, -0.3586, 1.8314, -0.1974, -0.5329, 0.0903, 0.6837, 0.1530],
  20129. [ 0.8656, -0.2259, 1.5109, -0.8773, -0.4817, -0.4769, 0.8409, 0.1297],
  20130. [ 0.8424, -0.2204, 1.4692, -1.0215, -0.1932, -1.1525, 0.6111, -0.0318],
  20131. [ 0.6292, -0.3355, 1.4958, -0.4595, -0.6039, -0.3666, 0.1997, 0.0866],
  20132. [-0.7059, -1.2208, 0.8391, -1.2020, -0.2385, -1.2981, 0.0876, 0.3023],
  20133. [ 0.7580, -0.2258, 1.6574, -0.7458, -0.0913, -1.0447, 0.4986, 0.0383]],
  20134. device='cuda:0', grad_fn=<AddmmBackward>)
  20135. landmarks are: tensor([[[ 0.6405, -0.3503, 1.8423, 0.1005, -0.4672, -0.6616, 1.1057,
  20136. 0.3692],
  20137. [ 0.5790, -0.4079, 1.5929, -1.0630, -0.4729, -1.0725, 0.4137,
  20138. 0.0807],
  20139. [ 0.6070, -0.4085, 1.8885, -0.2921, -0.6289, -0.1843, 0.6356,
  20140. 0.1390],
  20141. [ 0.6092, -0.4249, 1.6402, -1.0465, -0.4672, -0.6693, 0.8827,
  20142. 0.1608],
  20143. [ 0.6132, -0.4122, 1.5478, -1.0619, -0.2709, -1.4314, 0.5500,
  20144. -0.0583],
  20145. [ 0.4992, -0.4525, 1.4035, -0.4768, -0.6924, -0.5923, 0.1465,
  20146. -0.1151],
  20147. [-2.2859, -2.2859, 0.6760, -1.4083, -0.3286, -1.4160, 0.2487,
  20148. 0.3469],
  20149. [ 0.6052, -0.3663, 1.7845, -0.8156, -0.0804, -1.4237, 0.5866,
  20150. 0.0051]]], device='cuda:0')
  20151. loss_train_step before backward: tensor(0.0800, device='cuda:0', grad_fn=<MseLossBackward>)
  20152. loss_train_step after backward: tensor(0.0800, device='cuda:0', grad_fn=<MseLossBackward>)
  20153. loss_train: 0.4566794792190194
  20154. step: 21
  20155. running loss: 0.02174664186757235
  20156. Train Steps: 21/90 Loss: 0.0217 torch.Size([8, 600, 800])
  20157. torch.Size([8, 8])
  20158. tensor([[0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895],
  20159. [0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297],
  20160. [0.6257, 0.4167, 0.8775, 0.3433, 0.3563, 0.4133, 0.6200, 0.5667],
  20161. [0.6262, 0.4163, 0.8850, 0.5183, 0.3763, 0.4150, 0.6025, 0.5500],
  20162. [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309],
  20163. [0.6179, 0.4118, 0.7278, 0.4237, 0.3588, 0.3400, 0.5675, 0.5917],
  20164. [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467],
  20165. [0.6200, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391]],
  20166. device='cuda:0', dtype=torch.float64)
  20167. predictions are: tensor([[ 0.7002, -0.3247, 1.2085, -1.4291, -0.0743, -1.4957, 0.4974, -0.0724],
  20168. [ 0.4774, -0.4847, 1.7084, 0.1631, -0.4217, -0.1387, 0.7403, 0.0630],
  20169. [ 0.5838, -0.3842, 1.6920, -0.6905, -0.5661, -0.2903, 0.6077, 0.2382],
  20170. [ 0.5264, -0.4329, 1.6813, -0.0333, -0.4499, -0.3091, 0.4960, 0.1460],
  20171. [ 0.7125, -0.3116, 1.7807, -0.0740, -0.3466, -0.4352, 0.7564, 0.0296],
  20172. [ 0.3954, -0.4781, 1.2758, -0.6132, -0.4806, -0.7404, 0.2370, 0.3153],
  20173. [ 0.7172, -0.2866, 1.5352, -0.8002, -0.5477, -0.3290, 0.5524, 0.1585],
  20174. [ 0.5864, -0.3569, 1.5866, -0.9917, -0.2150, -1.0676, 0.5041, 0.0485]],
  20175. device='cuda:0', grad_fn=<AddmmBackward>)
  20176. landmarks are: tensor([[[ 0.5828, -0.4417, 1.2476, -1.3929, -0.1727, -1.5700, 0.4694,
  20177. -0.0248],
  20178. [ 0.6504, -0.3647, 1.7730, 0.2930, -0.6058, -0.2382, 0.7109,
  20179. 0.1608],
  20180. [ 0.6040, -0.3614, 1.7672, -0.7001, -0.6404, -0.3768, 0.5778,
  20181. 0.3315],
  20182. [ 0.6066, -0.3632, 1.8018, 0.1082, -0.5480, -0.3691, 0.4970,
  20183. 0.2545],
  20184. [ 0.6520, -0.3623, 1.8885, 0.0313, -0.5538, -0.5384, 0.6926,
  20185. 0.1661],
  20186. [ 0.5680, -0.3840, 1.0756, -0.3290, -0.6289, -0.7155, 0.3353,
  20187. 0.4470],
  20188. [ 0.5484, -0.4591, 1.5940, -0.7771, -0.6693, -0.3460, 0.4219,
  20189. 0.2391],
  20190. [ 0.5776, -0.3987, 1.6005, -0.9121, -0.3844, -1.2358, 0.4247,
  20191. 0.2043]]], device='cuda:0')
  20192. loss_train_step before backward: tensor(0.0122, device='cuda:0', grad_fn=<MseLossBackward>)
  20193. loss_train_step after backward: tensor(0.0122, device='cuda:0', grad_fn=<MseLossBackward>)
  20194. loss_train: 0.46885613538324833
  20195. step: 22
  20196. running loss: 0.02131164251742038
  20197. Train Steps: 22/90 Loss: 0.0213 torch.Size([8, 600, 800])
  20198. torch.Size([8, 8])
  20199. tensor([[0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250],
  20200. [0.6289, 0.4019, 0.8113, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332],
  20201. [0.6199, 0.4071, 0.7600, 0.2117, 0.4037, 0.2767, 0.6138, 0.5550],
  20202. [0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350],
  20203. [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207],
  20204. [0.6090, 0.4010, 0.7838, 0.3483, 0.3538, 0.3783, 0.5462, 0.5077],
  20205. [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967],
  20206. [0.6219, 0.4097, 0.8738, 0.3400, 0.3563, 0.4117, 0.5975, 0.5683]],
  20207. device='cuda:0', dtype=torch.float64)
  20208. predictions are: tensor([[ 0.6196, -0.3860, 1.7935, 0.0901, -0.3193, 0.3917, 0.6185, 0.0979],
  20209. [ 0.4539, -0.5063, 1.5252, 0.2251, -0.4727, 0.1093, 0.7826, 0.1797],
  20210. [ 0.6377, -0.3392, 1.3309, -1.1378, -0.4062, -1.0118, 0.6385, 0.1555],
  20211. [ 0.4766, -0.4483, 1.3842, -1.0258, -0.1972, -1.3189, 0.4875, 0.1217],
  20212. [ 0.6683, -0.3750, 1.4588, -0.9875, -0.2644, -1.2481, 0.7489, 0.0369],
  20213. [ 0.4246, -0.4852, 1.4205, -0.7114, -0.6263, -0.5640, 0.2431, 0.1097],
  20214. [ 0.5375, -0.3941, 1.6243, -1.0296, -0.0724, -1.3532, 0.6481, 0.0064],
  20215. [ 0.5658, -0.3992, 1.6959, -0.5945, -0.7086, -0.3629, 0.4398, 0.2220]],
  20216. device='cuda:0', grad_fn=<AddmmBackward>)
  20217. landmarks are: tensor([[[ 5.9913e-01, -3.8029e-01, 1.8018e+00, -5.3426e-02, -3.4596e-01,
  20218. 1.8522e-01, 5.3741e-01, 1.3903e-01],
  20219. [ 6.1907e-01, -4.2971e-01, 1.4612e+00, 2.3911e-01, -4.9607e-01,
  20220. 3.1255e-02, 1.1166e+00, 1.7680e-01],
  20221. [ 5.7754e-01, -4.0539e-01, 1.2245e+00, -1.3082e+00, -4.2102e-01,
  20222. -1.0080e+00, 5.4896e-01, 2.7760e-01],
  20223. [ 5.7679e-01, -4.0308e-01, 1.3838e+00, -1.1527e+00, -2.1876e-01,
  20224. -1.4216e+00, 4.3790e-01, 1.8502e-01],
  20225. [ 6.1322e-01, -4.2479e-01, 1.5247e+00, -1.1620e+00, -2.8822e-01,
  20226. -1.3159e+00, 6.5445e-01, 1.1931e-01],
  20227. [ 5.2702e-01, -4.3356e-01, 1.3342e+00, -6.7698e-01, -6.5196e-01,
  20228. -5.3841e-01, 2.3702e-01, 5.9193e-02],
  20229. [ 6.0479e-01, -3.7229e-01, 1.6517e+00, -1.0773e+00, 4.6189e-04,
  20230. -1.5161e+00, 5.8660e-01, 8.0947e-03],
  20231. [ 5.8672e-01, -3.9369e-01, 1.7499e+00, -7.1547e-01, -6.4042e-01,
  20232. -3.8445e-01, 4.7390e-01, 3.3918e-01]]], device='cuda:0')
  20233. loss_train_step before backward: tensor(0.0080, device='cuda:0', grad_fn=<MseLossBackward>)
  20234. loss_train_step after backward: tensor(0.0080, device='cuda:0', grad_fn=<MseLossBackward>)
  20235. loss_train: 0.4768888894468546
  20236. step: 23
  20237. running loss: 0.02073429954116759
  20238. Train Steps: 23/90 Loss: 0.0207 torch.Size([8, 600, 800])
  20239. torch.Size([8, 8])
  20240. tensor([[0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222],
  20241. [0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904],
  20242. [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426],
  20243. [ nan, nan, 0.8850, 0.2817, 0.5112, 0.2183, 0.7184, 0.5436],
  20244. [0.6263, 0.4233, 0.7924, 0.4626, 0.3788, 0.2883, 0.5573, 0.6047],
  20245. [0.6200, 0.4059, 0.8700, 0.4900, 0.4163, 0.5000, 0.6162, 0.5467],
  20246. [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012],
  20247. [0.6136, 0.4085, 0.6688, 0.2317, 0.3862, 0.2367, 0.5517, 0.5783]],
  20248. device='cuda:0', dtype=torch.float64)
  20249. predictions are: tensor([[ 6.4015e-01, -3.2741e-01, 1.5286e+00, -5.8203e-01, -2.0736e-01,
  20250. -1.0390e+00, 3.7237e-01, 3.1638e-01],
  20251. [ 7.2093e-01, -2.9143e-01, 1.7759e+00, -1.1026e+00, -2.6972e-01,
  20252. -1.1049e+00, 7.2239e-01, -1.2084e-01],
  20253. [ 8.0535e-01, -2.6231e-01, 1.2904e+00, -1.1777e+00, -3.5370e-01,
  20254. -1.2656e+00, 5.2912e-01, -1.9021e-02],
  20255. [-5.5371e-01, -1.1874e+00, 1.8828e+00, -1.1686e+00, -5.8424e-03,
  20256. -1.0682e+00, 1.1814e+00, 1.2185e-01],
  20257. [ 5.2261e-01, -4.1431e-01, 1.4245e+00, -1.9491e-01, -5.6756e-01,
  20258. -6.1366e-01, 3.0487e-01, 4.5116e-01],
  20259. [ 5.1992e-01, -4.8953e-01, 1.7885e+00, -5.3513e-02, -4.4506e-01,
  20260. 3.5400e-01, 6.2226e-01, 1.5005e-01],
  20261. [ 8.1954e-01, -3.3110e-01, 1.7693e+00, 2.3936e-01, -6.8419e-01,
  20262. -1.6167e-03, 7.2799e-01, 3.1313e-02],
  20263. [ 6.6607e-01, -3.6078e-01, 9.4744e-01, -1.1733e+00, -5.2266e-01,
  20264. -1.1170e+00, 3.4295e-01, 1.2331e-01]], device='cuda:0',
  20265. grad_fn=<AddmmBackward>)
  20266. landmarks are: tensor([[[ 0.6237, -0.2983, 1.3919, -0.4691, -0.0457, -1.2313, 0.2453,
  20267. 0.5882],
  20268. [ 0.6126, -0.3871, 1.6229, -1.0773, -0.2132, -1.3698, 0.5829,
  20269. -0.0209],
  20270. [ 0.5784, -0.4153, 1.2972, -1.2541, -0.2265, -1.4553, 0.4741,
  20271. 0.2203],
  20272. [-2.2859, -2.2859, 1.8018, -0.9849, 0.0755, -1.2774, 1.0326,
  20273. 0.2249],
  20274. [ 0.6069, -0.3309, 1.3742, -0.1493, -0.5365, -0.9541, 0.2884,
  20275. 0.5071],
  20276. [ 0.5776, -0.4109, 1.7326, -0.0226, -0.3633, 0.0236, 0.5605,
  20277. 0.2391],
  20278. [ 0.6249, -0.4352, 1.8018, 0.2545, -0.6173, -0.1997, 0.6401,
  20279. 0.0291],
  20280. [ 0.5483, -0.3991, 0.8030, -1.2159, -0.5018, -1.1928, 0.2624,
  20281. 0.3852]]], device='cuda:0')
  20282. loss_train_step before backward: tensor(0.0837, device='cuda:0', grad_fn=<MseLossBackward>)
  20283. loss_train_step after backward: tensor(0.0837, device='cuda:0', grad_fn=<MseLossBackward>)
  20284. loss_train: 0.5606125611811876
  20285. step: 24
  20286. running loss: 0.02335885671588282
  20287.  
  20288. Train Steps: 24/90 Loss: 0.0234 torch.Size([8, 600, 800])
  20289. torch.Size([8, 8])
  20290. tensor([[0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578],
  20291. [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650],
  20292. [0.6252, 0.4158, 0.8988, 0.4083, 0.3788, 0.4783, 0.6225, 0.5633],
  20293. [ nan, nan, 0.6793, 0.2110, 0.4012, 0.2167, 0.5112, 0.5583],
  20294. [0.6250, 0.4236, 0.8638, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799],
  20295. [0.6204, 0.4110, 0.7913, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367],
  20296. [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5413, 0.5433],
  20297. [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100]],
  20298. device='cuda:0', dtype=torch.float64)
  20299. predictions are: tensor([[ 0.7548, -0.3042, 1.0104, -1.2783, -0.3081, -1.4168, 0.4733, 0.0882],
  20300. [ 0.3597, -0.5199, 1.6652, -0.3600, -0.5313, -0.7199, 0.4997, 0.3743],
  20301. [ 1.0061, -0.1646, 1.8598, -0.2678, -0.5175, 0.0687, 0.8785, 0.2038],
  20302. [-1.9778, -2.1261, 1.0573, -1.3015, -0.3206, -1.1825, 0.1999, 0.2042],
  20303. [ 0.6407, -0.3564, 1.7933, -0.4605, -0.3934, -0.7640, 0.5918, 0.2892],
  20304. [ 0.9847, -0.1499, 1.5094, -1.0364, -0.2792, -1.1041, 0.7410, 0.0412],
  20305. [ 0.8230, -0.2356, 1.6608, -0.5814, -0.5504, -0.7949, 0.4448, 0.0940],
  20306. [ 0.9102, -0.2794, 1.8353, -0.3655, -0.5575, -0.4388, 0.9005, 0.0544]],
  20307. device='cuda:0', grad_fn=<AddmmBackward>)
  20308. landmarks are: tensor([[[ 0.5232, -0.4189, 0.9358, -1.4006, -0.3921, -1.3698, 0.2555,
  20309. 0.2906],
  20310. [ 0.5894, -0.3503, 1.6402, -0.3614, -0.5827, -0.7925, 0.3238,
  20311. 0.3238],
  20312. [ 0.6017, -0.3654, 1.8654, -0.3998, -0.5365, -0.0765, 0.5894,
  20313. 0.3161],
  20314. [-2.2859, -2.2859, 0.8516, -1.3112, -0.4326, -1.2851, 0.0755,
  20315. 0.2930],
  20316. [ 0.6008, -0.3293, 1.7037, -0.5461, -0.4152, -0.8309, 0.3234,
  20317. 0.3928],
  20318. [ 0.5796, -0.3878, 1.3688, -1.0542, -0.4095, -1.1312, 0.5894,
  20319. 0.1929],
  20320. [ 0.5894, -0.3479, 1.7730, -0.6847, -0.5538, -0.7155, 0.2141,
  20321. 0.2237],
  20322. [ 0.6220, -0.4436, 1.8711, -0.3691, -0.6173, -0.4999, 0.6702,
  20323. 0.0697]]], device='cuda:0')
  20324. loss_train_step before backward: tensor(0.0251, device='cuda:0', grad_fn=<MseLossBackward>)
  20325. loss_train_step after backward: tensor(0.0251, device='cuda:0', grad_fn=<MseLossBackward>)
  20326. loss_train: 0.5856962967664003
  20327. step: 25
  20328. running loss: 0.023427851870656012
  20329. Train Steps: 25/90 Loss: 0.0234 torch.Size([8, 600, 800])
  20330. torch.Size([8, 8])
  20331. tensor([[ nan, nan, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650],
  20332. [0.6200, 0.4098, 0.8237, 0.2917, 0.4012, 0.2967, 0.6000, 0.5683],
  20333. [0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950],
  20334. [0.6197, 0.3986, 0.8800, 0.4617, 0.4188, 0.4783, 0.5687, 0.5550],
  20335. [0.6243, 0.4128, 0.7762, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417],
  20336. [0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567],
  20337. [0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773],
  20338. [0.6079, 0.3964, 0.7420, 0.2958, 0.3563, 0.2917, 0.5351, 0.4980]],
  20339. device='cuda:0', dtype=torch.float64)
  20340. predictions are: tensor([[-2.1749, -2.2414, 1.2203, -1.2753, -0.2672, -1.2905, 0.4293, 0.2614],
  20341. [ 0.8052, -0.2676, 1.6369, -0.9690, -0.5289, -1.0819, 0.6849, 0.2336],
  20342. [ 0.7062, -0.3927, 1.7734, 0.1701, -0.3351, 0.0147, 0.6362, 0.0818],
  20343. [ 0.6538, -0.3901, 1.7930, -0.1344, -0.4454, -0.0750, 0.6346, 0.2179],
  20344. [ 0.7714, -0.2682, 1.5192, -1.0016, -0.5511, -1.1028, 0.7895, 0.1417],
  20345. [ 0.8999, -0.1971, 1.8160, -0.1320, -0.1471, 0.0496, 0.6020, 0.2215],
  20346. [ 0.9350, -0.1469, 1.1757, -0.9296, -0.3252, -1.4202, 0.5431, 0.3288],
  20347. [ 0.6527, -0.3576, 1.3160, -0.9865, -0.6031, -1.0946, 0.3418, -0.0273]],
  20348. device='cuda:0', grad_fn=<AddmmBackward>)
  20349. landmarks are: tensor([[[-2.2859e+00, -2.2859e+00, 9.1750e-01, -1.3947e+00, -3.6905e-01,
  20350. -1.2467e+00, 2.3141e-01, 3.2379e-01],
  20351. [ 5.7783e-01, -3.9299e-01, 1.5189e+00, -9.3872e-01, -4.3256e-01,
  20352. -9.1563e-01, 4.8545e-01, 3.3918e-01],
  20353. [ 5.8199e-01, -4.7544e-01, 1.7095e+00, 1.9292e-01, -2.5358e-01,
  20354. 5.4350e-02, 4.9700e-01, 4.6189e-04],
  20355. [ 5.7627e-01, -4.4503e-01, 1.7788e+00, -1.5350e-01, -3.5173e-01,
  20356. -7.6520e-02, 3.4111e-01, 2.7760e-01],
  20357. [ 5.9766e-01, -3.7916e-01, 1.2995e+00, -1.0311e+00, -5.1917e-01,
  20358. -8.3865e-01, 5.8360e-01, 2.1601e-01],
  20359. [ 5.5381e-01, -4.1386e-01, 1.7557e+00, -1.8430e-01, -4.5897e-02,
  20360. 1.2417e-01, 4.2194e-01, 2.8530e-01],
  20361. [ 5.8412e-01, -3.5743e-01, 1.0859e+00, -9.5412e-01, -2.8245e-01,
  20362. -1.2851e+00, 3.4601e-01, 3.8081e-01],
  20363. [ 5.2194e-01, -4.5504e-01, 1.1415e+00, -9.1962e-01, -6.4042e-01,
  20364. -9.3872e-01, 1.8562e-01, 1.4106e-02]]], device='cuda:0')
  20365. loss_train_step before backward: tensor(0.0203, device='cuda:0', grad_fn=<MseLossBackward>)
  20366. loss_train_step after backward: tensor(0.0203, device='cuda:0', grad_fn=<MseLossBackward>)
  20367. loss_train: 0.606036888435483
  20368. step: 26
  20369. running loss: 0.02330911109367242
  20370. Train Steps: 26/90 Loss: 0.0233 torch.Size([8, 600, 800])
  20371. torch.Size([8, 8])
  20372. tensor([[0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461],
  20373. [0.6233, 0.4091, 0.8100, 0.2950, 0.3563, 0.3883, 0.6013, 0.5200],
  20374. [0.6260, 0.4214, 0.8538, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983],
  20375. [ nan, nan, 0.7335, 0.2569, 0.3788, 0.2667, 0.5066, 0.5578],
  20376. [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320],
  20377. [0.6205, 0.4016, 0.8350, 0.2717, 0.3987, 0.2550, 0.5787, 0.5133],
  20378. [0.6112, 0.4029, 0.8638, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567],
  20379. [0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412]],
  20380. device='cuda:0', dtype=torch.float64)
  20381. predictions are: tensor([[ 8.0656e-01, -2.6370e-01, 1.5694e+00, -7.0661e-01, -6.7750e-01,
  20382. -6.1102e-01, 4.4837e-01, 1.7196e-01],
  20383. [ 8.3816e-01, -2.6449e-01, 1.4815e+00, -9.9043e-01, -6.0364e-01,
  20384. -6.2400e-01, 7.4873e-01, 1.6229e-01],
  20385. [ 7.6048e-01, -3.0167e-01, 1.4651e+00, 1.3264e-01, -4.7946e-01,
  20386. -7.3808e-01, 4.7905e-01, 5.3412e-01],
  20387. [-2.4250e+00, -2.4390e+00, 1.1478e+00, -1.2399e+00, -4.5536e-01,
  20388. -1.1712e+00, 1.5048e-01, 2.7632e-01],
  20389. [ 9.2886e-01, -2.3300e-01, 1.6681e+00, 3.6170e-02, -4.2982e-01,
  20390. -2.4203e-02, 1.0275e+00, 1.7328e-01],
  20391. [ 7.3659e-01, -3.2591e-01, 1.6391e+00, -1.1892e+00, -4.0055e-01,
  20392. -1.3278e+00, 5.5345e-01, 4.1518e-02],
  20393. [ 7.4341e-01, -3.3253e-01, 1.7754e+00, -2.0403e-01, -2.2468e-03,
  20394. -5.2091e-02, 6.5368e-01, 2.7859e-01],
  20395. [ 7.0976e-01, -3.3546e-01, 1.8470e+00, -7.5960e-01, -4.6156e-01,
  20396. -1.2449e+00, 5.1825e-01, 1.3950e-01]], device='cuda:0',
  20397. grad_fn=<AddmmBackward>)
  20398. landmarks are: tensor([[[ 0.5124, -0.4483, 1.5420, -0.6231, -0.7155, -0.4152, 0.2679,
  20399. 0.2365],
  20400. [ 0.5932, -0.3962, 1.4554, -0.9233, -0.6404, -0.4922, 0.4912,
  20401. 0.1159],
  20402. [ 0.6055, -0.3393, 1.6575, 0.2545, -0.5942, -0.5461, 0.2949,
  20403. 0.4778],
  20404. [-2.2859, -2.2859, 1.1020, -1.0994, -0.5365, -1.0542, 0.0542,
  20405. 0.2905],
  20406. [ 0.6273, -0.4393, 1.6402, 0.1313, -0.5076, 0.0467, 1.1532,
  20407. 0.1715],
  20408. [ 0.5800, -0.4312, 1.5709, -1.0311, -0.4441, -1.1081, 0.3873,
  20409. 0.0851],
  20410. [ 0.5371, -0.4252, 1.7037, -0.0688, -0.0342, 0.0620, 0.3758,
  20411. 0.2853],
  20412. [ 0.5784, -0.4006, 1.7911, -0.5701, -0.5192, -1.0331, 0.4137,
  20413. 0.2139]]], device='cuda:0')
  20414. loss_train_step before backward: tensor(0.0194, device='cuda:0', grad_fn=<MseLossBackward>)
  20415. loss_train_step after backward: tensor(0.0194, device='cuda:0', grad_fn=<MseLossBackward>)
  20416. loss_train: 0.6254050750285387
  20417. step: 27
  20418. running loss: 0.023163150926982914
  20419. Train Steps: 27/90 Loss: 0.0232 torch.Size([8, 600, 800])
  20420. torch.Size([8, 8])
  20421. tensor([[ nan, nan, 0.6900, 0.1917, 0.3937, 0.2367, 0.5240, 0.5246],
  20422. [0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433],
  20423. [0.6084, 0.3981, 0.8588, 0.5233, 0.4600, 0.5367, 0.5680, 0.5006],
  20424. [0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689],
  20425. [0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5787, 0.5600],
  20426. [0.6339, 0.4149, 0.8800, 0.5000, 0.3900, 0.5283, 0.7541, 0.5424],
  20427. [0.6268, 0.4052, 0.8175, 0.2250, 0.4688, 0.1917, 0.6375, 0.5267],
  20428. [0.6117, 0.4019, 0.8538, 0.4067, 0.3513, 0.3583, 0.5663, 0.5133]],
  20429. device='cuda:0', dtype=torch.float64)
  20430. predictions are: tensor([[-2.3912, -2.4006, 1.1279, -1.3163, -0.4300, -1.4060, 0.2300, 0.2982],
  20431. [ 0.8388, -0.2334, 1.3732, -1.0759, -0.6614, -0.8011, 0.6057, 0.2207],
  20432. [ 0.7985, -0.2911, 1.6881, 0.1210, -0.2120, -0.0144, 0.3557, 0.1805],
  20433. [ 0.5698, -0.3962, 1.3641, -0.7297, -0.7064, -0.7499, 0.3289, 0.3488],
  20434. [ 0.6580, -0.3809, 1.7488, -0.0879, -0.1178, -0.1244, 0.4801, 0.3834],
  20435. [ 1.0398, -0.1747, 1.8704, 0.0921, -0.5262, 0.1024, 1.1594, 0.2341],
  20436. [ 0.6474, -0.3854, 1.4655, -1.3019, -0.1727, -1.7691, 0.7337, 0.1894],
  20437. [ 0.5760, -0.4059, 1.7574, -0.5579, -0.7147, -0.9041, 0.3207, 0.1653]],
  20438. device='cuda:0', grad_fn=<AddmmBackward>)
  20439. landmarks are: tensor([[[-2.2859, -2.2859, 0.9012, -1.4006, -0.4672, -1.1928, 0.1342,
  20440. 0.1373],
  20441. [ 0.5665, -0.3963, 1.3457, -0.9926, -0.6058, -0.6308, 0.5721,
  20442. 0.2237],
  20443. [ 0.5241, -0.4470, 1.6806, 0.1313, -0.1612, 0.1929, 0.3378,
  20444. 0.0261],
  20445. [ 0.5327, -0.4381, 1.2880, -0.6308, -0.6866, -0.5307, 0.2658,
  20446. 0.3417],
  20447. [ 0.5505, -0.4207, 1.7095, -0.0534, -0.0509, 0.1050, 0.3873,
  20448. 0.3007],
  20449. [ 0.6421, -0.3695, 1.7788, 0.0236, -0.4845, 0.1544, 1.1971,
  20450. 0.2196],
  20451. [ 0.6092, -0.4143, 1.4901, -1.2467, -0.1208, -1.4006, 0.6587,
  20452. 0.1467],
  20453. [ 0.5393, -0.4294, 1.6575, -0.4075, -0.6635, -0.6308, 0.3296,
  20454. 0.0851]]], device='cuda:0')
  20455. loss_train_step before backward: tensor(0.0175, device='cuda:0', grad_fn=<MseLossBackward>)
  20456. loss_train_step after backward: tensor(0.0175, device='cuda:0', grad_fn=<MseLossBackward>)
  20457. loss_train: 0.6429185438901186
  20458. step: 28
  20459. running loss: 0.022961376567504237
  20460.  
  20461. Train Steps: 28/90 Loss: 0.0230 torch.Size([8, 600, 800])
  20462. torch.Size([8, 8])
  20463. tensor([[0.6228, 0.4119, 0.7938, 0.2233, 0.4674, 0.1773, 0.6188, 0.5433],
  20464. [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116],
  20465. [0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400],
  20466. [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402],
  20467. [0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6262, 0.5167],
  20468. [0.6231, 0.3973, 0.8650, 0.3950, 0.3625, 0.3183, 0.5837, 0.5167],
  20469. [0.6289, 0.4032, 0.8419, 0.5446, 0.4075, 0.5017, 0.6312, 0.5117],
  20470. [0.6151, 0.4125, 0.8738, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483]],
  20471. device='cuda:0', dtype=torch.float64)
  20472. predictions are: tensor([[ 0.3600, -0.5704, 1.3240, -1.3796, -0.1798, -1.6076, 0.5185, 0.2673],
  20473. [ 0.5784, -0.3890, 1.7528, -0.3822, -0.4831, 0.4132, 0.5336, 0.2449],
  20474. [ 0.4459, -0.5053, 1.4762, -1.3180, -0.2036, -1.4446, 0.5637, 0.2456],
  20475. [ 0.2246, -0.7033, 1.5671, 0.0254, -0.4486, -0.2696, 0.4713, 0.3428],
  20476. [ 0.5284, -0.4739, 1.8616, -0.5123, -0.5799, -0.8014, 0.6375, 0.2973],
  20477. [ 0.3499, -0.5930, 1.6678, -0.6075, -0.6182, -0.9334, 0.5179, 0.2470],
  20478. [ 0.6549, -0.4208, 1.5748, 0.0875, -0.4565, -0.0380, 0.7645, 0.2367],
  20479. [ 0.0652, -0.7456, 1.5443, -0.4409, -0.7221, -0.7115, 0.0813, 0.4008]],
  20480. device='cuda:0', grad_fn=<AddmmBackward>)
  20481. landmarks are: tensor([[[ 0.5908, -0.3832, 1.3804, -1.2543, -0.1270, -1.4671, 0.5721,
  20482. 0.2237],
  20483. [ 0.5376, -0.3903, 1.7095, -0.3229, -0.4730, 0.4701, 0.3871,
  20484. 0.0772],
  20485. [ 0.5958, -0.3818, 1.5536, -1.1081, -0.2074, -1.3390, 0.5663,
  20486. 0.2083],
  20487. [ 0.5780, -0.4565, 1.6221, 0.2532, -0.3728, -0.1718, 0.4357,
  20488. 0.2091],
  20489. [ 0.6094, -0.3947, 1.8885, -0.2998, -0.5769, -0.6770, 0.6067,
  20490. 0.1005],
  20491. [ 0.5924, -0.4507, 1.7095, -0.4614, -0.6115, -0.8156, 0.4104,
  20492. 0.1005],
  20493. [ 0.6189, -0.4238, 1.6026, 0.2295, -0.4037, 0.0313, 0.6298,
  20494. 0.0774],
  20495. [ 0.5551, -0.3807, 1.7499, -0.2459, -0.6346, -0.5384, 0.0871,
  20496. 0.2468]]], device='cuda:0')
  20497. loss_train_step before backward: tensor(0.0222, device='cuda:0', grad_fn=<MseLossBackward>)
  20498. loss_train_step after backward: tensor(0.0222, device='cuda:0', grad_fn=<MseLossBackward>)
  20499. loss_train: 0.6651229802519083
  20500. step: 29
  20501. running loss: 0.022935275181100286
  20502. Train Steps: 29/90 Loss: 0.0229 torch.Size([8, 600, 800])
  20503. torch.Size([8, 8])
  20504. tensor([[0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283],
  20505. [0.6136, 0.4085, 0.6688, 0.2317, 0.3862, 0.2367, 0.5517, 0.5783],
  20506. [0.6098, 0.3991, 0.8638, 0.4717, 0.4263, 0.4967, 0.5212, 0.5650],
  20507. [0.6304, 0.4024, 0.8925, 0.4800, 0.3937, 0.4817, 0.7485, 0.5297],
  20508. [0.6156, 0.4125, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084],
  20509. [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726],
  20510. [0.6364, 0.4154, 0.8938, 0.3717, 0.4500, 0.2583, 0.6448, 0.5285],
  20511. [ nan, nan, 0.7525, 0.2291, 0.3838, 0.3017, 0.6050, 0.5667]],
  20512. device='cuda:0', dtype=torch.float64)
  20513. predictions are: tensor([[ 0.6743, -0.3775, 1.9492, -0.5144, -0.6510, -0.6040, 0.7542, 0.2366],
  20514. [ 0.4084, -0.5484, 0.8144, -1.2229, -0.4932, -1.3891, 0.2318, 0.3599],
  20515. [ 0.3393, -0.5796, 1.6478, -0.0665, -0.3251, -0.0515, 0.2467, 0.3376],
  20516. [ 0.7025, -0.3827, 1.7598, -0.0448, -0.4856, -0.1389, 0.9078, 0.2428],
  20517. [ 0.4087, -0.5155, 1.7943, -0.0300, -0.2215, 0.2901, 0.3834, 0.2715],
  20518. [ 0.5483, -0.4365, 1.2142, -1.0301, -0.6042, -1.1357, 0.1894, 0.0886],
  20519. [ 0.5886, -0.3989, 1.8939, -0.6754, -0.2331, -1.3781, 0.6624, 0.3241],
  20520. [-1.2400, -1.6479, 1.1698, -1.2162, -0.5687, -1.1057, 0.4281, 0.4191]],
  20521. device='cuda:0', grad_fn=<AddmmBackward>)
  20522. landmarks are: tensor([[[ 0.6073, -0.4058, 1.8885, -0.4999, -0.5942, -0.4768, 0.6413,
  20523. 0.1544],
  20524. [ 0.5483, -0.3991, 0.8030, -1.2159, -0.5018, -1.1928, 0.2624,
  20525. 0.3852],
  20526. [ 0.5308, -0.4425, 1.7037, -0.1073, -0.3171, 0.0082, 0.1217,
  20527. 0.3238],
  20528. [ 0.6257, -0.4273, 1.8365, -0.0688, -0.4672, -0.0611, 1.1715,
  20529. 0.1608],
  20530. [ 0.5573, -0.3808, 1.8018, -0.0534, -0.1843, 0.4008, 0.3769,
  20531. 0.0622],
  20532. [ 0.4997, -0.4446, 1.1610, -0.9772, -0.6058, -1.0311, 0.1404,
  20533. -0.1031],
  20534. [ 0.6537, -0.3671, 1.8423, -0.5692, -0.2074, -1.0927, 0.6926,
  20535. 0.1554],
  20536. [-2.2859, -2.2859, 1.1898, -1.2278, -0.5134, -0.8925, 0.5085,
  20537. 0.3315]]], device='cuda:0')
  20538. loss_train_step before backward: tensor(0.0337, device='cuda:0', grad_fn=<MseLossBackward>)
  20539. loss_train_step after backward: tensor(0.0337, device='cuda:0', grad_fn=<MseLossBackward>)
  20540. loss_train: 0.6988462079316378
  20541. step: 30
  20542. running loss: 0.02329487359772126
  20543. Train Steps: 30/90 Loss: 0.0233 torch.Size([8, 600, 800])
  20544. torch.Size([8, 8])
  20545. tensor([[0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986],
  20546. [0.6145, 0.4007, 0.8775, 0.4533, 0.4562, 0.5533, 0.6088, 0.5533],
  20547. [0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400],
  20548. [0.6127, 0.4084, 0.8700, 0.4467, 0.3987, 0.4317, 0.5013, 0.5471],
  20549. [0.6114, 0.4018, 0.7213, 0.1967, 0.3763, 0.2700, 0.5875, 0.5533],
  20550. [0.6087, 0.3976, 0.8337, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300],
  20551. [0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461],
  20552. [0.6304, 0.4024, 0.8925, 0.4800, 0.3937, 0.4817, 0.7485, 0.5297]],
  20553. device='cuda:0', dtype=torch.float64)
  20554. predictions are: tensor([[ 0.6474, -0.3811, 1.9037, -0.6287, -0.2022, -1.5291, 0.5346, 0.2287],
  20555. [ 0.5797, -0.4225, 1.7740, -0.1589, -0.1278, 0.2441, 0.5273, 0.3165],
  20556. [ 0.3275, -0.6175, 1.7052, -0.4999, -0.6740, -0.2583, 0.5826, 0.3372],
  20557. [ 0.0698, -0.7592, 1.6376, -0.3294, -0.4746, -0.4258, 0.0776, 0.3240],
  20558. [ 0.0830, -0.7637, 0.9233, -1.3943, -0.5709, -1.1773, 0.2773, 0.2900],
  20559. [ 0.2631, -0.6277, 1.5495, -0.5722, -0.5926, -0.8965, 0.4296, 0.3226],
  20560. [ 0.3554, -0.5687, 1.5380, -0.6126, -0.7502, -0.5382, 0.2942, 0.2498],
  20561. [ 0.5028, -0.5089, 1.7308, -0.0581, -0.4410, -0.0810, 0.8824, 0.2342]],
  20562. device='cuda:0', grad_fn=<AddmmBackward>)
  20563. landmarks are: tensor([[[ 0.6054, -0.3641, 1.8711, -0.5153, -0.2305, -1.3005, 0.5948,
  20564. 0.0171],
  20565. [ 0.5526, -0.4351, 1.7672, -0.1920, -0.1785, 0.2699, 0.5259,
  20566. 0.2699],
  20567. [ 0.5778, -0.4393, 1.8018, -0.4614, -0.6693, -0.1381, 0.5490,
  20568. 0.2083],
  20569. [ 0.5441, -0.3997, 1.7326, -0.2228, -0.4441, -0.2921, 0.0296,
  20570. 0.2409],
  20571. [ 0.5380, -0.4302, 1.0455, -1.3775, -0.5480, -1.0388, 0.4277,
  20572. 0.2699],
  20573. [ 0.5255, -0.4495, 1.5651, -0.4999, -0.5711, -0.8463, 0.4566,
  20574. 0.1621],
  20575. [ 0.5124, -0.4483, 1.5420, -0.6231, -0.7155, -0.4152, 0.2679,
  20576. 0.2365],
  20577. [ 0.6257, -0.4273, 1.8365, -0.0688, -0.4672, -0.0611, 1.1715,
  20578. 0.1608]]], device='cuda:0')
  20579. loss_train_step before backward: tensor(0.0212, device='cuda:0', grad_fn=<MseLossBackward>)
  20580. loss_train_step after backward: tensor(0.0212, device='cuda:0', grad_fn=<MseLossBackward>)
  20581. loss_train: 0.7200348284095526
  20582. step: 31
  20583. running loss: 0.023226929948695244
  20584. Train Steps: 31/90 Loss: 0.0232 torch.Size([8, 600, 800])
  20585. torch.Size([8, 8])
  20586. tensor([[0.6274, 0.4003, 0.8638, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661],
  20587. [0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6262, 0.5167],
  20588. [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5513, 0.5750],
  20589. [0.6113, 0.4088, 0.6859, 0.2208, 0.4363, 0.1700, 0.5188, 0.5533],
  20590. [0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058],
  20591. [0.6193, 0.4050, 0.7313, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656],
  20592. [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118],
  20593. [0.6216, 0.4099, 0.7225, 0.2033, 0.4188, 0.2217, 0.5975, 0.5283]],
  20594. device='cuda:0', dtype=torch.float64)
  20595. predictions are: tensor([[ 0.6033, -0.4310, 1.8310, 0.4091, -0.5672, 0.0558, 0.7620, 0.1504],
  20596. [ 0.8779, -0.2438, 2.1329, -0.3252, -0.5712, -0.5649, 0.7217, 0.2011],
  20597. [-1.6352, -1.8747, 0.9210, -1.2440, -0.4025, -1.2805, 0.1456, 0.4208],
  20598. [-0.6913, -1.2824, 1.0063, -1.2538, -0.3921, -1.3718, 0.0689, 0.3489],
  20599. [ 0.7124, -0.3332, 1.8602, -0.3045, -0.5819, -0.0313, 0.3551, 0.0675],
  20600. [ 0.4150, -0.5009, 1.2186, -1.2239, -0.4369, -1.1337, 0.3174, 0.2731],
  20601. [ 0.6908, -0.3370, 1.6717, 0.3058, -0.4995, -0.0272, 0.5719, 0.5020],
  20602. [ 0.5956, -0.4108, 1.1865, -1.2724, -0.4142, -1.2213, 0.2497, 0.2512]],
  20603. device='cuda:0', grad_fn=<AddmmBackward>)
  20604. landmarks are: tensor([[[ 0.6120, -0.4371, 1.7037, 0.4701, -0.5827, -0.0226, 0.5354,
  20605. -0.1331],
  20606. [ 0.6094, -0.3947, 1.8885, -0.2998, -0.5769, -0.6770, 0.6067,
  20607. 0.1005],
  20608. [ 0.5663, -0.4396, 0.8261, -1.1312, -0.2940, -1.3929, 0.2603,
  20609. 0.3700],
  20610. [ 0.5377, -0.3978, 0.8824, -1.2663, -0.2709, -1.5007, 0.1102,
  20611. 0.2699],
  20612. [ 0.5245, -0.4361, 1.5940, -0.2921, -0.5480, -0.0919, 0.2432,
  20613. 0.0502],
  20614. [ 0.5746, -0.4153, 1.0917, -1.1620, -0.4037, -1.3082, 0.3234,
  20615. 0.3267],
  20616. [ 0.6121, -0.3844, 1.4556, 0.3936, -0.4383, -0.1689, 0.2925,
  20617. 0.5401],
  20618. [ 0.5850, -0.3925, 1.0513, -1.3467, -0.3517, -1.2620, 0.4739,
  20619. 0.1544]]], device='cuda:0')
  20620. loss_train_step before backward: tensor(0.1585, device='cuda:0', grad_fn=<MseLossBackward>)
  20621. loss_train_step after backward: tensor(0.1585, device='cuda:0', grad_fn=<MseLossBackward>)
  20622. loss_train: 0.8785099294036627
  20623. step: 32
  20624. running loss: 0.02745343529386446
  20625.  
  20626. Train Steps: 32/90 Loss: 0.0275 torch.Size([8, 600, 800])
  20627. torch.Size([8, 8])
  20628. tensor([[0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166],
  20629. [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383],
  20630. [0.6068, 0.3963, 0.8650, 0.4317, 0.4037, 0.5083, 0.5253, 0.4999],
  20631. [0.6142, 0.4127, 0.7575, 0.3067, 0.3438, 0.4383, 0.5778, 0.5207],
  20632. [0.6151, 0.4125, 0.8738, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483],
  20633. [0.6127, 0.4084, 0.8700, 0.4467, 0.3987, 0.4317, 0.5013, 0.5471],
  20634. [0.6286, 0.4055, 0.9000, 0.4717, 0.3763, 0.4683, 0.7018, 0.5494],
  20635. [0.6332, 0.4118, 0.9238, 0.4267, 0.4012, 0.4733, 0.7525, 0.5436]],
  20636. device='cuda:0', dtype=torch.float64)
  20637. predictions are: tensor([[ 5.2940e-01, -4.4512e-01, 1.6529e+00, 1.7090e-01, -2.1387e-01,
  20638. 7.7426e-03, 6.9037e-01, 1.7068e-01],
  20639. [ 3.4147e-01, -5.6399e-01, 1.5299e+00, -8.4593e-01, -6.4986e-01,
  20640. -7.0577e-01, 4.4995e-01, 2.9445e-01],
  20641. [-1.7507e-03, -7.9282e-01, 1.6644e+00, -4.0141e-01, -4.2820e-01,
  20642. -1.7175e-01, 1.8599e-01, 1.2021e-01],
  20643. [ 2.9183e-01, -5.5249e-01, 1.0889e+00, -9.9410e-01, -6.0846e-01,
  20644. -4.8721e-01, 2.1527e-01, 3.2583e-01],
  20645. [ 3.6955e-01, -5.1423e-01, 1.5982e+00, -2.9987e-01, -6.4598e-01,
  20646. -8.5654e-01, -3.9592e-02, 2.8502e-01],
  20647. [ 3.0083e-01, -5.7450e-01, 1.6468e+00, -3.0167e-01, -4.3349e-01,
  20648. -5.5919e-01, -3.3376e-02, 2.7009e-01],
  20649. [ 5.5085e-01, -4.4781e-01, 1.8877e+00, -1.4398e-01, -5.1621e-01,
  20650. -3.5603e-01, 8.2913e-01, 1.9243e-01],
  20651. [ 4.9237e-01, -4.7212e-01, 1.7585e+00, -3.4924e-01, -4.0652e-01,
  20652. -2.1576e-01, 8.1910e-01, 3.0801e-01]], device='cuda:0',
  20653. grad_fn=<AddmmBackward>)
  20654. landmarks are: tensor([[[ 0.6042, -0.4273, 1.6920, 0.1860, -0.2717, 0.1406, 0.7997,
  20655. 0.1004],
  20656. [ 0.5779, -0.4275, 1.5894, -0.8362, -0.6577, -0.5153, 0.5605,
  20657. 0.2006],
  20658. [ 0.5168, -0.4556, 1.7095, -0.2921, -0.4210, 0.0620, 0.1404,
  20659. 0.0231],
  20660. [ 0.5509, -0.3798, 1.2129, -0.8694, -0.6982, -0.2613, 0.3830,
  20661. 0.1193],
  20662. [ 0.5551, -0.3807, 1.7499, -0.2459, -0.6346, -0.5384, 0.0871,
  20663. 0.2468],
  20664. [ 0.5441, -0.3997, 1.7326, -0.2228, -0.4441, -0.2921, 0.0296,
  20665. 0.2409],
  20666. [ 0.6174, -0.4129, 1.8711, -0.1073, -0.5480, -0.1227, 0.9558,
  20667. 0.2516],
  20668. [ 0.6388, -0.3840, 1.9808, -0.3152, -0.4326, -0.0996, 1.1897,
  20669. 0.2249]]], device='cuda:0')
  20670. loss_train_step before backward: tensor(0.0250, device='cuda:0', grad_fn=<MseLossBackward>)
  20671. loss_train_step after backward: tensor(0.0250, device='cuda:0', grad_fn=<MseLossBackward>)
  20672. loss_train: 0.9034913182258606
  20673. step: 33
  20674. running loss: 0.02737852479472305
  20675. Train Steps: 33/90 Loss: 0.0274 torch.Size([8, 600, 800])
  20676. torch.Size([8, 8])
  20677. tensor([[ nan, nan, 0.7512, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617],
  20678. [0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6138, 0.5450],
  20679. [0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367],
  20680. [0.6127, 0.4066, 0.8550, 0.5567, 0.4662, 0.5141, 0.5070, 0.5412],
  20681. [0.6115, 0.4005, 0.8838, 0.3867, 0.3763, 0.4700, 0.5800, 0.5550],
  20682. [0.6263, 0.4233, 0.7924, 0.4626, 0.3788, 0.2883, 0.5573, 0.6047],
  20683. [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332],
  20684. [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767]],
  20685. device='cuda:0', dtype=torch.float64)
  20686. predictions are: tensor([[-1.9099, -2.0520, 0.9409, -1.3096, -0.3855, -1.2464, 0.1902, 0.2886],
  20687. [ 0.5154, -0.4495, 1.7163, -0.2839, -0.3076, 0.1952, 0.4860, 0.1330],
  20688. [ 0.5209, -0.4609, 1.6231, -1.0219, -0.5351, -1.0031, 0.5578, 0.1162],
  20689. [ 0.5516, -0.4181, 1.5766, 0.2001, -0.1115, -0.0064, 0.2084, 0.1623],
  20690. [ 0.4060, -0.5142, 1.6260, -0.6436, -0.6285, -0.1200, 0.2998, 0.1566],
  20691. [ 0.7592, -0.2388, 1.3822, -0.2121, -0.4769, -0.7826, 0.2621, 0.4588],
  20692. [ 0.6406, -0.4045, 1.7779, -0.2040, -0.5504, -0.4587, 0.7782, 0.1341],
  20693. [ 0.6576, -0.3083, 1.6359, -0.1900, -0.5963, -0.8407, 0.2722, 0.2160]],
  20694. device='cuda:0', grad_fn=<AddmmBackward>)
  20695. landmarks are: tensor([[[-2.2859, -2.2859, 1.1841, -1.3082, -0.3055, -1.3621, 0.3007,
  20696. 0.3084],
  20697. [ 0.5960, -0.4102, 1.8018, -0.1612, -0.3344, 0.1159, 0.5490,
  20698. 0.2314],
  20699. [ 0.6174, -0.4201, 1.7309, -0.8784, -0.4735, -0.9524, 0.6242,
  20700. 0.1931],
  20701. [ 0.5443, -0.4079, 1.6633, 0.2853, -0.1325, 0.0888, 0.0558,
  20702. 0.2138],
  20703. [ 0.5384, -0.4361, 1.7961, -0.4999, -0.5480, -0.1150, 0.3931,
  20704. 0.2776],
  20705. [ 0.6069, -0.3309, 1.3742, -0.1493, -0.5365, -0.9541, 0.2884,
  20706. 0.5071],
  20707. [ 0.6520, -0.4056, 1.9173, -0.0765, -0.5596, -0.4537, 0.7949,
  20708. 0.1768],
  20709. [ 0.5781, -0.3848, 1.7441, -0.0996, -0.5769, -0.8002, 0.2021,
  20710. 0.3778]]], device='cuda:0')
  20711. loss_train_step before backward: tensor(0.0114, device='cuda:0', grad_fn=<MseLossBackward>)
  20712. loss_train_step after backward: tensor(0.0114, device='cuda:0', grad_fn=<MseLossBackward>)
  20713. loss_train: 0.9148482903838158
  20714. step: 34
  20715. running loss: 0.026907302658347523
  20716. Train Steps: 34/90 Loss: 0.0269 torch.Size([8, 600, 800])
  20717. torch.Size([8, 8])
  20718. tensor([[0.6346, 0.4086, 0.7938, 0.5500, 0.3962, 0.4867, 0.7343, 0.5702],
  20719. [0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385],
  20720. [0.6128, 0.4084, 0.8738, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397],
  20721. [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092],
  20722. [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600],
  20723. [0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901],
  20724. [0.6201, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044],
  20725. [0.6332, 0.4118, 0.9238, 0.4267, 0.4012, 0.4733, 0.7525, 0.5436]],
  20726. device='cuda:0', dtype=torch.float64)
  20727. predictions are: tensor([[ 3.4211e-01, -5.5436e-01, 1.4170e+00, 6.1026e-02, -4.8594e-01,
  20728. -1.1325e-01, 7.7582e-01, 3.2171e-01],
  20729. [ 1.4285e-01, -6.6580e-01, 1.4183e+00, -1.1495e+00, -3.2002e-01,
  20730. -1.2133e+00, 6.1003e-01, 2.7405e-01],
  20731. [ 3.6311e-01, -4.9852e-01, 1.6690e+00, -1.9997e-01, -6.7210e-01,
  20732. -5.8863e-01, -4.3973e-02, 1.5990e-01],
  20733. [ 4.6845e-01, -4.8369e-01, 1.6665e+00, -1.5795e-02, -4.8026e-01,
  20734. -1.5836e-03, 2.7498e-01, 9.0587e-02],
  20735. [ 6.5483e-01, -3.1392e-01, 1.6959e+00, 2.6545e-01, -4.1626e-01,
  20736. -8.7995e-02, 4.2136e-01, 2.3797e-01],
  20737. [ 1.4710e-01, -6.7734e-01, 1.7470e+00, -3.3103e-01, -4.3827e-01,
  20738. -2.9471e-01, -9.0514e-03, 3.3535e-02],
  20739. [ 3.4537e-01, -5.5135e-01, 1.1529e+00, -1.3529e+00, -3.9643e-01,
  20740. -1.2332e+00, 2.8598e-01, 1.4165e-01],
  20741. [ 3.7841e-01, -5.3315e-01, 1.7237e+00, -4.0560e-01, -4.5712e-01,
  20742. -7.7054e-02, 7.7512e-01, 2.6182e-01]], device='cuda:0',
  20743. grad_fn=<AddmmBackward>)
  20744. landmarks are: tensor([[[ 0.6454, -0.3984, 1.3804, 0.2545, -0.4557, -0.0380, 1.1057,
  20745. 0.3478],
  20746. [ 0.6174, -0.3936, 1.4586, -1.1709, -0.2420, -1.1389, 0.8296,
  20747. 0.2012],
  20748. [ 0.5443, -0.3994, 1.7499, -0.1227, -0.6173, -0.5769, 0.0049,
  20749. 0.2071],
  20750. [ 0.5783, -0.4363, 1.7724, 0.0432, -0.5153, 0.0871, 0.4840,
  20751. 0.0663],
  20752. [ 0.5883, -0.3594, 1.7557, 0.2545, -0.4152, -0.0611, 0.3353,
  20753. 0.3007],
  20754. [ 0.5663, -0.3955, 1.7788, -0.2382, -0.4037, -0.2690, 0.0828,
  20755. -0.0220],
  20756. [ 0.5785, -0.4129, 1.2785, -1.3996, -0.3227, -1.3259, 0.4258,
  20757. 0.0438],
  20758. [ 0.6388, -0.3840, 1.9808, -0.3152, -0.4326, -0.0996, 1.1897,
  20759. 0.2249]]], device='cuda:0')
  20760. loss_train_step before backward: tensor(0.0243, device='cuda:0', grad_fn=<MseLossBackward>)
  20761. loss_train_step after backward: tensor(0.0243, device='cuda:0', grad_fn=<MseLossBackward>)
  20762. loss_train: 0.9391738977283239
  20763. step: 35
  20764. running loss: 0.02683353993509497
  20765. Train Steps: 35/90 Loss: 0.0268 torch.Size([8, 600, 800])
  20766. torch.Size([8, 8])
  20767. tensor([[0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222],
  20768. [0.6268, 0.4061, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300],
  20769. [0.6111, 0.3995, 0.8788, 0.4567, 0.3813, 0.4833, 0.5450, 0.5700],
  20770. [0.6234, 0.4179, 0.7825, 0.3450, 0.3813, 0.2867, 0.5675, 0.5617],
  20771. [ nan, nan, 0.7240, 0.2722, 0.3900, 0.2567, 0.5168, 0.5933],
  20772. [ nan, nan, 0.7515, 0.2708, 0.3987, 0.2267, 0.5162, 0.5567],
  20773. [0.6228, 0.4004, 0.8750, 0.5250, 0.3825, 0.5233, 0.6362, 0.5000],
  20774. [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400]],
  20775. device='cuda:0', dtype=torch.float64)
  20776. predictions are: tensor([[ 0.9853, -0.0791, 1.4641, -0.3702, -0.1446, -1.0620, 0.2074, 0.3532],
  20777. [ 0.7470, -0.2884, 1.5556, -0.9633, -0.2606, -1.1213, 0.5230, 0.1062],
  20778. [ 0.7365, -0.3068, 1.7563, -0.0710, -0.4981, -0.0292, 0.3810, 0.1003],
  20779. [ 0.8464, -0.1846, 1.4369, -0.5832, -0.4998, -0.7952, 0.3272, 0.2525],
  20780. [-1.5989, -1.7973, 0.9184, -0.9933, -0.5291, -0.9933, 0.0321, 0.2860],
  20781. [-1.6922, -1.8783, 1.0217, -1.0603, -0.4737, -1.0298, 0.0809, 0.2151],
  20782. [ 0.7276, -0.3266, 1.7957, 0.2800, -0.4631, 0.1283, 0.7769, -0.0928],
  20783. [ 0.7260, -0.2884, 1.6461, -0.5658, -0.5884, -0.0756, 0.7037, 0.1972]],
  20784. device='cuda:0', grad_fn=<AddmmBackward>)
  20785. landmarks are: tensor([[[ 0.6237, -0.2983, 1.3919, -0.4691, -0.0457, -1.2313, 0.2453,
  20786. 0.5882],
  20787. [ 0.6093, -0.4104, 1.5709, -1.1620, -0.1727, -1.2313, 0.6471,
  20788. 0.1621],
  20789. [ 0.5368, -0.4406, 1.7730, -0.1766, -0.5249, -0.0534, 0.2314,
  20790. 0.3469],
  20791. [ 0.5935, -0.3558, 1.3284, -0.6924, -0.5249, -0.9618, 0.3353,
  20792. 0.3084],
  20793. [-2.2859, -2.2859, 1.0580, -1.0288, -0.4845, -1.1004, 0.1011,
  20794. 0.4543],
  20795. [-2.2859, -2.2859, 1.1854, -1.0352, -0.4441, -1.2390, 0.0986,
  20796. 0.2853],
  20797. [ 0.5908, -0.4366, 1.7557, 0.1390, -0.5192, 0.1313, 0.6529,
  20798. 0.0236],
  20799. [ 0.5631, -0.4008, 1.6113, -0.7309, -0.6693, -0.1304, 0.5836,
  20800. 0.2083]]], device='cuda:0')
  20801. loss_train_step before backward: tensor(0.0352, device='cuda:0', grad_fn=<MseLossBackward>)
  20802. loss_train_step after backward: tensor(0.0352, device='cuda:0', grad_fn=<MseLossBackward>)
  20803. loss_train: 0.9743791241198778
  20804. step: 36
  20805. running loss: 0.027066086781107716
  20806.  
  20807. Train Steps: 36/90 Loss: 0.0271 torch.Size([8, 600, 800])
  20808. torch.Size([8, 8])
  20809. tensor([[0.6203, 0.4021, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405],
  20810. [0.6268, 0.4102, 0.8938, 0.3667, 0.4025, 0.2833, 0.6275, 0.5183],
  20811. [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450],
  20812. [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833],
  20813. [0.6256, 0.4199, 0.8638, 0.5800, 0.3987, 0.4383, 0.5600, 0.5950],
  20814. [0.6263, 0.4065, 0.9038, 0.4317, 0.3588, 0.4550, 0.6325, 0.5250],
  20815. [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6088, 0.5583],
  20816. [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749]],
  20817. device='cuda:0', dtype=torch.float64)
  20818. predictions are: tensor([[ 0.6996, -0.3346, 1.8437, 0.1672, -0.5292, -0.3474, 0.3822, -0.0497],
  20819. [ 0.3493, -0.5383, 1.9132, -0.5951, -0.3706, -0.8605, 0.4897, 0.1431],
  20820. [ 0.2542, -0.6049, 1.8101, -0.1032, -0.4832, -0.2194, 0.4467, 0.1163],
  20821. [-0.0646, -0.8031, 0.9565, -0.8911, -0.5063, -0.8299, 0.2433, 0.3494],
  20822. [ 0.7911, -0.2365, 1.7099, 0.5081, -0.3259, -0.1652, 0.2937, 0.3079],
  20823. [ 0.5108, -0.4289, 1.9848, -0.2359, -0.5005, -0.0401, 0.5715, 0.0395],
  20824. [ 0.3191, -0.5725, 1.0183, -1.2563, -0.4272, -0.9802, 0.4102, 0.2090],
  20825. [ 0.2063, -0.6487, 1.2193, -0.8771, -0.5484, -0.6147, 0.2950, 0.2426]],
  20826. device='cuda:0', grad_fn=<AddmmBackward>)
  20827. landmarks are: tensor([[[ 0.5791, -0.4289, 1.7694, 0.0379, -0.5923, -0.4927, 0.4126,
  20828. 0.2107],
  20829. [ 0.6092, -0.3913, 1.8423, -0.5923, -0.4268, -0.9772, 0.6125,
  20830. 0.1082],
  20831. [ 0.5785, -0.4252, 1.7676, -0.1602, -0.5845, -0.3446, 0.4566,
  20832. 0.2314],
  20833. [ 0.5614, -0.4080, 0.7741, -0.8848, -0.5423, -0.9156, 0.3584,
  20834. 0.4085],
  20835. [ 0.6038, -0.3464, 1.7037, 0.3931, -0.4441, -0.2613, 0.3007,
  20836. 0.4624],
  20837. [ 0.6070, -0.4085, 1.8885, -0.2921, -0.6289, -0.1843, 0.6356,
  20838. 0.1390],
  20839. [ 0.5992, -0.3968, 0.9307, -1.3497, -0.4730, -1.0465, 0.5259,
  20840. 0.2930],
  20841. [ 0.5419, -0.4160, 1.1810, -0.8939, -0.6808, -0.7463, 0.2485,
  20842. 0.3694]]], device='cuda:0')
  20843. loss_train_step before backward: tensor(0.0252, device='cuda:0', grad_fn=<MseLossBackward>)
  20844. loss_train_step after backward: tensor(0.0252, device='cuda:0', grad_fn=<MseLossBackward>)
  20845. loss_train: 0.9995399247854948
  20846. step: 37
  20847. running loss: 0.02701459256177013
  20848. Train Steps: 37/90 Loss: 0.0270 torch.Size([8, 600, 800])
  20849. torch.Size([8, 8])
  20850. tensor([[0.6310, 0.4017, 0.8563, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006],
  20851. [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209],
  20852. [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283],
  20853. [0.6198, 0.4164, 0.8700, 0.5067, 0.4625, 0.5650, 0.5464, 0.5197],
  20854. [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567],
  20855. [0.6204, 0.4110, 0.7913, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367],
  20856. [0.6222, 0.3937, 0.8350, 0.5617, 0.4138, 0.4600, 0.5800, 0.5233],
  20857. [0.6218, 0.4185, 0.7338, 0.2650, 0.4625, 0.1950, 0.5687, 0.5800]],
  20858. device='cuda:0', dtype=torch.float64)
  20859. predictions are: tensor([[ 0.4743, -0.4986, 1.7012, 0.2645, -0.5962, -0.0472, 0.6053, -0.0269],
  20860. [ 0.4422, -0.4338, 1.6112, -0.2641, -0.3141, -0.9310, 0.3228, 0.4357],
  20861. [ 0.4612, -0.4605, 1.7693, -0.4304, -0.6406, -0.4055, 0.3712, 0.1926],
  20862. [ 0.3480, -0.5056, 1.7185, -0.0225, -0.2127, 0.2593, 0.2440, 0.0680],
  20863. [ 0.2740, -0.6009, 1.6386, -0.3549, -0.6436, -0.3869, 0.3592, 0.2311],
  20864. [ 0.6209, -0.3838, 1.4577, -1.1062, -0.4147, -0.9839, 0.5374, 0.0921],
  20865. [ 0.1334, -0.7052, 1.5866, 0.1677, -0.4499, -0.1311, 0.4455, 0.1111],
  20866. [ 0.3481, -0.5082, 1.0693, -1.1434, -0.2080, -1.2564, 0.3309, 0.3095]],
  20867. device='cuda:0', grad_fn=<AddmmBackward>)
  20868. landmarks are: tensor([[[ 0.6286, -0.4303, 1.6691, 0.3931, -0.5827, -0.0919, 0.6359,
  20869. 0.0261],
  20870. [ 0.6174, -0.3118, 1.6402, -0.2074, -0.1958, -1.0927, 0.2267,
  20871. 0.5822],
  20872. [ 0.5880, -0.3887, 1.8423, -0.3306, -0.6231, -0.5230, 0.4046,
  20873. 0.1544],
  20874. [ 0.5770, -0.3624, 1.7326, 0.0543, -0.1497, 0.3238, 0.2378,
  20875. 0.1146],
  20876. [ 0.5432, -0.4336, 1.7095, -0.1766, -0.5942, -0.4845, 0.3007,
  20877. 0.2853],
  20878. [ 0.5796, -0.3878, 1.3688, -1.0542, -0.4095, -1.1312, 0.5894,
  20879. 0.1929],
  20880. [ 0.5880, -0.4676, 1.5709, 0.3084, -0.3748, -0.1612, 0.3931,
  20881. 0.1313],
  20882. [ 0.5862, -0.3530, 1.1032, -1.0619, -0.1497, -1.3852, 0.3411,
  20883. 0.3931]]], device='cuda:0')
  20884. loss_train_step before backward: tensor(0.0145, device='cuda:0', grad_fn=<MseLossBackward>)
  20885. loss_train_step after backward: tensor(0.0145, device='cuda:0', grad_fn=<MseLossBackward>)
  20886. loss_train: 1.0140107097104192
  20887. step: 38
  20888. running loss: 0.026684492360800505
  20889. Train Steps: 38/90 Loss: 0.0267 torch.Size([8, 600, 800])
  20890. torch.Size([8, 8])
  20891. tensor([[0.6230, 0.4152, 0.7588, 0.2283, 0.4012, 0.2883, 0.6200, 0.5767],
  20892. [0.6310, 0.4017, 0.8563, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006],
  20893. [ nan, nan, 0.6412, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700],
  20894. [0.6223, 0.4028, 0.8988, 0.4200, 0.3763, 0.5733, 0.6375, 0.5167],
  20895. [0.6202, 0.4079, 0.8025, 0.2500, 0.3763, 0.3217, 0.6125, 0.5533],
  20896. [ nan, nan, 0.8900, 0.3217, 0.5038, 0.2233, 0.6694, 0.5609],
  20897. [0.6248, 0.4185, 0.8500, 0.5767, 0.4463, 0.4550, 0.5613, 0.5917],
  20898. [0.6260, 0.4153, 0.9000, 0.4533, 0.4025, 0.2633, 0.6223, 0.4967]],
  20899. device='cuda:0', dtype=torch.float64)
  20900. predictions are: tensor([[ 0.8678, -0.2009, 1.2378, -0.9185, -0.5161, -0.8553, 0.4915, 0.3090],
  20901. [ 0.9830, -0.1437, 1.6857, 0.4170, -0.5662, -0.0471, 0.5264, -0.0337],
  20902. [-1.3197, -1.6164, 0.8604, -1.1805, -0.3298, -1.2823, 0.1823, 0.3035],
  20903. [ 0.6939, -0.3165, 1.7800, -0.2526, -0.5378, 0.3783, 0.5597, 0.0969],
  20904. [ 0.5649, -0.3764, 1.3641, -0.9624, -0.5757, -0.6402, 0.3433, 0.2319],
  20905. [-1.4869, -1.6870, 1.6115, -0.8906, -0.0766, -1.0723, 0.6136, 0.3026],
  20906. [ 0.9856, -0.0815, 1.6037, 0.4419, -0.2505, -0.1491, 0.2663, 0.3167],
  20907. [ 0.7317, -0.2633, 1.8326, -0.1786, -0.4594, -0.8693, 0.3603, 0.0575]],
  20908. device='cuda:0', grad_fn=<AddmmBackward>)
  20909. landmarks are: tensor([[[ 0.5915, -0.3682, 1.2187, -1.2313, -0.4326, -0.9541, 0.5778,
  20910. 0.3777],
  20911. [ 0.6286, -0.4303, 1.6691, 0.3931, -0.5827, -0.0919, 0.6359,
  20912. 0.0261],
  20913. [-2.2859, -2.2859, 0.6760, -1.4083, -0.3286, -1.4160, 0.2487,
  20914. 0.3469],
  20915. [ 0.5886, -0.4253, 1.8654, -0.3460, -0.5480, 0.3623, 0.6587,
  20916. 0.1005],
  20917. [ 0.5788, -0.4020, 1.4208, -1.1312, -0.5480, -0.8002, 0.5432,
  20918. 0.2699],
  20919. [-2.2859, -2.2859, 1.8249, -0.8002, 0.0409, -1.2543, 0.8059,
  20920. 0.3050],
  20921. [ 0.5999, -0.3530, 1.6402, 0.3777, -0.2247, -0.1843, 0.3065,
  20922. 0.4470],
  20923. [ 0.6055, -0.3676, 1.8711, -0.1920, -0.4268, -1.0696, 0.5887,
  20924. 0.0081]]], device='cuda:0')
  20925. loss_train_step before backward: tensor(0.0564, device='cuda:0', grad_fn=<MseLossBackward>)
  20926. loss_train_step after backward: tensor(0.0564, device='cuda:0', grad_fn=<MseLossBackward>)
  20927. loss_train: 1.070364617742598
  20928. step: 39
  20929. running loss: 0.027445246608784564
  20930. Train Steps: 39/90 Loss: 0.0274 torch.Size([8, 600, 800])
  20931. torch.Size([8, 8])
  20932. tensor([[0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183],
  20933. [0.6137, 0.4038, 0.8563, 0.4050, 0.3813, 0.2550, 0.5106, 0.4954],
  20934. [ nan, nan, 0.6512, 0.1717, 0.4100, 0.1983, 0.5253, 0.5240],
  20935. [0.6151, 0.4085, 0.8750, 0.4367, 0.3887, 0.4367, 0.5066, 0.5846],
  20936. [ nan, nan, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819],
  20937. [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212],
  20938. [0.6277, 0.4103, 0.8087, 0.5717, 0.4188, 0.4750, 0.5663, 0.6083],
  20939. [0.6246, 0.4028, 0.8738, 0.4867, 0.4088, 0.5667, 0.6362, 0.5200]],
  20940. device='cuda:0', dtype=torch.float64)
  20941. predictions are: tensor([[ 0.5133, -0.4663, 1.8136, 0.0464, -0.6048, -0.3059, 0.4851, 0.1031],
  20942. [ 0.8636, -0.1945, 1.7741, -0.3626, -0.4596, -0.9356, 0.1393, 0.1064],
  20943. [-1.3675, -1.6747, 0.9541, -1.3097, -0.3199, -1.3510, 0.3645, 0.2582],
  20944. [ 0.8342, -0.1883, 1.7027, -0.1683, -0.4789, -0.1760, 0.1873, 0.3101],
  20945. [-0.7556, -1.2731, 0.9561, -0.9565, -0.4540, -1.0619, 0.3168, 0.3183],
  20946. [ 0.6719, -0.3621, 1.5153, -0.9662, -0.3847, -0.8820, 0.7084, 0.1236],
  20947. [ 0.8573, -0.1882, 1.5670, 0.4527, -0.3737, 0.0536, 0.4860, 0.3803],
  20948. [ 0.7862, -0.2566, 1.8451, 0.0945, -0.3737, 0.4714, 0.6448, 0.1194]],
  20949. device='cuda:0', grad_fn=<AddmmBackward>)
  20950. landmarks are: tensor([[[ 5.9440e-01, -4.5427e-01, 1.8018e+00, 8.1601e-03, -6.0577e-01,
  20951. -4.3064e-01, 4.1617e-01, 1.0824e-01],
  20952. [ 5.4850e-01, -4.2094e-01, 1.6691e+00, -4.1524e-01, -5.2494e-01,
  20953. -1.1081e+00, 7.2521e-02, 2.0831e-03],
  20954. [-2.2859e+00, -2.2859e+00, 7.2217e-01, -1.4930e+00, -3.9215e-01,
  20955. -1.3698e+00, 1.4038e-01, 1.3434e-01],
  20956. [ 5.5525e-01, -3.9923e-01, 1.7557e+00, -2.6898e-01, -4.9030e-01,
  20957. -2.6898e-01, 5.4227e-02, 4.1446e-01],
  20958. [-2.2859e+00, -2.2859e+00, 9.4385e-01, -9.9666e-01, -4.6143e-01,
  20959. -1.1851e+00, 2.4679e-01, 4.0188e-01],
  20960. [ 6.1248e-01, -4.3453e-01, 1.4308e+00, -1.1384e+00, -4.2133e-01,
  20961. -1.0031e+00, 7.1897e-01, 1.2136e-01],
  20962. [ 6.1339e-01, -3.9099e-01, 1.4497e+00, 3.5458e-01, -3.5173e-01,
  20963. -9.1917e-02, 3.2956e-01, 5.2394e-01],
  20964. [ 5.9902e-01, -4.2556e-01, 1.7499e+00, -3.8029e-02, -3.9792e-01,
  20965. 3.3149e-01, 6.5289e-01, 1.1594e-01]]], device='cuda:0')
  20966. loss_train_step before backward: tensor(0.0867, device='cuda:0', grad_fn=<MseLossBackward>)
  20967. loss_train_step after backward: tensor(0.0867, device='cuda:0', grad_fn=<MseLossBackward>)
  20968. loss_train: 1.1570822978392243
  20969. step: 40
  20970. running loss: 0.02892705744598061
  20971.  
  20972. Train Steps: 40/90 Loss: 0.0289 torch.Size([8, 600, 800])
  20973. torch.Size([8, 8])
  20974. tensor([[0.6145, 0.4008, 0.8750, 0.5383, 0.3975, 0.4650, 0.5563, 0.5533],
  20975. [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050],
  20976. [ nan, nan, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563],
  20977. [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209],
  20978. [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5837, 0.5500],
  20979. [0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767],
  20980. [0.6346, 0.4165, 0.9138, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471],
  20981. [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398]],
  20982. device='cuda:0', dtype=torch.float64)
  20983. predictions are: tensor([[ 0.6094, -0.3628, 1.5813, 0.1503, -0.3496, -0.1345, 0.4158, 0.2550],
  20984. [ 0.6345, -0.3623, 1.7048, -0.3339, -0.5686, -0.4320, 0.3378, 0.0178],
  20985. [-1.7347, -1.9036, 1.1396, -0.9692, -0.5907, -0.9591, 0.2255, 0.2204],
  20986. [ 0.7350, -0.2342, 1.5293, -0.2934, -0.1893, -1.0755, 0.4044, 0.5002],
  20987. [ 0.5159, -0.4508, 1.6632, -0.3091, -0.3144, 0.0313, 0.3703, 0.1639],
  20988. [ 0.6471, -0.3102, 1.6732, -0.2953, -0.5726, -0.5657, 0.2926, 0.3481],
  20989. [ 0.3940, -0.5209, 1.8598, -0.4337, -0.3804, -0.2356, 0.9161, 0.1679],
  20990. [ 0.8231, -0.2564, 1.5874, 0.1697, -0.5144, -0.3017, 0.5055, 0.0932]],
  20991. device='cuda:0', grad_fn=<AddmmBackward>)
  20992. landmarks are: tensor([[[ 0.5526, -0.4347, 1.7557, 0.2006, -0.4499, -0.1381, 0.2834,
  20993. 0.2699],
  20994. [ 0.5844, -0.4466, 1.8423, -0.1997, -0.5942, -0.3998, 0.4219,
  20995. 0.0467],
  20996. [-2.2859, -2.2859, 1.4006, -0.8105, -0.6115, -0.8232, 0.0419,
  20997. 0.2837],
  20998. [ 0.6174, -0.3118, 1.6402, -0.2074, -0.1958, -1.0927, 0.2267,
  20999. 0.5822],
  21000. [ 0.5604, -0.4620, 1.7961, -0.1997, -0.3229, 0.1082, 0.4104,
  21001. 0.2545],
  21002. [ 0.6058, -0.3216, 1.8423, -0.2536, -0.5885, -0.6000, 0.3353,
  21003. 0.3777],
  21004. [ 0.6454, -0.3623, 1.9346, -0.4460, -0.4961, -0.2921, 1.1642,
  21005. 0.2409],
  21006. [ 0.5786, -0.4463, 1.6655, 0.2216, -0.5115, -0.2675, 0.4236,
  21007. 0.2075]]], device='cuda:0')
  21008. loss_train_step before backward: tensor(0.0186, device='cuda:0', grad_fn=<MseLossBackward>)
  21009. loss_train_step after backward: tensor(0.0186, device='cuda:0', grad_fn=<MseLossBackward>)
  21010. loss_train: 1.175681502558291
  21011. step: 41
  21012. running loss: 0.028675158598982706
  21013. Train Steps: 41/90 Loss: 0.0287 torch.Size([8, 600, 800])
  21014. torch.Size([8, 8])
  21015. tensor([[0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426],
  21016. [0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283],
  21017. [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333],
  21018. [ nan, nan, 0.8888, 0.3100, 0.5262, 0.2817, 0.7145, 0.6003],
  21019. [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672],
  21020. [ nan, nan, 0.7335, 0.2569, 0.3788, 0.2667, 0.5066, 0.5578],
  21021. [0.6353, 0.4128, 0.8488, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540],
  21022. [0.6233, 0.4091, 0.8100, 0.2950, 0.3563, 0.3883, 0.6013, 0.5200]],
  21023. device='cuda:0', dtype=torch.float64)
  21024. predictions are: tensor([[ 1.0867, -0.0981, 1.1628, -0.9375, -0.2822, -1.2437, 0.4180, 0.2015],
  21025. [ 0.9307, -0.1722, 1.8503, -0.2443, -0.6467, -0.3693, 0.4760, 0.2033],
  21026. [ 0.9050, -0.1645, 1.7764, 0.1137, -0.5736, 0.1382, 0.2548, 0.2371],
  21027. [-1.1627, -1.4897, 1.6139, -0.6616, 0.0053, -0.9658, 0.8065, 0.4578],
  21028. [ 1.1114, -0.1139, 1.6950, 0.5100, -0.6381, -0.2213, 0.4249, 0.0464],
  21029. [-1.7602, -1.8987, 1.0080, -0.9613, -0.5353, -1.0438, 0.1584, 0.2675],
  21030. [-0.5237, -1.0626, 1.4513, -1.0196, 0.0815, -1.1053, 0.7325, 0.3350],
  21031. [ 0.8871, -0.2171, 1.4334, -0.6789, -0.6656, -0.4187, 0.4968, 0.2156]],
  21032. device='cuda:0', grad_fn=<AddmmBackward>)
  21033. landmarks are: tensor([[[ 0.5784, -0.4153, 1.2972, -1.2541, -0.2265, -1.4553, 0.4741,
  21034. 0.2203],
  21035. [ 0.6073, -0.4058, 1.8885, -0.4999, -0.5942, -0.4768, 0.6413,
  21036. 0.1544],
  21037. [ 0.6003, -0.3747, 1.8885, -0.1997, -0.5018, -0.0149, 0.5490,
  21038. 0.1775],
  21039. [-2.2859, -2.2859, 1.8192, -0.8540, 0.1448, -0.9849, 1.0143,
  21040. 0.4867],
  21041. [ 0.6128, -0.4375, 1.7788, 0.2699, -0.6346, -0.2536, 0.5463,
  21042. -0.1278],
  21043. [-2.2859, -2.2859, 1.1020, -1.0994, -0.5365, -1.0542, 0.0542,
  21044. 0.2905],
  21045. [ 0.6487, -0.3792, 1.6344, -1.0850, 0.2659, -1.5397, 0.8059,
  21046. 0.2730],
  21047. [ 0.5932, -0.3962, 1.4554, -0.9233, -0.6404, -0.4922, 0.4912,
  21048. 0.1159]]], device='cuda:0')
  21049. loss_train_step before backward: tensor(0.1006, device='cuda:0', grad_fn=<MseLossBackward>)
  21050. loss_train_step after backward: tensor(0.1006, device='cuda:0', grad_fn=<MseLossBackward>)
  21051. loss_train: 1.2762999096885324
  21052. step: 42
  21053. running loss: 0.0303880930878222
  21054. Train Steps: 42/90 Loss: 0.0304 torch.Size([8, 600, 800])
  21055. torch.Size([8, 8])
  21056. tensor([[0.6163, 0.4114, 0.7650, 0.2017, 0.3763, 0.2867, 0.5631, 0.5071],
  21057. [0.6128, 0.4115, 0.7934, 0.3778, 0.3450, 0.4033, 0.5337, 0.5456],
  21058. [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578],
  21059. [ nan, nan, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819],
  21060. [0.6259, 0.4156, 0.8812, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960],
  21061. [0.6204, 0.4007, 0.7838, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167],
  21062. [ nan, nan, 0.7515, 0.2708, 0.3987, 0.2267, 0.5162, 0.5567],
  21063. [0.6202, 0.4064, 0.7879, 0.2179, 0.4567, 0.1725, 0.5955, 0.5478]],
  21064. device='cuda:0', dtype=torch.float64)
  21065. predictions are: tensor([[ 0.5861, -0.4517, 1.3542, -0.8687, -0.6011, -0.7119, 0.5024, 0.1682],
  21066. [ 0.7095, -0.3213, 1.5826, -0.2131, -0.7450, -0.2024, 0.2908, 0.2557],
  21067. [ 0.6829, -0.3612, 1.0833, -1.0107, -0.4194, -1.0598, 0.4923, 0.3245],
  21068. [-1.2285, -1.5793, 1.0233, -0.6672, -0.4874, -0.9407, 0.3913, 0.3521],
  21069. [ 0.8831, -0.2063, 1.9609, -0.4306, -0.1119, -0.9870, 0.5996, 0.1698],
  21070. [ 0.6656, -0.3915, 1.3575, -0.8236, -0.2261, -1.1903, 0.6117, 0.1950],
  21071. [-1.9736, -2.0531, 1.1866, -0.7312, -0.4399, -0.9817, 0.2848, 0.2794],
  21072. [ 0.5745, -0.4375, 1.3779, -0.8132, -0.2105, -1.1064, 0.5409, 0.2714]],
  21073. device='cuda:0', grad_fn=<AddmmBackward>)
  21074. landmarks are: tensor([[[ 0.5606, -0.3856, 1.2476, -1.3544, -0.5480, -0.9618, 0.3152,
  21075. 0.0562],
  21076. [ 0.5444, -0.3852, 1.3786, -0.5409, -0.6924, -0.4229, 0.1791,
  21077. 0.2341],
  21078. [ 0.5232, -0.4189, 0.9358, -1.4006, -0.3921, -1.3698, 0.2555,
  21079. 0.2906],
  21080. [-2.2859, -2.2859, 0.9438, -0.9967, -0.4614, -1.1851, 0.2468,
  21081. 0.4019],
  21082. [ 0.6052, -0.3663, 1.7845, -0.8156, -0.0804, -1.4237, 0.5866,
  21083. 0.0051],
  21084. [ 0.5796, -0.4351, 1.3342, -1.3159, -0.2189, -1.4853, 0.4046,
  21085. 0.1005],
  21086. [-2.2859, -2.2859, 1.1854, -1.0352, -0.4441, -1.2390, 0.0986,
  21087. 0.2853],
  21088. [ 0.5787, -0.4086, 1.3535, -1.2794, -0.1764, -1.4891, 0.4645,
  21089. 0.2442]]], device='cuda:0')
  21090. loss_train_step before backward: tensor(0.0686, device='cuda:0', grad_fn=<MseLossBackward>)
  21091. loss_train_step after backward: tensor(0.0686, device='cuda:0', grad_fn=<MseLossBackward>)
  21092. loss_train: 1.3449399406090379
  21093. step: 43
  21094. running loss: 0.03127767303741948
  21095. Train Steps: 43/90 Loss: 0.0313 torch.Size([8, 600, 800])
  21096. torch.Size([8, 8])
  21097. tensor([[0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901],
  21098. [0.6198, 0.4164, 0.8700, 0.5067, 0.4625, 0.5650, 0.5464, 0.5197],
  21099. [0.6126, 0.3954, 0.8538, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350],
  21100. [0.6299, 0.4008, 0.8450, 0.5350, 0.4213, 0.5000, 0.6350, 0.5100],
  21101. [0.6204, 0.4013, 0.8075, 0.2400, 0.4313, 0.2050, 0.5800, 0.5150],
  21102. [0.6255, 0.4017, 0.8688, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901],
  21103. [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098],
  21104. [ nan, nan, 0.8363, 0.3317, 0.3563, 0.3367, 0.5329, 0.5142]],
  21105. device='cuda:0', dtype=torch.float64)
  21106. predictions are: tensor([[ 0.5280, -0.4528, 1.7985, -0.1869, -0.4356, -0.3924, 0.1520, 0.1324],
  21107. [ 0.4553, -0.4516, 1.6858, 0.0901, -0.1598, 0.0426, 0.3004, 0.2708],
  21108. [ 0.6611, -0.3818, 1.6391, 0.0695, -0.3394, -0.2443, 0.2703, 0.2813],
  21109. [ 0.5227, -0.4709, 1.6064, 0.1950, -0.3658, -0.1808, 0.7545, 0.2764],
  21110. [ 0.2876, -0.6182, 1.3079, -1.2030, -0.2005, -1.5260, 0.6327, 0.2942],
  21111. [ 0.4826, -0.5128, 1.6825, -0.7884, -0.6276, -0.8875, 0.7476, 0.1812],
  21112. [ 0.6377, -0.3845, 1.6858, -0.0881, -0.5576, -0.0028, 0.5573, 0.2380],
  21113. [-1.4403, -1.7426, 1.4023, -0.8333, -0.6012, -1.0315, 0.3907, 0.2474]],
  21114. device='cuda:0', grad_fn=<AddmmBackward>)
  21115. landmarks are: tensor([[[ 0.5663, -0.3955, 1.7788, -0.2382, -0.4037, -0.2690, 0.0828,
  21116. -0.0220],
  21117. [ 0.5770, -0.3624, 1.7326, 0.0543, -0.1497, 0.3238, 0.2378,
  21118. 0.1146],
  21119. [ 0.5435, -0.4597, 1.6575, 0.0159, -0.3229, -0.1150, 0.1877,
  21120. 0.1854],
  21121. [ 0.6236, -0.4344, 1.6171, 0.1852, -0.3402, 0.0236, 0.6471,
  21122. 0.0697],
  21123. [ 0.5796, -0.4326, 1.4439, -1.1774, -0.2940, -1.3390, 0.3931,
  21124. 0.0928],
  21125. [ 0.6031, -0.4307, 1.7268, -0.8002, -0.6058, -0.6462, 0.6442,
  21126. -0.0220],
  21127. [ 0.5466, -0.4706, 1.7198, -0.0903, -0.5712, 0.1261, 0.4733,
  21128. 0.0688],
  21129. [-2.2859, -2.2859, 1.5767, -0.7540, -0.6404, -0.7309, 0.1753,
  21130. 0.0893]]], device='cuda:0')
  21131. loss_train_step before backward: tensor(0.0320, device='cuda:0', grad_fn=<MseLossBackward>)
  21132. loss_train_step after backward: tensor(0.0320, device='cuda:0', grad_fn=<MseLossBackward>)
  21133. loss_train: 1.3768937001004815
  21134. step: 44
  21135. running loss: 0.031293038638647304
  21136.  
  21137. Train Steps: 44/90 Loss: 0.0313 torch.Size([8, 600, 800])
  21138. torch.Size([8, 8])
  21139. tensor([[ nan, nan, 0.8463, 0.2550, 0.5850, 0.2133, 0.7129, 0.6072],
  21140. [0.6212, 0.4171, 0.7875, 0.3633, 0.3813, 0.2933, 0.5675, 0.5700],
  21141. [0.6329, 0.4196, 0.9238, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748],
  21142. [0.6263, 0.4039, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804],
  21143. [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400],
  21144. [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426],
  21145. [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524],
  21146. [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583]],
  21147. device='cuda:0', dtype=torch.float64)
  21148. predictions are: tensor([[-1.9036, -2.0651, 1.5527, -1.0373, 0.3358, -1.2065, 1.1035, 0.3857],
  21149. [ 0.7392, -0.3105, 1.2319, -0.7023, -0.4938, -0.9192, 0.4006, 0.3703],
  21150. [ 0.2073, -0.6865, 1.8889, -0.1462, -0.2306, -1.0921, 0.7502, 0.2691],
  21151. [ 0.3774, -0.5846, 1.7424, -0.3205, -0.6016, -0.1874, 0.4186, -0.0711],
  21152. [ 0.6282, -0.4141, 1.4711, -0.7965, -0.6304, -0.2250, 0.5941, 0.2709],
  21153. [ 0.3742, -0.5652, 1.6887, -0.3577, -0.5015, -0.8006, 0.2380, 0.1651],
  21154. [ 0.6018, -0.4007, 1.6369, 0.0416, -0.4545, -0.3304, 0.0900, 0.1834],
  21155. [ 0.3541, -0.5593, 1.5228, -0.5119, -0.6283, -0.5439, 0.4191, 0.2913]],
  21156. device='cuda:0', grad_fn=<AddmmBackward>)
  21157. landmarks are: tensor([[[-2.2859, -2.2859, 1.6229, -1.1081, 0.4162, -1.3005, 1.0070,
  21158. 0.5188],
  21159. [ 0.5832, -0.3593, 1.3515, -0.6077, -0.5249, -0.9310, 0.3353,
  21160. 0.3469],
  21161. [ 0.6372, -0.3479, 1.9808, -0.1689, -0.3171, -0.9310, 1.1057,
  21162. 0.3692],
  21163. [ 0.6072, -0.4206, 1.8711, -0.2536, -0.6115, -0.1304, 0.6812,
  21164. -0.0670],
  21165. [ 0.5631, -0.4008, 1.6113, -0.7309, -0.6693, -0.1304, 0.5836,
  21166. 0.2083],
  21167. [ 0.5788, -0.4135, 1.8214, -0.2408, -0.6039, -0.7854, 0.4115,
  21168. 0.2203],
  21169. [ 0.5529, -0.3811, 1.7788, -0.0380, -0.5307, -0.2074, 0.0727,
  21170. 0.2657],
  21171. [ 0.5761, -0.3966, 1.6171, -0.4845, -0.6346, -0.4691, 0.4739,
  21172. 0.2930]]], device='cuda:0')
  21173. loss_train_step before backward: tensor(0.0201, device='cuda:0', grad_fn=<MseLossBackward>)
  21174. loss_train_step after backward: tensor(0.0201, device='cuda:0', grad_fn=<MseLossBackward>)
  21175. loss_train: 1.3970262268558145
  21176. step: 45
  21177. running loss: 0.031045027263462543
  21178. Train Steps: 45/90 Loss: 0.0310 torch.Size([8, 600, 800])
  21179. torch.Size([8, 8])
  21180. tensor([[0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5837, 0.5583],
  21181. [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656],
  21182. [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5687, 0.5567],
  21183. [0.6236, 0.4084, 0.7738, 0.2133, 0.3663, 0.3233, 0.5813, 0.5567],
  21184. [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833],
  21185. [ nan, nan, 0.8525, 0.2217, 0.5413, 0.2367, 0.7367, 0.5482],
  21186. [ nan, nan, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737],
  21187. [0.6346, 0.4165, 0.9138, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471]],
  21188. device='cuda:0', dtype=torch.float64)
  21189. predictions are: tensor([[ 0.7793, -0.3316, 1.8120, -0.0791, -0.2213, -0.0297, 0.2753, 0.1217],
  21190. [ 0.7511, -0.3460, 1.6510, -0.1584, -0.7081, -0.2885, 0.1719, 0.1961],
  21191. [ 0.5894, -0.4178, 1.1553, -0.9393, -0.3990, -1.2565, 0.3335, 0.3389],
  21192. [ 0.9679, -0.2385, 1.2764, -1.0876, -0.6151, -0.9484, 0.3991, 0.2344],
  21193. [ 0.7129, -0.3717, 0.9215, -0.8002, -0.5720, -1.0232, 0.1904, 0.3818],
  21194. [-2.0511, -2.0903, 1.6392, -1.0022, 0.1126, -1.3098, 1.0287, 0.2427],
  21195. [-2.1435, -2.1390, 1.6120, -0.9052, 0.1158, -1.2869, 0.8389, 0.2812],
  21196. [ 0.6420, -0.4414, 1.9973, -0.1654, -0.5391, -0.2443, 0.8296, 0.0792]],
  21197. device='cuda:0', grad_fn=<AddmmBackward>)
  21198. landmarks are: tensor([[[ 0.5223, -0.4336, 1.7557, -0.2074, -0.0573, 0.3084, 0.4104,
  21199. 0.2930],
  21200. [ 0.5283, -0.4429, 1.5940, -0.2844, -0.5827, -0.1458, 0.2823,
  21201. 0.3267],
  21202. [ 0.5824, -0.3625, 1.0405, -0.9325, -0.4268, -1.1389, 0.3411,
  21203. 0.2853],
  21204. [ 0.5946, -0.3995, 1.2880, -1.3005, -0.5942, -0.7925, 0.3988,
  21205. 0.2853],
  21206. [ 0.5614, -0.4080, 0.7741, -0.8848, -0.5423, -0.9156, 0.3584,
  21207. 0.4085],
  21208. [-2.2859, -2.2859, 1.6517, -1.2620, 0.2141, -1.1928, 1.1166,
  21209. 0.2463],
  21210. [-2.2859, -2.2859, 1.6979, -1.1081, 0.2487, -1.2697, 1.1020,
  21211. 0.3638],
  21212. [ 0.6454, -0.3623, 1.9346, -0.4460, -0.4961, -0.2921, 1.1642,
  21213. 0.2409]]], device='cuda:0')
  21214. loss_train_step before backward: tensor(0.0219, device='cuda:0', grad_fn=<MseLossBackward>)
  21215. loss_train_step after backward: tensor(0.0219, device='cuda:0', grad_fn=<MseLossBackward>)
  21216. loss_train: 1.4189534233883023
  21217. step: 46
  21218. running loss: 0.030846813551919615
  21219. Train Steps: 46/90 Loss: 0.0308 torch.Size([8, 600, 800])
  21220. torch.Size([8, 8])
  21221. tensor([[0.6176, 0.3911, 0.8738, 0.4217, 0.3488, 0.4033, 0.6025, 0.4817],
  21222. [0.6201, 0.3970, 0.8413, 0.4950, 0.4413, 0.5183, 0.6088, 0.5400],
  21223. [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6038, 0.4833],
  21224. [0.6179, 0.4082, 0.6688, 0.2667, 0.3588, 0.3317, 0.5750, 0.5783],
  21225. [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810],
  21226. [0.6246, 0.4028, 0.8738, 0.4867, 0.4088, 0.5667, 0.6362, 0.5200],
  21227. [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167],
  21228. [0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5737, 0.5350]],
  21229. device='cuda:0', dtype=torch.float64)
  21230. predictions are: tensor([[-0.1889, -0.9780, 1.7402, -0.4658, -0.6240, -0.6522, 0.4429, 0.1077],
  21231. [ 0.0686, -0.7909, 1.6863, -0.0267, -0.1338, -0.0739, 0.4993, 0.3025],
  21232. [ 0.1904, -0.7449, 1.7578, -0.1275, -0.4826, -0.2689, 0.4755, 0.1531],
  21233. [ 0.1659, -0.7163, 0.8696, -1.2486, -0.6219, -0.9096, 0.4073, 0.4252],
  21234. [ 0.7232, -0.3879, 1.6732, -1.1300, -0.3453, -1.2824, 0.7050, 0.0514],
  21235. [ 0.3838, -0.5893, 1.7987, -0.0436, -0.2872, 0.2436, 0.6770, 0.2404],
  21236. [ 0.0072, -0.8378, 1.6215, -0.9220, -0.4324, -1.1284, 0.3920, 0.1993],
  21237. [ 0.2877, -0.6304, 1.7888, -0.1657, -0.3686, -1.1141, 0.4204, 0.3036]],
  21238. device='cuda:0', grad_fn=<AddmmBackward>)
  21239. landmarks are: tensor([[[ 5.6692e-01, -4.7937e-01, 1.7499e+00, -3.3826e-01, -6.7506e-01,
  21240. -4.2294e-01, 4.9700e-01, -6.1124e-02],
  21241. [ 5.7812e-01, -4.5219e-01, 1.5998e+00, 4.6189e-04, -2.4781e-01,
  21242. 1.0824e-01, 5.2587e-01, 2.0831e-01],
  21243. [ 5.7263e-01, -4.7952e-01, 1.7788e+00, -1.4935e-02, -5.7113e-01,
  21244. -6.8822e-02, 5.0277e-01, -5.3426e-02],
  21245. [ 5.6796e-01, -4.0046e-01, 8.0300e-01, -1.0542e+00, -6.2887e-01,
  21246. -7.5396e-01, 3.6998e-01, 3.8537e-01],
  21247. [ 6.0641e-01, -3.9900e-01, 1.6113e+00, -8.3095e-01, -4.2679e-01,
  21248. -1.0696e+00, 6.4212e-01, -6.4044e-02],
  21249. [ 5.9902e-01, -4.2556e-01, 1.7499e+00, -3.8029e-02, -3.9792e-01,
  21250. 3.3149e-01, 6.5289e-01, 1.1594e-01],
  21251. [ 5.4825e-01, -4.2490e-01, 1.5305e+00, -7.3857e-01, -6.1155e-01,
  21252. -8.6944e-01, 3.3533e-01, 1.0054e-01],
  21253. [ 5.9007e-01, -4.0000e-01, 1.8423e+00, -6.8822e-02, -5.3072e-01,
  21254. -9.2333e-01, 3.6420e-01, 1.8522e-01]]], device='cuda:0')
  21255. loss_train_step before backward: tensor(0.0493, device='cuda:0', grad_fn=<MseLossBackward>)
  21256. loss_train_step after backward: tensor(0.0493, device='cuda:0', grad_fn=<MseLossBackward>)
  21257. loss_train: 1.4682444604113698
  21258. step: 47
  21259. running loss: 0.031239243838539783
  21260. Train Steps: 47/90 Loss: 0.0312 torch.Size([8, 600, 800])
  21261. torch.Size([8, 8])
  21262. tensor([[0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600],
  21263. [ nan, nan, 0.7525, 0.2291, 0.3838, 0.3017, 0.6050, 0.5667],
  21264. [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483],
  21265. [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446],
  21266. [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167],
  21267. [0.6053, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168],
  21268. [0.6257, 0.4034, 0.8287, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901],
  21269. [0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317]],
  21270. device='cuda:0', dtype=torch.float64)
  21271. predictions are: tensor([[ 0.8121, -0.2857, 1.8553, -0.0534, -0.1376, 0.1054, 0.5305, 0.2714],
  21272. [-2.7636, -2.6309, 1.3694, -0.9074, -0.3612, -1.1750, 0.5475, 0.2775],
  21273. [ 0.6600, -0.4128, 1.1770, -1.2402, -0.3847, -1.2091, 0.5563, 0.2726],
  21274. [ 0.4320, -0.4988, 1.7527, -0.0355, -0.5599, -0.0758, 0.2460, 0.2057],
  21275. [ 0.6434, -0.4167, 1.8805, 0.2187, -0.4504, 0.0262, 0.4597, 0.1198],
  21276. [ 0.0658, -0.7735, 0.9937, -1.2910, -0.3749, -1.4518, 0.4108, 0.1898],
  21277. [ 0.5898, -0.4576, 1.5912, -1.0806, -0.4569, -1.1443, 0.6867, 0.0462],
  21278. [-0.1678, -0.9171, 1.3631, -1.1337, -0.2762, -1.3283, 0.5493, 0.2750]],
  21279. device='cuda:0', grad_fn=<AddmmBackward>)
  21280. landmarks are: tensor([[[ 0.5652, -0.4325, 1.7388, -0.1429, -0.0688, 0.3469, 0.4970,
  21281. 0.3007],
  21282. [-2.2859, -2.2859, 1.1898, -1.2278, -0.5134, -0.8925, 0.5085,
  21283. 0.3315],
  21284. [ 0.5784, -0.4085, 1.0859, -1.3929, -0.4037, -1.1158, 0.5605,
  21285. 0.2468],
  21286. [ 0.5444, -0.3846, 1.6171, -0.1689, -0.5885, -0.0380, 0.1791,
  21287. 0.2296],
  21288. [ 0.5828, -0.4066, 1.7557, 0.0774, -0.3691, -0.0226, 0.4277,
  21289. 0.1005],
  21290. [ 0.5102, -0.4223, 0.8999, -1.4011, -0.4383, -1.3082, 0.2267,
  21291. 0.1013],
  21292. [ 0.6042, -0.4225, 1.5420, -1.2082, -0.4730, -1.0311, 0.6380,
  21293. -0.0220],
  21294. [ 0.5849, -0.3836, 1.2649, -1.2236, -0.3460, -1.2313, 0.4508,
  21295. 0.1698]]], device='cuda:0')
  21296. loss_train_step before backward: tensor(0.0356, device='cuda:0', grad_fn=<MseLossBackward>)
  21297. loss_train_step after backward: tensor(0.0356, device='cuda:0', grad_fn=<MseLossBackward>)
  21298. loss_train: 1.5038310894742608
  21299. step: 48
  21300. running loss: 0.0313298143640471
  21301.  
  21302. Train Steps: 48/90 Loss: 0.0313 torch.Size([8, 600, 800])
  21303. torch.Size([8, 8])
  21304. tensor([[0.6143, 0.4034, 0.8800, 0.4833, 0.4512, 0.5367, 0.5289, 0.5097],
  21305. [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050],
  21306. [0.6310, 0.4017, 0.8563, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006],
  21307. [0.6113, 0.4088, 0.6859, 0.2208, 0.4363, 0.1700, 0.5188, 0.5533],
  21308. [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750],
  21309. [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869],
  21310. [0.6274, 0.4003, 0.8638, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661],
  21311. [0.6115, 0.3998, 0.7063, 0.2383, 0.4037, 0.1950, 0.5320, 0.4993]],
  21312. device='cuda:0', dtype=torch.float64)
  21313. predictions are: tensor([[ 0.2730, -0.6282, 1.9380, -0.1759, -0.0889, 0.0257, 0.2360, 0.1262],
  21314. [ 0.5148, -0.5094, 1.9247, -0.4342, -0.5837, -0.4938, 0.5429, 0.0917],
  21315. [ 0.3095, -0.6562, 1.8017, 0.1408, -0.6035, -0.1369, 0.7252, 0.1669],
  21316. [-0.0043, -0.8408, 1.0899, -1.4017, -0.2449, -1.5586, 0.4209, 0.3265],
  21317. [ 0.3186, -0.6128, 1.1535, -1.3408, -0.5048, -1.0129, 0.7331, 0.4692],
  21318. [ 0.0875, -0.7587, 1.8255, 0.1383, -0.4714, -0.2924, 0.2298, 0.1498],
  21319. [-0.2003, -0.9530, 1.7477, 0.2234, -0.5687, -0.0849, 0.5988, 0.1860],
  21320. [ 0.1707, -0.7150, 1.1711, -1.5067, -0.3215, -1.4711, 0.4139, 0.1151]],
  21321. device='cuda:0', grad_fn=<AddmmBackward>)
  21322. landmarks are: tensor([[[ 0.5515, -0.4225, 1.7788, -0.0534, -0.2016, 0.1929, 0.1568,
  21323. 0.0682],
  21324. [ 0.5844, -0.4466, 1.8423, -0.1997, -0.5942, -0.3998, 0.4219,
  21325. 0.0467],
  21326. [ 0.6286, -0.4303, 1.6691, 0.3931, -0.5827, -0.0919, 0.6359,
  21327. 0.0261],
  21328. [ 0.5377, -0.3978, 0.8824, -1.2663, -0.2709, -1.5007, 0.1102,
  21329. 0.2699],
  21330. [ 0.5805, -0.3818, 1.0282, -1.1774, -0.4903, -0.9310, 0.5894,
  21331. 0.3700],
  21332. [ 0.5320, -0.4488, 1.6633, 0.3315, -0.5018, -0.2459, 0.0828,
  21333. -0.0370],
  21334. [ 0.6120, -0.4371, 1.7037, 0.4701, -0.5827, -0.0226, 0.5354,
  21335. -0.1331],
  21336. [ 0.5384, -0.4393, 0.9762, -1.1851, -0.4210, -1.3852, 0.1712,
  21337. 0.0201]]], device='cuda:0')
  21338. loss_train_step before backward: tensor(0.0528, device='cuda:0', grad_fn=<MseLossBackward>)
  21339. loss_train_step after backward: tensor(0.0528, device='cuda:0', grad_fn=<MseLossBackward>)
  21340. loss_train: 1.5566692175343633
  21341. step: 49
  21342. running loss: 0.031768759541517616
  21343. Train Steps: 49/90 Loss: 0.0318 torch.Size([8, 600, 800])
  21344. torch.Size([8, 8])
  21345. tensor([[0.6138, 0.4101, 0.8800, 0.5083, 0.4637, 0.5950, 0.5587, 0.5077],
  21346. [0.6267, 0.4080, 0.8438, 0.2633, 0.4763, 0.1800, 0.6259, 0.5240],
  21347. [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383],
  21348. [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100],
  21349. [0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689],
  21350. [ nan, nan, 0.9088, 0.3783, 0.4562, 0.2617, 0.6741, 0.5575],
  21351. [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841],
  21352. [0.6278, 0.4253, 0.8875, 0.5017, 0.4113, 0.2750, 0.5413, 0.6196]],
  21353. device='cuda:0', dtype=torch.float64)
  21354. predictions are: tensor([[ 0.3308, -0.5791, 1.6990, -0.1509, -0.1670, 0.1825, 0.3153, 0.1267],
  21355. [ 0.4098, -0.5453, 1.5226, -1.3711, -0.0952, -1.5752, 0.6124, 0.0944],
  21356. [ 0.1309, -0.7259, 1.6941, -0.4025, -0.3071, 0.3146, 0.4373, 0.1410],
  21357. [ 0.5440, -0.4765, 1.5739, -0.0572, -0.5781, -0.0772, 0.4984, 0.0258],
  21358. [ 0.1205, -0.6980, 1.2362, -0.9301, -0.7810, -0.5665, 0.1186, 0.2302],
  21359. [-1.3172, -1.6646, 1.9684, -0.6178, -0.2131, -1.3039, 0.8897, 0.1631],
  21360. [ 0.5473, -0.4987, 1.7234, -0.6042, -0.4938, -0.6665, 0.9100, 0.1871],
  21361. [ 0.4775, -0.4528, 1.5372, -0.2494, -0.4017, -0.9764, 0.2984, 0.4618]],
  21362. device='cuda:0', grad_fn=<AddmmBackward>)
  21363. landmarks are: tensor([[[ 0.5491, -0.3918, 1.7788, 0.0620, -0.1439, 0.4624, 0.2946,
  21364. 0.0592],
  21365. [ 0.6088, -0.4015, 1.6113, -1.0696, -0.0861, -1.4545, 0.6051,
  21366. 0.1343],
  21367. [ 0.5664, -0.4321, 1.8249, -0.2074, -0.2651, 0.4162, 0.5663,
  21368. 0.2006],
  21369. [ 0.6125, -0.4273, 1.6402, 0.2006, -0.4788, -0.0303, 0.6182,
  21370. 0.0697],
  21371. [ 0.5327, -0.4381, 1.2880, -0.6308, -0.6866, -0.5307, 0.2658,
  21372. 0.3417],
  21373. [-2.2859, -2.2859, 1.9115, -0.5384, -0.1785, -1.0773, 0.8278,
  21374. 0.2890],
  21375. [ 0.6438, -0.3936, 2.0039, -0.2690, -0.2940, -0.6231, 1.0618,
  21376. 0.4119],
  21377. [ 0.6139, -0.3216, 1.8134, 0.0313, -0.3864, -1.0157, 0.2144,
  21378. 0.5762]]], device='cuda:0')
  21379. loss_train_step before backward: tensor(0.0497, device='cuda:0', grad_fn=<MseLossBackward>)
  21380. loss_train_step after backward: tensor(0.0497, device='cuda:0', grad_fn=<MseLossBackward>)
  21381. loss_train: 1.6064054938033223
  21382. step: 50
  21383. running loss: 0.032128109876066444
  21384. Train Steps: 50/90 Loss: 0.0321 torch.Size([8, 600, 800])
  21385. torch.Size([8, 8])
  21386. tensor([[0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5837, 0.5500],
  21387. [0.6277, 0.4013, 0.8888, 0.4767, 0.3600, 0.3567, 0.6148, 0.4672],
  21388. [0.6099, 0.4030, 0.8638, 0.5117, 0.4983, 0.4965, 0.5086, 0.5388],
  21389. [0.6137, 0.4038, 0.8563, 0.4050, 0.3813, 0.2550, 0.5106, 0.4954],
  21390. [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290],
  21391. [0.6307, 0.4029, 0.8988, 0.4817, 0.3937, 0.3500, 0.7311, 0.5378],
  21392. [0.6250, 0.3993, 0.9138, 0.4333, 0.3763, 0.5217, 0.6995, 0.5320],
  21393. [ nan, nan, 0.7512, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617]],
  21394. device='cuda:0', dtype=torch.float64)
  21395. predictions are: tensor([[ 0.4499, -0.5199, 1.6845, -0.3348, -0.3485, 0.1269, 0.3157, 0.2084],
  21396. [ 0.6747, -0.3542, 1.6891, -0.3486, -0.6206, -0.6618, 0.4907, 0.0787],
  21397. [ 0.6204, -0.3971, 1.6820, -0.0836, -0.0091, -0.0460, 0.1695, 0.1800],
  21398. [ 0.6887, -0.3090, 1.6003, -0.6434, -0.4770, -1.1085, 0.1139, 0.1487],
  21399. [ 0.4262, -0.4737, 1.1590, -1.4543, -0.3002, -1.2801, 0.6216, 0.1746],
  21400. [ 0.4292, -0.5376, 1.8991, -0.2834, -0.4674, -0.6994, 0.8851, 0.1040],
  21401. [ 0.4682, -0.5118, 1.7642, -0.4536, -0.5753, 0.2361, 0.9281, 0.1673],
  21402. [-2.8165, -2.6293, 1.1189, -1.1107, -0.3079, -1.3337, 0.3717, 0.2679]],
  21403. device='cuda:0', grad_fn=<AddmmBackward>)
  21404. landmarks are: tensor([[[ 5.6039e-01, -4.6197e-01, 1.7961e+00, -1.9969e-01, -3.2286e-01,
  21405. 1.0824e-01, 4.1039e-01, 2.5450e-01],
  21406. [ 6.1322e-01, -4.3241e-01, 1.8192e+00, -8.4219e-02, -6.2309e-01,
  21407. -6.3849e-01, 5.5366e-01, -1.2778e-01],
  21408. [ 5.3095e-01, -4.2456e-01, 1.7037e+00, 7.7444e-02, 1.5763e-02,
  21409. 7.5237e-03, 6.3480e-02, 2.0256e-01],
  21410. [ 5.4850e-01, -4.2094e-01, 1.6691e+00, -4.1524e-01, -5.2494e-01,
  21411. -1.1081e+00, 7.2521e-02, 2.0831e-03],
  21412. [ 6.1248e-01, -4.2731e-01, 1.2807e+00, -1.3253e+00, -2.5737e-01,
  21413. -1.2542e+00, 6.8644e-01, 1.5750e-01],
  21414. [ 6.2730e-01, -4.2490e-01, 1.8654e+00, -6.1124e-02, -4.6721e-01,
  21415. -6.6928e-01, 1.0910e+00, 1.9818e-01],
  21416. [ 6.0095e-01, -4.4175e-01, 1.9346e+00, -2.8437e-01, -5.4804e-01,
  21417. 1.2363e-01, 9.4481e-01, 1.7146e-01],
  21418. [-2.2859e+00, -2.2859e+00, 1.1841e+00, -1.3082e+00, -3.0554e-01,
  21419. -1.3621e+00, 3.0069e-01, 3.0839e-01]]], device='cuda:0')
  21420. loss_train_step before backward: tensor(0.0178, device='cuda:0', grad_fn=<MseLossBackward>)
  21421. loss_train_step after backward: tensor(0.0178, device='cuda:0', grad_fn=<MseLossBackward>)
  21422. loss_train: 1.6242366721853614
  21423. step: 51
  21424. running loss: 0.03184777788598748
  21425. Train Steps: 51/90 Loss: 0.0318 torch.Size([8, 600, 800])
  21426. torch.Size([8, 8])
  21427. tensor([[ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600],
  21428. [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986],
  21429. [0.6257, 0.4034, 0.8287, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901],
  21430. [0.6144, 0.4032, 0.8563, 0.3283, 0.3525, 0.4200, 0.5775, 0.5583],
  21431. [0.6256, 0.4199, 0.8638, 0.5800, 0.3987, 0.4383, 0.5600, 0.5950],
  21432. [ nan, nan, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552],
  21433. [0.6125, 0.4076, 0.8488, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505],
  21434. [ nan, nan, 0.7268, 0.2333, 0.4125, 0.1933, 0.5112, 0.5383]],
  21435. device='cuda:0', dtype=torch.float64)
  21436. predictions are: tensor([[-1.2281, -1.5360, 1.2552, -0.9789, -0.3538, -0.9091, 0.2718, 0.2497],
  21437. [ 1.1770, 0.0143, 1.8093, -0.5852, -0.1964, -1.1055, 0.6296, 0.0711],
  21438. [ 1.0187, -0.1383, 1.4644, -1.1879, -0.4770, -0.8713, 0.7727, -0.0471],
  21439. [ 1.4524, 0.1545, 1.5627, -0.7437, -0.6868, -0.2045, 0.5216, 0.1345],
  21440. [ 1.1354, -0.0355, 1.5601, 0.3532, -0.4657, -0.0231, 0.3850, 0.3366],
  21441. [-2.4087, -2.3484, 1.7172, -1.0239, 0.0941, -1.0496, 0.9831, 0.2392],
  21442. [ 0.8638, -0.1748, 1.5705, -0.4505, -0.6372, -0.3944, 0.0331, 0.1580],
  21443. [-1.9988, -2.0419, 0.9688, -1.1190, -0.3952, -1.2128, 0.0812, 0.2310]],
  21444. device='cuda:0', grad_fn=<AddmmBackward>)
  21445. landmarks are: tensor([[[-2.2859, -2.2859, 1.4266, -1.1389, -0.4499, -1.1235, 0.2891,
  21446. 0.3007],
  21447. [ 0.6054, -0.3641, 1.8711, -0.5153, -0.2305, -1.3005, 0.5948,
  21448. 0.0171],
  21449. [ 0.6042, -0.4225, 1.5420, -1.2082, -0.4730, -1.0311, 0.6380,
  21450. -0.0220],
  21451. [ 0.5520, -0.4237, 1.6691, -0.7694, -0.6577, -0.3460, 0.3815,
  21452. 0.2930],
  21453. [ 0.6038, -0.3464, 1.7037, 0.3931, -0.4441, -0.2613, 0.3007,
  21454. 0.4624],
  21455. [-2.2859, -2.2859, 1.7557, -1.1466, 0.0871, -1.0773, 1.1239,
  21456. 0.2783],
  21457. [ 0.5433, -0.4032, 1.6344, -0.4922, -0.5769, -0.5846, 0.0357,
  21458. 0.2567],
  21459. [-2.2859, -2.2859, 1.0712, -1.2085, -0.3806, -1.3929, 0.0755,
  21460. 0.2006]]], device='cuda:0')
  21461. loss_train_step before backward: tensor(0.0748, device='cuda:0', grad_fn=<MseLossBackward>)
  21462. loss_train_step after backward: tensor(0.0748, device='cuda:0', grad_fn=<MseLossBackward>)
  21463. loss_train: 1.6990172257646918
  21464. step: 52
  21465. running loss: 0.032673408187782534
  21466.  
  21467. Train Steps: 52/90 Loss: 0.0327 torch.Size([8, 600, 800])
  21468. torch.Size([8, 8])
  21469. tensor([[0.6125, 0.4076, 0.8488, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505],
  21470. [0.6110, 0.3984, 0.8750, 0.4933, 0.4625, 0.4950, 0.5578, 0.5676],
  21471. [0.6148, 0.4076, 0.8666, 0.4820, 0.4138, 0.5067, 0.5250, 0.5767],
  21472. [0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033],
  21473. [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333],
  21474. [0.6212, 0.4171, 0.7875, 0.3633, 0.3813, 0.2933, 0.5675, 0.5700],
  21475. [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749],
  21476. [0.6153, 0.4119, 0.8463, 0.3833, 0.3600, 0.3200, 0.5106, 0.5563]],
  21477. device='cuda:0', dtype=torch.float64)
  21478. predictions are: tensor([[-0.1634, -0.8578, 1.6840, -0.5581, -0.5848, -0.7154, 0.0890, 0.1144],
  21479. [ 0.4708, -0.4999, 1.8414, -0.1520, -0.0150, -0.1076, 0.5225, 0.1580],
  21480. [ 0.2425, -0.6167, 1.7604, -0.0780, -0.2799, 0.0465, 0.4614, 0.2296],
  21481. [ 0.4320, -0.5397, 1.8275, -0.0124, -0.2925, 0.0979, 0.6848, 0.0197],
  21482. [ 0.1199, -0.6968, 1.9107, -0.3066, -0.4233, -0.0754, 0.6837, 0.0897],
  21483. [ 0.6303, -0.3232, 1.3665, -0.7849, -0.5192, -0.9144, 0.5360, 0.3355],
  21484. [ 0.5060, -0.4484, 1.1837, -1.1058, -0.6574, -0.7888, 0.4394, 0.1935],
  21485. [-0.2262, -0.9020, 1.6467, -0.6310, -0.5496, -0.9648, 0.1842, 0.1176]],
  21486. device='cuda:0', grad_fn=<AddmmBackward>)
  21487. landmarks are: tensor([[[ 5.4331e-01, -4.0323e-01, 1.6344e+00, -4.9222e-01, -5.7691e-01,
  21488. -5.8460e-01, 3.5720e-02, 2.5666e-01],
  21489. [ 5.3637e-01, -4.4573e-01, 1.7557e+00, -7.2363e-03, -1.4965e-01,
  21490. 4.6189e-04, 2.9049e-01, 3.3573e-01],
  21491. [ 5.5393e-01, -4.0308e-01, 1.7168e+00, -5.9398e-02, -3.7483e-01,
  21492. 5.4350e-02, 1.3903e-01, 3.7768e-01],
  21493. [ 5.6680e-01, -4.3056e-01, 1.7730e+00, 6.9746e-02, -4.0370e-01,
  21494. 1.3903e-01, 4.4503e-01, 3.8953e-02],
  21495. [ 6.0035e-01, -3.7467e-01, 1.8885e+00, -1.9969e-01, -5.0185e-01,
  21496. -1.4935e-02, 5.4896e-01, 1.7752e-01],
  21497. [ 5.8320e-01, -3.5928e-01, 1.3515e+00, -6.0770e-01, -5.2494e-01,
  21498. -9.3102e-01, 3.3533e-01, 3.4688e-01],
  21499. [ 5.4186e-01, -4.1601e-01, 1.1810e+00, -8.9394e-01, -6.8083e-01,
  21500. -7.4627e-01, 2.4855e-01, 3.6938e-01],
  21501. [ 5.5606e-01, -3.8337e-01, 1.6229e+00, -5.1532e-01, -6.2309e-01,
  21502. -8.0785e-01, 7.2734e-02, 2.8371e-01]]], device='cuda:0')
  21503. loss_train_step before backward: tensor(0.0443, device='cuda:0', grad_fn=<MseLossBackward>)
  21504. loss_train_step after backward: tensor(0.0443, device='cuda:0', grad_fn=<MseLossBackward>)
  21505. loss_train: 1.7433188473805785
  21506. step: 53
  21507. running loss: 0.03289280844114299
  21508. Train Steps: 53/90 Loss: 0.0329 torch.Size([8, 600, 800])
  21509. torch.Size([8, 8])
  21510. tensor([[0.6091, 0.3997, 0.8314, 0.4334, 0.3788, 0.4550, 0.5213, 0.5656],
  21511. [0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240],
  21512. [0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494],
  21513. [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600],
  21514. [0.6142, 0.3982, 0.8650, 0.4883, 0.3912, 0.4317, 0.5315, 0.5350],
  21515. [0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835],
  21516. [0.6273, 0.4143, 0.8750, 0.5700, 0.3987, 0.4717, 0.6013, 0.5467],
  21517. [0.6250, 0.3993, 0.9138, 0.4333, 0.3763, 0.5217, 0.6995, 0.5320]],
  21518. device='cuda:0', dtype=torch.float64)
  21519. predictions are: tensor([[ 0.4969, -0.4439, 1.6245, -0.3381, -0.5079, -0.0939, 0.2119, 0.2477],
  21520. [-0.6186, -1.1474, 1.1579, -1.4484, -0.2859, -1.4703, 0.1573, 0.1548],
  21521. [ 0.0448, -0.7621, 1.7929, 0.1244, -0.4749, -0.2750, 0.8188, 0.1702],
  21522. [ 0.3577, -0.5050, 1.1513, -1.2380, -0.5836, -0.7809, 0.4390, 0.2825],
  21523. [ 0.5176, -0.4453, 1.7902, -0.0143, -0.4640, -0.1327, 0.1576, 0.1083],
  21524. [ 0.6601, -0.3388, 1.9378, -0.6587, -0.2862, -1.3708, 0.4818, 0.0354],
  21525. [-0.0375, -0.7674, 1.7133, 0.1232, -0.3605, -0.0799, 0.3183, 0.2255],
  21526. [ 0.5661, -0.4524, 1.9309, -0.3699, -0.4641, 0.2918, 1.0282, 0.1330]],
  21527. device='cuda:0', grad_fn=<AddmmBackward>)
  21528. landmarks are: tensor([[[ 0.5275, -0.4396, 1.5543, -0.2841, -0.5365, -0.1843, 0.1221,
  21529. 0.3265],
  21530. [ 0.5676, -0.4112, 1.1898, -1.2467, -0.2940, -1.4622, 0.2103,
  21531. 0.1343],
  21532. [ 0.6224, -0.4032, 1.5940, 0.2930, -0.5769, -0.2690, 0.8900,
  21533. 0.2516],
  21534. [ 0.5466, -0.4080, 1.0668, -1.1764, -0.6289, -0.7617, 0.4855,
  21535. 0.3007],
  21536. [ 0.5513, -0.4467, 1.7095, -0.0303, -0.4788, -0.2921, 0.1692,
  21537. 0.1854],
  21538. [ 0.6117, -0.3877, 1.8249, -0.5230, -0.3517, -1.2851, 0.4200,
  21539. -0.0529],
  21540. [ 0.6115, -0.3724, 1.7557, 0.3469, -0.4441, -0.1073, 0.4912,
  21541. 0.2391],
  21542. [ 0.6010, -0.4417, 1.9346, -0.2844, -0.5480, 0.1236, 0.9448,
  21543. 0.1715]]], device='cuda:0')
  21544. loss_train_step before backward: tensor(0.0540, device='cuda:0', grad_fn=<MseLossBackward>)
  21545. loss_train_step after backward: tensor(0.0540, device='cuda:0', grad_fn=<MseLossBackward>)
  21546. loss_train: 1.7973417667672038
  21547. step: 54
  21548. running loss: 0.033284106791985256
  21549. Train Steps: 54/90 Loss: 0.0333 torch.Size([8, 600, 800])
  21550. torch.Size([8, 8])
  21551. tensor([[0.6207, 0.4081, 0.7662, 0.2067, 0.3962, 0.3200, 0.6312, 0.5300],
  21552. [0.6289, 0.4024, 0.9088, 0.4567, 0.3937, 0.5633, 0.7058, 0.5609],
  21553. [0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901],
  21554. [0.6148, 0.3996, 0.8488, 0.3867, 0.3488, 0.4067, 0.5863, 0.5000],
  21555. [0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133],
  21556. [0.6286, 0.4060, 0.9188, 0.4333, 0.3675, 0.4167, 0.7034, 0.5528],
  21557. [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377],
  21558. [ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600]],
  21559. device='cuda:0', dtype=torch.float64)
  21560. predictions are: tensor([[ 0.6193, -0.2979, 1.2371, -1.2785, -0.4778, -0.9031, 0.4626, 0.2020],
  21561. [ 0.6740, -0.3263, 1.8320, -0.1743, -0.3139, 0.2781, 0.8246, 0.2859],
  21562. [ 0.3821, -0.4890, 1.8163, -0.2113, -0.3938, -0.2797, 0.0329, 0.0312],
  21563. [ 0.8386, -0.1654, 1.7004, -0.5014, -0.6979, -0.4120, 0.3056, 0.0958],
  21564. [ 0.6805, -0.3059, 1.7077, 0.1778, -0.4999, -0.0904, 0.5120, 0.1148],
  21565. [ 0.6076, -0.3326, 1.9301, -0.3129, -0.4152, -0.5163, 0.7245, 0.2601],
  21566. [ 0.6592, -0.3361, 1.6481, 0.1911, -0.3928, -0.1796, 0.3788, 0.1469],
  21567. [-2.6177, -2.5506, 1.3313, -1.1060, -0.2942, -1.1176, 0.2407, 0.2369]],
  21568. device='cuda:0', grad_fn=<AddmmBackward>)
  21569. landmarks are: tensor([[[ 0.5809, -0.4011, 1.2533, -1.3313, -0.4557, -0.8079, 0.6298,
  21570. 0.1621],
  21571. [ 0.6191, -0.4273, 1.9115, -0.1766, -0.4672, 0.3161, 0.9741,
  21572. 0.3050],
  21573. [ 0.5663, -0.3955, 1.7788, -0.2382, -0.4037, -0.2690, 0.0828,
  21574. -0.0220],
  21575. [ 0.5538, -0.4400, 1.6344, -0.4999, -0.6751, -0.4075, 0.4219,
  21576. 0.0236],
  21577. [ 0.6148, -0.4130, 1.6864, 0.1698, -0.5307, -0.1150, 0.6125,
  21578. 0.0851],
  21579. [ 0.6174, -0.4105, 1.9577, -0.2844, -0.5885, -0.3614, 0.9631,
  21580. 0.2676],
  21581. [ 0.5775, -0.4607, 1.6741, 0.1962, -0.4036, -0.1212, 0.4588,
  21582. 0.1979],
  21583. [-2.2859, -2.2859, 1.4266, -1.1389, -0.4499, -1.1235, 0.2891,
  21584. 0.3007]]], device='cuda:0')
  21585. loss_train_step before backward: tensor(0.0118, device='cuda:0', grad_fn=<MseLossBackward>)
  21586. loss_train_step after backward: tensor(0.0118, device='cuda:0', grad_fn=<MseLossBackward>)
  21587. loss_train: 1.8091550571843982
  21588. step: 55
  21589. running loss: 0.0328937283124436
  21590. Train Steps: 55/90 Loss: 0.0329 torch.Size([8, 600, 800])
  21591. torch.Size([8, 8])
  21592. tensor([[0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117],
  21593. [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600],
  21594. [0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124],
  21595. [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5637, 0.5633],
  21596. [0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494],
  21597. [0.6203, 0.4096, 0.8862, 0.4267, 0.3538, 0.4117, 0.6025, 0.5650],
  21598. [0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431],
  21599. [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5687, 0.5567]],
  21600. device='cuda:0', dtype=torch.float64)
  21601. predictions are: tensor([[ 0.1492, -0.6670, 1.5802, 0.2910, -0.4135, 0.0303, 0.4409, 0.4137],
  21602. [ 0.0904, -0.6878, 1.2486, -1.1853, -0.5766, -0.7854, 0.4236, 0.2221],
  21603. [ 0.7112, -0.3543, 1.6138, -1.0944, -0.4004, -1.0451, 0.4618, -0.0073],
  21604. [ 0.6569, -0.3063, 1.7665, -0.0456, -0.6110, -0.5325, 0.3404, 0.3160],
  21605. [ 0.0950, -0.7382, 1.8449, 0.1953, -0.5241, -0.2591, 0.8265, 0.0738],
  21606. [ 0.1203, -0.6797, 1.9646, -0.3627, -0.5742, -0.3932, 0.5246, 0.2486],
  21607. [ 0.3048, -0.5792, 1.8250, -0.0522, -0.2013, 0.0844, 0.1208, 0.0362],
  21608. [ 0.4090, -0.4900, 1.2181, -1.1421, -0.3401, -1.0661, 0.3066, 0.2915]],
  21609. device='cuda:0', grad_fn=<AddmmBackward>)
  21610. landmarks are: tensor([[[ 0.6118, -0.3983, 1.5824, 0.3469, -0.4268, -0.0688, 0.3469,
  21611. 0.5393],
  21612. [ 0.5466, -0.4080, 1.0668, -1.1764, -0.6289, -0.7617, 0.4855,
  21613. 0.3007],
  21614. [ 0.5790, -0.4079, 1.5929, -1.0630, -0.4729, -1.0725, 0.4137,
  21615. 0.0807],
  21616. [ 0.5902, -0.3493, 1.7961, -0.0072, -0.5942, -0.5615, 0.3180,
  21617. 0.3161],
  21618. [ 0.6224, -0.4032, 1.5940, 0.2930, -0.5769, -0.2690, 0.8900,
  21619. 0.2516],
  21620. [ 0.5790, -0.3940, 1.8076, -0.3152, -0.6520, -0.3844, 0.4970,
  21621. 0.3238],
  21622. [ 0.5351, -0.4153, 1.7326, -0.0457, -0.2214, -0.0466, 0.0434,
  21623. 0.2228],
  21624. [ 0.5824, -0.3625, 1.0405, -0.9325, -0.4268, -1.1389, 0.3411,
  21625. 0.2853]]], device='cuda:0')
  21626. loss_train_step before backward: tensor(0.0284, device='cuda:0', grad_fn=<MseLossBackward>)
  21627. loss_train_step after backward: tensor(0.0284, device='cuda:0', grad_fn=<MseLossBackward>)
  21628. loss_train: 1.837601481936872
  21629. step: 56
  21630. running loss: 0.03281431217744414
  21631.  
  21632. Train Steps: 56/90 Loss: 0.0328 torch.Size([8, 600, 800])
  21633. torch.Size([8, 8])
  21634. tensor([[0.6197, 0.4091, 0.8800, 0.4783, 0.3538, 0.4767, 0.5950, 0.5550],
  21635. [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617],
  21636. [0.6357, 0.4118, 0.8400, 0.2500, 0.5413, 0.1633, 0.6725, 0.5586],
  21637. [0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350],
  21638. [0.6159, 0.4085, 0.6900, 0.2283, 0.4088, 0.1950, 0.5123, 0.5397],
  21639. [0.6286, 0.4078, 0.8063, 0.2267, 0.4788, 0.1533, 0.5953, 0.4913],
  21640. [0.6260, 0.4214, 0.8538, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983],
  21641. [0.6151, 0.4085, 0.8750, 0.4367, 0.3887, 0.4367, 0.5066, 0.5846]],
  21642. device='cuda:0', dtype=torch.float64)
  21643. predictions are: tensor([[ 1.2243e-01, -7.1268e-01, 1.8774e+00, 6.4187e-02, -6.6555e-01,
  21644. 2.1508e-01, 6.6481e-01, 2.6001e-01],
  21645. [ 1.7363e-01, -6.6750e-01, 1.9018e+00, -4.6164e-02, -6.1875e-01,
  21646. 1.2968e-02, 3.6536e-01, 1.5884e-01],
  21647. [ 9.2766e-02, -7.3770e-01, 1.7641e+00, -1.0756e+00, 6.6186e-02,
  21648. -1.2018e+00, 9.7435e-01, 1.6731e-01],
  21649. [ 8.2140e-01, -2.5627e-01, 1.3991e+00, -1.0600e+00, -3.0525e-01,
  21650. -1.0604e+00, 4.5021e-01, 1.6918e-01],
  21651. [ 1.3324e-01, -6.6895e-01, 9.5730e-01, -1.1570e+00, -5.2132e-01,
  21652. -1.1929e+00, 1.6117e-03, 1.6140e-01],
  21653. [ 2.9194e-01, -5.9492e-01, 1.3999e+00, -1.0445e+00, -2.3796e-01,
  21654. -1.2431e+00, 3.2231e-01, 5.8498e-02],
  21655. [ 3.9386e-01, -4.8484e-01, 1.6187e+00, 4.3420e-01, -6.6633e-01,
  21656. -2.1455e-01, 3.6329e-01, 4.0566e-01],
  21657. [ 4.8454e-01, -4.5625e-01, 1.7858e+00, -7.6451e-02, -6.4586e-01,
  21658. 6.5811e-02, 2.6243e-01, 2.7170e-01]], device='cuda:0',
  21659. grad_fn=<AddmmBackward>)
  21660. landmarks are: tensor([[[ 0.5763, -0.3963, 1.7788, -0.0765, -0.6520, -0.0842, 0.4624,
  21661. 0.2776],
  21662. [ 0.5488, -0.4221, 1.8018, -0.2459, -0.4845, -0.3075, 0.2309,
  21663. 0.3087],
  21664. [ 0.6504, -0.3840, 1.5940, -1.1312, 0.2141, -1.5315, 0.8205,
  21665. 0.2944],
  21666. [ 0.5768, -0.4031, 1.3838, -1.1527, -0.2188, -1.4216, 0.4379,
  21667. 0.1850],
  21668. [ 0.5591, -0.3990, 0.9012, -1.2313, -0.3979, -1.3852, 0.0804,
  21669. 0.2071],
  21670. [ 0.6177, -0.4022, 1.4381, -1.2390, -0.0746, -1.5777, 0.4639,
  21671. -0.0168],
  21672. [ 0.6055, -0.3393, 1.6575, 0.2545, -0.5942, -0.5461, 0.2949,
  21673. 0.4778],
  21674. [ 0.5553, -0.3992, 1.7557, -0.2690, -0.4903, -0.2690, 0.0542,
  21675. 0.4145]]], device='cuda:0')
  21676. loss_train_step before backward: tensor(0.0451, device='cuda:0', grad_fn=<MseLossBackward>)
  21677. loss_train_step after backward: tensor(0.0451, device='cuda:0', grad_fn=<MseLossBackward>)
  21678. loss_train: 1.8827112754806876
  21679. step: 57
  21680. running loss: 0.03303002237685417
  21681. Train Steps: 57/90 Loss: 0.0330 torch.Size([8, 600, 800])
  21682. torch.Size([8, 8])
  21683. tensor([[0.6236, 0.4084, 0.7738, 0.2133, 0.3663, 0.3233, 0.5813, 0.5567],
  21684. [0.6200, 0.3978, 0.8900, 0.4550, 0.3775, 0.5200, 0.6150, 0.5367],
  21685. [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383],
  21686. [ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600],
  21687. [0.6229, 0.4066, 0.7612, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350],
  21688. [0.6263, 0.4065, 0.9038, 0.4317, 0.3588, 0.4550, 0.6325, 0.5250],
  21689. [ nan, nan, 0.6512, 0.1717, 0.4100, 0.1983, 0.5253, 0.5240],
  21690. [0.6200, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148]],
  21691. device='cuda:0', dtype=torch.float64)
  21692. predictions are: tensor([[ 1.0962, -0.0636, 1.3366, -0.8499, -0.5666, -0.6745, 0.4527, 0.2407],
  21693. [ 0.5959, -0.3922, 1.8984, 0.1088, -0.5427, 0.2683, 0.5982, 0.1676],
  21694. [ 1.0458, -0.0688, 1.2324, -0.9085, -0.1895, -1.3337, 0.2850, 0.2403],
  21695. [-1.5538, -1.7963, 1.3983, -0.8904, -0.4080, -0.8523, 0.2596, 0.2820],
  21696. [ 0.7834, -0.1845, 1.3360, -0.7645, -0.3736, -1.0244, 0.2668, 0.2252],
  21697. [ 1.0495, -0.0705, 2.0186, 0.0522, -0.6195, -0.0223, 0.7219, 0.1308],
  21698. [-2.0917, -2.1628, 1.0026, -1.2713, -0.3786, -1.3109, 0.1309, 0.2826],
  21699. [ 0.8242, -0.2368, 1.9117, 0.2111, -0.6423, -0.1351, 0.4206, 0.0966]],
  21700. device='cuda:0', grad_fn=<AddmmBackward>)
  21701. landmarks are: tensor([[[ 0.5946, -0.3995, 1.2880, -1.3005, -0.5942, -0.7925, 0.3988,
  21702. 0.2853],
  21703. [ 0.5776, -0.4484, 1.8249, -0.1843, -0.5423, 0.1159, 0.5547,
  21704. 0.1929],
  21705. [ 0.5911, -0.3984, 1.1956, -1.0850, -0.0804, -1.5392, 0.4393,
  21706. 0.2006],
  21707. [-2.2859, -2.2859, 1.4266, -1.1389, -0.4499, -1.1235, 0.2891,
  21708. 0.3007],
  21709. [ 0.5911, -0.4080, 1.2303, -0.9156, -0.3229, -1.2851, 0.4508,
  21710. 0.1852],
  21711. [ 0.6070, -0.4085, 1.8885, -0.2921, -0.6289, -0.1843, 0.6356,
  21712. 0.1390],
  21713. [-2.2859, -2.2859, 0.7222, -1.4930, -0.3921, -1.3698, 0.1404,
  21714. 0.1343],
  21715. [ 0.5781, -0.4205, 1.8159, -0.0691, -0.6116, -0.3069, 0.4236,
  21716. 0.0919]]], device='cuda:0')
  21717. loss_train_step before backward: tensor(0.0496, device='cuda:0', grad_fn=<MseLossBackward>)
  21718. loss_train_step after backward: tensor(0.0496, device='cuda:0', grad_fn=<MseLossBackward>)
  21719. loss_train: 1.932359985075891
  21720. step: 58
  21721. running loss: 0.03331655146682571
  21722. Train Steps: 58/90 Loss: 0.0333 torch.Size([8, 600, 800])
  21723. torch.Size([8, 8])
  21724. tensor([[0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400],
  21725. [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374],
  21726. [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617],
  21727. [0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283],
  21728. [0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835],
  21729. [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150],
  21730. [0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433],
  21731. [0.6275, 0.4048, 0.8488, 0.2883, 0.4463, 0.2033, 0.6321, 0.5155]],
  21732. device='cuda:0', dtype=torch.float64)
  21733. predictions are: tensor([[ 0.3482, -0.5847, 1.6769, -0.4344, -0.6912, -0.1510, 0.5420, 0.2736],
  21734. [ 0.4134, -0.4870, 1.6019, -0.5267, -0.6652, -0.7825, 0.3372, 0.1593],
  21735. [ 0.5419, -0.4151, 1.6395, -0.1511, -0.5665, -0.2487, 0.1434, 0.2999],
  21736. [ 0.0570, -0.7853, 1.8228, -0.4909, -0.6013, -0.4838, 0.6780, 0.1995],
  21737. [ 0.5814, -0.4066, 1.7251, -0.4970, -0.4215, -1.3153, 0.4057, 0.0920],
  21738. [ 0.6046, -0.3934, 1.6968, -0.0545, -0.2929, 0.4666, 0.5360, 0.3136],
  21739. [ 0.5320, -0.4514, 1.5537, 0.1591, -0.1326, 0.0026, 0.1167, 0.3526],
  21740. [ 0.3389, -0.5410, 1.5506, -0.9298, -0.3028, -1.3098, 0.5715, 0.1343]],
  21741. device='cuda:0', grad_fn=<AddmmBackward>)
  21742. landmarks are: tensor([[[ 0.5778, -0.4393, 1.8018, -0.4614, -0.6693, -0.1381, 0.5490,
  21743. 0.2083],
  21744. [ 0.5777, -0.4416, 1.7044, -0.5827, -0.5962, -0.8361, 0.4862,
  21745. 0.1963],
  21746. [ 0.5488, -0.4221, 1.8018, -0.2459, -0.4845, -0.3075, 0.2309,
  21747. 0.3087],
  21748. [ 0.6073, -0.4058, 1.8885, -0.4999, -0.5942, -0.4768, 0.6413,
  21749. 0.1544],
  21750. [ 0.6117, -0.3877, 1.8249, -0.5230, -0.3517, -1.2851, 0.4200,
  21751. -0.0529],
  21752. [ 0.6014, -0.3883, 1.8192, -0.1150, -0.2882, 0.4008, 0.5201,
  21753. 0.0928],
  21754. [ 0.5295, -0.4373, 1.6553, 0.1011, 0.0380, 0.0671, 0.0813,
  21755. 0.2237],
  21756. [ 0.6126, -0.4161, 1.6344, -0.9541, -0.2247, -1.3467, 0.6339,
  21757. 0.0953]]], device='cuda:0')
  21758. loss_train_step before backward: tensor(0.0151, device='cuda:0', grad_fn=<MseLossBackward>)
  21759. loss_train_step after backward: tensor(0.0151, device='cuda:0', grad_fn=<MseLossBackward>)
  21760. loss_train: 1.9474248168990016
  21761. step: 59
  21762. running loss: 0.03300720028642375
  21763. Train Steps: 59/90 Loss: 0.0330 torch.Size([8, 600, 800])
  21764. torch.Size([8, 8])
  21765. tensor([[0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901],
  21766. [0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183],
  21767. [0.6219, 0.3934, 0.8688, 0.5267, 0.4313, 0.4967, 0.5988, 0.4983],
  21768. [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938],
  21769. [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833],
  21770. [0.6076, 0.3953, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954],
  21771. [0.6231, 0.3973, 0.8650, 0.3950, 0.3625, 0.3183, 0.5837, 0.5167],
  21772. [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663]],
  21773. device='cuda:0', dtype=torch.float64)
  21774. predictions are: tensor([[ 0.3817, -0.5736, 1.9059, -0.1225, -0.4497, -0.2011, 0.2088, 0.1125],
  21775. [ 0.9177, -0.2117, 1.3480, -0.8393, -0.3956, -0.8119, 0.5936, 0.5077],
  21776. [ 0.4320, -0.5735, 1.8279, 0.1515, -0.3319, -0.0194, 0.5533, 0.1154],
  21777. [ 0.4054, -0.5492, 1.9774, 0.1779, -0.5931, -0.5311, 0.6125, 0.0498],
  21778. [ 0.6835, -0.3405, 0.9233, -0.8949, -0.5273, -0.9222, 0.1870, 0.4747],
  21779. [ 0.4931, -0.4520, 1.6901, -0.4726, -0.6711, -0.3948, 0.3562, 0.0879],
  21780. [ 0.4073, -0.5244, 1.8318, -0.3861, -0.5785, -0.7754, 0.4366, 0.1827],
  21781. [-0.1445, -0.8772, 1.1068, -1.2608, -0.4086, -1.1984, 0.1743, 0.3005]],
  21782. device='cuda:0', grad_fn=<AddmmBackward>)
  21783. landmarks are: tensor([[[ 0.5663, -0.3955, 1.7788, -0.2382, -0.4037, -0.2690, 0.0828,
  21784. -0.0220],
  21785. [ 0.5759, -0.3948, 1.2072, -0.9849, -0.4672, -0.8156, 0.4912,
  21786. 0.5701],
  21787. [ 0.5864, -0.4690, 1.7268, 0.1467, -0.2940, 0.0082, 0.4797,
  21788. 0.0159],
  21789. [ 0.6273, -0.4105, 1.8480, 0.1082, -0.5538, -0.5076, 0.6414,
  21790. -0.0049],
  21791. [ 0.5614, -0.4080, 0.7741, -0.8848, -0.5423, -0.9156, 0.3584,
  21792. 0.4085],
  21793. [ 0.5206, -0.4603, 1.4670, -0.4768, -0.6577, -0.3998, 0.1836,
  21794. 0.0021],
  21795. [ 0.5924, -0.4507, 1.7095, -0.4614, -0.6115, -0.8156, 0.4104,
  21796. 0.1005],
  21797. [ 0.5269, -0.4176, 1.0628, -1.3159, -0.4037, -1.2236, 0.2432,
  21798. 0.3297]]], device='cuda:0')
  21799. loss_train_step before backward: tensor(0.0221, device='cuda:0', grad_fn=<MseLossBackward>)
  21800. loss_train_step after backward: tensor(0.0221, device='cuda:0', grad_fn=<MseLossBackward>)
  21801. loss_train: 1.969504582695663
  21802. step: 60
  21803. running loss: 0.03282507637826105
  21804.  
  21805. Train Steps: 60/90 Loss: 0.0328 torch.Size([8, 600, 800])
  21806. torch.Size([8, 8])
  21807. tensor([[0.6129, 0.4069, 0.8750, 0.5067, 0.3875, 0.4233, 0.5235, 0.5881],
  21808. [0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124],
  21809. [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483],
  21810. [ nan, nan, 0.7192, 0.2346, 0.4037, 0.2050, 0.5138, 0.5650],
  21811. [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633],
  21812. [0.6346, 0.4086, 0.7938, 0.5500, 0.3962, 0.4867, 0.7343, 0.5702],
  21813. [0.6097, 0.4000, 0.7325, 0.2667, 0.3450, 0.3517, 0.5284, 0.5045],
  21814. [0.6228, 0.4004, 0.8750, 0.5250, 0.3825, 0.5233, 0.6362, 0.5000]],
  21815. device='cuda:0', dtype=torch.float64)
  21816. predictions are: tensor([[ 0.7551, -0.2848, 1.7650, 0.1631, -0.5179, -0.3995, 0.1350, 0.3288],
  21817. [ 0.6782, -0.3846, 1.5860, -1.0260, -0.3997, -1.1093, 0.5274, 0.0610],
  21818. [ 0.6320, -0.3760, 1.7877, -0.3926, -0.4922, -0.0774, 0.3800, 0.2165],
  21819. [-1.5177, -1.8227, 1.0843, -1.3910, -0.4651, -1.3811, 0.1120, 0.2572],
  21820. [ 0.8237, -0.2159, 1.7595, 0.2419, -0.5153, -0.2390, 0.3561, 0.3817],
  21821. [ 0.8449, -0.2290, 1.5374, 0.1570, -0.4488, -0.2010, 0.8125, 0.3838],
  21822. [ 0.3734, -0.5396, 1.2457, -1.0386, -0.6445, -0.7832, 0.1265, 0.0847],
  21823. [ 0.8764, -0.2458, 1.8436, 0.1623, -0.4635, -0.0212, 0.6347, 0.1237]],
  21824. device='cuda:0', grad_fn=<AddmmBackward>)
  21825. landmarks are: tensor([[[ 0.5452, -0.4067, 1.7557, 0.0543, -0.4961, -0.3306, 0.1323,
  21826. 0.4306],
  21827. [ 0.5790, -0.4079, 1.5929, -1.0630, -0.4729, -1.0725, 0.4137,
  21828. 0.0807],
  21829. [ 0.5546, -0.4620, 1.7788, -0.4229, -0.5192, -0.0226, 0.4277,
  21830. 0.2468],
  21831. [-2.2859, -2.2859, 1.0361, -1.2021, -0.4210, -1.3390, 0.0871,
  21832. 0.3238],
  21833. [ 0.5879, -0.3591, 1.8018, 0.1236, -0.5596, -0.1612, 0.3469,
  21834. 0.3161],
  21835. [ 0.6454, -0.3984, 1.3804, 0.2545, -0.4557, -0.0380, 1.1057,
  21836. 0.3478],
  21837. [ 0.5303, -0.4384, 1.0975, -1.0542, -0.6924, -0.6616, 0.1548,
  21838. 0.0442],
  21839. [ 0.5908, -0.4366, 1.7557, 0.1390, -0.5192, 0.1313, 0.6529,
  21840. 0.0236]]], device='cuda:0')
  21841. loss_train_step before backward: tensor(0.0242, device='cuda:0', grad_fn=<MseLossBackward>)
  21842. loss_train_step after backward: tensor(0.0242, device='cuda:0', grad_fn=<MseLossBackward>)
  21843. loss_train: 1.9937451435253024
  21844. step: 61
  21845. running loss: 0.03268434661516889
  21846. Train Steps: 61/90 Loss: 0.0327 torch.Size([8, 600, 800])
  21847. torch.Size([8, 8])
  21848. tensor([[0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5837, 0.5500],
  21849. [0.6124, 0.4075, 0.7696, 0.4153, 0.3475, 0.3767, 0.5157, 0.5427],
  21850. [0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947],
  21851. [0.6112, 0.4029, 0.8638, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567],
  21852. [0.6293, 0.4097, 0.8800, 0.2517, 0.5262, 0.2600, 0.7430, 0.5378],
  21853. [0.6216, 0.4099, 0.7225, 0.2033, 0.4188, 0.2217, 0.5975, 0.5283],
  21854. [0.6128, 0.4084, 0.8738, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397],
  21855. [0.6075, 0.4000, 0.8513, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280]],
  21856. device='cuda:0', dtype=torch.float64)
  21857. predictions are: tensor([[ 0.3201, -0.6410, 1.7896, -0.1694, -0.4898, 0.0609, 0.4582, 0.2323],
  21858. [ 0.4722, -0.4604, 1.3808, -0.3858, -0.7707, -0.5247, 0.1395, 0.3130],
  21859. [ 0.5414, -0.4888, 1.3641, -1.3285, -0.4477, -1.3795, 0.6508, 0.0505],
  21860. [ 0.3127, -0.6277, 1.7601, -0.0538, -0.2350, -0.0423, 0.4265, 0.3210],
  21861. [ 0.2437, -0.6771, 1.7701, -1.0139, -0.0564, -1.2747, 1.0509, 0.1679],
  21862. [ 0.5388, -0.4299, 1.0035, -1.1597, -0.4715, -1.3582, 0.1529, 0.2767],
  21863. [ 0.8609, -0.2031, 1.7233, 0.0268, -0.7773, -0.6240, 0.2071, 0.1623],
  21864. [ 0.5464, -0.4439, 1.6349, 0.1478, -0.3280, 0.1001, 0.1455, 0.2697]],
  21865. device='cuda:0', grad_fn=<AddmmBackward>)
  21866. landmarks are: tensor([[[ 5.6039e-01, -4.6197e-01, 1.7961e+00, -1.9969e-01, -3.2286e-01,
  21867. 1.0824e-01, 4.1039e-01, 2.5450e-01],
  21868. [ 5.4267e-01, -4.0354e-01, 1.2688e+00, -3.6754e-01, -6.8083e-01,
  21869. -5.4611e-01, 9.5867e-02, 2.2059e-01],
  21870. [ 6.0739e-01, -4.0631e-01, 1.3630e+00, -1.5238e+00, -3.1132e-01,
  21871. -1.2390e+00, 6.1950e-01, -9.2270e-04],
  21872. [ 5.3712e-01, -4.2517e-01, 1.7037e+00, -6.8822e-02, -3.4180e-02,
  21873. 6.2048e-02, 3.7575e-01, 2.8530e-01],
  21874. [ 6.2072e-01, -3.9360e-01, 1.7788e+00, -1.1235e+00, 1.4480e-01,
  21875. -1.0850e+00, 1.1459e+00, 1.9818e-01],
  21876. [ 5.8505e-01, -3.9246e-01, 1.0513e+00, -1.3467e+00, -3.5173e-01,
  21877. -1.2620e+00, 4.7390e-01, 1.5443e-01],
  21878. [ 5.4434e-01, -3.9938e-01, 1.7499e+00, -1.2271e-01, -6.1732e-01,
  21879. -5.7691e-01, 4.8756e-03, 2.0706e-01],
  21880. [ 5.2021e-01, -4.3818e-01, 1.6460e+00, 1.0824e-01, -2.0286e-01,
  21881. 1.7544e-01, 1.0666e-01, 1.5296e-01]]], device='cuda:0')
  21882. loss_train_step before backward: tensor(0.0184, device='cuda:0', grad_fn=<MseLossBackward>)
  21883. loss_train_step after backward: tensor(0.0184, device='cuda:0', grad_fn=<MseLossBackward>)
  21884. loss_train: 2.0121765499934554
  21885. step: 62
  21886. running loss: 0.03245446048376541
  21887. Train Steps: 62/90 Loss: 0.0325 torch.Size([8, 600, 800])
  21888. torch.Size([8, 8])
  21889. tensor([[0.6129, 0.4069, 0.8750, 0.5067, 0.3875, 0.4233, 0.5235, 0.5881],
  21890. [0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904],
  21891. [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100],
  21892. [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167],
  21893. [0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6362, 0.5200],
  21894. [0.6241, 0.4143, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550],
  21895. [0.6286, 0.4055, 0.9000, 0.4717, 0.3763, 0.4683, 0.7018, 0.5494],
  21896. [0.6265, 0.4088, 0.8025, 0.1850, 0.4163, 0.2500, 0.6290, 0.4947]],
  21897. device='cuda:0', dtype=torch.float64)
  21898. predictions are: tensor([[ 0.5197, -0.4527, 1.5950, 0.1716, -0.5711, -0.4149, 0.0090, 0.3377],
  21899. [ 0.3744, -0.5529, 1.5663, -1.0771, -0.2237, -1.2708, 0.4707, 0.0513],
  21900. [ 0.5448, -0.4714, 1.7032, -0.2847, -0.6567, -0.5911, 0.5097, 0.0959],
  21901. [ 0.6185, -0.3506, 1.0383, -0.7511, -0.1287, -1.3023, 0.0891, 0.4953],
  21902. [ 0.3769, -0.5777, 1.6381, -0.5963, -0.5831, 0.0237, 0.5714, 0.2753],
  21903. [ 0.5147, -0.4490, 1.7073, -0.1214, -0.4517, 0.0542, 0.3266, 0.3011],
  21904. [ 0.5869, -0.4673, 1.8068, -0.0764, -0.5900, -0.2953, 0.7764, 0.1528],
  21905. [ 0.4651, -0.5418, 1.3631, -1.2766, -0.3952, -1.2271, 0.5598, 0.0800]],
  21906. device='cuda:0', grad_fn=<AddmmBackward>)
  21907. landmarks are: tensor([[[ 5.4515e-01, -4.0670e-01, 1.7557e+00, 5.4350e-02, -4.9607e-01,
  21908. -3.3056e-01, 1.3228e-01, 4.3063e-01],
  21909. [ 6.1264e-01, -3.8707e-01, 1.6229e+00, -1.0773e+00, -2.1316e-01,
  21910. -1.3698e+00, 5.8291e-01, -2.0913e-02],
  21911. [ 6.2200e-01, -4.4357e-01, 1.8711e+00, -3.6905e-01, -6.1732e-01,
  21912. -4.9992e-01, 6.7021e-01, 6.9746e-02],
  21913. [ 6.1611e-01, -3.0754e-01, 1.1678e+00, -6.5000e-01, 8.1293e-02,
  21914. -1.4006e+00, 2.5450e-01, 5.6243e-01],
  21915. [ 5.8412e-01, -4.1986e-01, 1.7961e+00, -6.6928e-01, -6.2309e-01,
  21916. 1.0824e-01, 6.5289e-01, 1.1594e-01],
  21917. [ 5.9677e-01, -3.7252e-01, 1.8423e+00, -1.3811e-01, -4.0370e-01,
  21918. 1.8522e-01, 6.0092e-01, 2.7760e-01],
  21919. [ 6.1742e-01, -4.1286e-01, 1.8711e+00, -1.0731e-01, -5.4804e-01,
  21920. -1.2271e-01, 9.5578e-01, 2.5161e-01],
  21921. [ 6.0785e-01, -3.9761e-01, 1.4208e+00, -1.4314e+00, -3.6328e-01,
  21922. -1.1312e+00, 6.1950e-01, -9.2270e-04]]], device='cuda:0')
  21923. loss_train_step before backward: tensor(0.0129, device='cuda:0', grad_fn=<MseLossBackward>)
  21924. loss_train_step after backward: tensor(0.0129, device='cuda:0', grad_fn=<MseLossBackward>)
  21925. loss_train: 2.0250426745042205
  21926. step: 63
  21927. running loss: 0.03214353451594001
  21928. Train Steps: 63/90 Loss: 0.0321 torch.Size([8, 600, 800])
  21929. torch.Size([8, 8])
  21930. tensor([[0.6213, 0.4001, 0.7712, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183],
  21931. [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114],
  21932. [0.6134, 0.4090, 0.6926, 0.2819, 0.3538, 0.3233, 0.5563, 0.5667],
  21933. [0.6113, 0.4088, 0.6859, 0.2208, 0.4363, 0.1700, 0.5188, 0.5533],
  21934. [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051],
  21935. [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290],
  21936. [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378],
  21937. [0.6250, 0.4131, 0.8688, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367]],
  21938. device='cuda:0', dtype=torch.float64)
  21939. predictions are: tensor([[ 0.5808, -0.4370, 1.2576, -1.1996, -0.3212, -1.3849, 0.3267, 0.0878],
  21940. [ 0.4906, -0.5313, 1.7542, 0.1598, -0.3029, 0.2531, 0.3300, 0.1995],
  21941. [ 0.3006, -0.5941, 1.0523, -0.8893, -0.6749, -0.6995, 0.1450, 0.3257],
  21942. [ 0.6383, -0.3974, 0.9691, -1.1039, -0.3556, -1.3894, 0.1058, 0.2528],
  21943. [ 0.5203, -0.4946, 1.8813, -0.0241, -0.6237, -0.5090, 0.5620, 0.1526],
  21944. [ 0.4117, -0.5442, 1.3097, -1.0924, -0.3429, -1.2326, 0.5587, 0.1343],
  21945. [ 0.6043, -0.4339, 1.7556, 0.0348, -0.4017, 0.2142, 0.6594, 0.2807],
  21946. [ 0.4688, -0.4926, 1.7983, -0.7584, -0.3287, -1.0427, 0.5270, 0.1824]],
  21947. device='cuda:0', grad_fn=<AddmmBackward>)
  21948. landmarks are: tensor([[[ 0.5838, -0.4379, 1.2764, -1.3082, -0.2824, -1.4545, 0.4162,
  21949. 0.1082],
  21950. [ 0.5450, -0.4730, 1.7420, 0.1372, -0.1919, 0.2614, 0.4976,
  21951. 0.0764],
  21952. [ 0.5473, -0.3966, 0.9131, -0.9838, -0.6520, -0.7925, 0.2834,
  21953. 0.3315],
  21954. [ 0.5377, -0.3978, 0.8824, -1.2663, -0.2709, -1.5007, 0.1102,
  21955. 0.2699],
  21956. [ 0.6204, -0.4336, 1.8654, -0.0688, -0.6058, -0.5230, 0.6503,
  21957. 0.0472],
  21958. [ 0.6125, -0.4273, 1.2807, -1.3253, -0.2574, -1.2542, 0.6864,
  21959. 0.1575],
  21960. [ 0.6141, -0.4345, 1.6864, -0.0303, -0.2882, 0.1544, 0.9521,
  21961. 0.1982],
  21962. [ 0.6011, -0.3778, 1.7268, -0.9079, -0.3113, -1.1928, 0.5605,
  21963. 0.1929]]], device='cuda:0')
  21964. loss_train_step before backward: tensor(0.0106, device='cuda:0', grad_fn=<MseLossBackward>)
  21965. loss_train_step after backward: tensor(0.0106, device='cuda:0', grad_fn=<MseLossBackward>)
  21966. loss_train: 2.035658926703036
  21967. step: 64
  21968. running loss: 0.031807170729734935
  21969.  
  21970. Train Steps: 64/90 Loss: 0.0318 torch.Size([8, 600, 800])
  21971. torch.Size([8, 8])
  21972. tensor([[0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250],
  21973. [ nan, nan, 0.6412, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700],
  21974. [ nan, nan, 0.7240, 0.2722, 0.3900, 0.2567, 0.5168, 0.5933],
  21975. [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083],
  21976. [0.6128, 0.4022, 0.8738, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064],
  21977. [0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783],
  21978. [0.6249, 0.4142, 0.8350, 0.3283, 0.3613, 0.3700, 0.6188, 0.5400],
  21979. [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483]],
  21980. device='cuda:0', dtype=torch.float64)
  21981. predictions are: tensor([[ 1.1619, -0.0507, 1.9458, 0.1117, -0.3936, 0.2390, 0.6420, 0.0908],
  21982. [-0.6746, -1.2440, 0.9450, -1.3181, -0.3175, -1.5303, 0.2282, 0.3076],
  21983. [-0.8266, -1.3301, 1.1158, -1.1625, -0.5268, -1.1397, 0.0964, 0.2674],
  21984. [ 0.9588, -0.1822, 1.7905, 0.1449, -0.5481, -0.2363, 0.5219, 0.3334],
  21985. [ 0.9507, -0.1960, 1.8109, 0.1169, -0.0843, 0.0336, 0.1701, 0.1119],
  21986. [-0.1099, -0.8441, 1.0848, -1.0951, -0.3097, -1.2919, 0.2902, 0.3028],
  21987. [ 1.0059, -0.1478, 1.7145, -0.5904, -0.6488, -0.6710, 0.6253, 0.1184],
  21988. [ 0.9850, -0.1714, 1.1831, -1.1515, -0.3543, -1.2117, 0.5893, 0.1447]],
  21989. device='cuda:0', grad_fn=<AddmmBackward>)
  21990. landmarks are: tensor([[[ 0.5991, -0.3803, 1.8018, -0.0534, -0.3460, 0.1852, 0.5374,
  21991. 0.1390],
  21992. [-2.2859, -2.2859, 0.6760, -1.4083, -0.3286, -1.4160, 0.2487,
  21993. 0.3469],
  21994. [-2.2859, -2.2859, 1.0580, -1.0288, -0.4845, -1.1004, 0.1011,
  21995. 0.4543],
  21996. [ 0.5713, -0.4105, 1.7557, 0.0467, -0.6520, -0.2690, 0.3988,
  21997. 0.5239],
  21998. [ 0.5446, -0.4280, 1.7499, 0.0543, 0.0156, 0.1301, 0.1918,
  21999. 0.0532],
  22000. [ 0.5742, -0.4474, 0.9834, -1.0159, -0.3229, -1.3159, 0.2314,
  22001. 0.3854],
  22002. [ 0.6006, -0.3728, 1.5709, -0.7694, -0.6173, -0.5769, 0.5721,
  22003. 0.2083],
  22004. [ 0.5784, -0.4085, 1.0859, -1.3929, -0.4037, -1.1158, 0.5605,
  22005. 0.2468]]], device='cuda:0')
  22006. loss_train_step before backward: tensor(0.1430, device='cuda:0', grad_fn=<MseLossBackward>)
  22007. loss_train_step after backward: tensor(0.1430, device='cuda:0', grad_fn=<MseLossBackward>)
  22008. loss_train: 2.1786455223336816
  22009. step: 65
  22010. running loss: 0.03351762342051818
  22011. Train Steps: 65/90 Loss: 0.0335 torch.Size([8, 600, 800])
  22012. torch.Size([8, 8])
  22013. tensor([[0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367],
  22014. [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901],
  22015. [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167],
  22016. [0.6079, 0.3964, 0.7420, 0.2958, 0.3563, 0.2917, 0.5351, 0.4980],
  22017. [0.6200, 0.4059, 0.8700, 0.4900, 0.4163, 0.5000, 0.6162, 0.5467],
  22018. [0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742],
  22019. [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390],
  22020. [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758]],
  22021. device='cuda:0', dtype=torch.float64)
  22022. predictions are: tensor([[ 0.4607, -0.5076, 1.5616, -0.9819, -0.3880, -1.0047, 0.6830, 0.2139],
  22023. [ 0.2895, -0.5994, 1.3120, -1.0558, -0.5012, -1.1000, 0.1815, 0.0953],
  22024. [ 0.7617, -0.2370, 1.0651, -0.8918, -0.0222, -1.3057, 0.3144, 0.5303],
  22025. [ 0.2737, -0.6040, 1.1433, -1.1580, -0.5218, -0.9390, 0.1412, 0.1010],
  22026. [ 0.6965, -0.3564, 1.6710, -0.2018, -0.2981, -0.0155, 0.4957, 0.2971],
  22027. [ 0.6458, -0.3963, 1.7657, 0.1448, -0.5660, -0.6313, 0.6119, 0.0351],
  22028. [ 0.5802, -0.4325, 1.6830, -0.1799, -0.3756, -0.1435, 0.3684, 0.2465],
  22029. [ 0.6696, -0.3913, 1.7502, -0.2301, -0.3841, -0.0126, 0.6902, 0.1319]],
  22030. device='cuda:0', grad_fn=<AddmmBackward>)
  22031. landmarks are: tensor([[[ 0.6174, -0.4201, 1.7309, -0.8784, -0.4735, -0.9524, 0.6242,
  22032. 0.1931],
  22033. [ 0.5515, -0.4129, 1.4785, -1.0080, -0.5192, -1.1004, 0.1034,
  22034. -0.0220],
  22035. [ 0.6161, -0.3075, 1.1678, -0.6500, 0.0813, -1.4006, 0.2545,
  22036. 0.5624],
  22037. [ 0.5219, -0.4550, 1.1415, -0.9196, -0.6404, -0.9387, 0.1856,
  22038. 0.0141],
  22039. [ 0.5776, -0.4109, 1.7326, -0.0226, -0.3633, 0.0236, 0.5605,
  22040. 0.2391],
  22041. [ 0.6113, -0.4057, 1.7499, 0.3007, -0.5885, -0.5384, 0.4513,
  22042. -0.0957],
  22043. [ 0.5250, -0.4661, 1.7383, -0.0765, -0.4268, -0.0226, 0.2535,
  22044. 0.2035],
  22045. [ 0.6072, -0.4250, 1.8249, -0.0072, -0.4037, 0.1082, 0.6730,
  22046. -0.0881]]], device='cuda:0')
  22047. loss_train_step before backward: tensor(0.0131, device='cuda:0', grad_fn=<MseLossBackward>)
  22048. loss_train_step after backward: tensor(0.0131, device='cuda:0', grad_fn=<MseLossBackward>)
  22049. loss_train: 2.1917926585301757
  22050. step: 66
  22051. running loss: 0.03320897967469963
  22052. Train Steps: 66/90 Loss: 0.0332 torch.Size([8, 600, 800])
  22053. torch.Size([8, 8])
  22054. tensor([[0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083],
  22055. [0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033],
  22056. [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398],
  22057. [0.6171, 0.4127, 0.8900, 0.4800, 0.4325, 0.5783, 0.5769, 0.5090],
  22058. [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5988, 0.5283],
  22059. [0.6262, 0.4163, 0.8850, 0.5183, 0.3763, 0.4150, 0.6025, 0.5500],
  22060. [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400],
  22061. [0.6095, 0.3970, 0.8688, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940]],
  22062. device='cuda:0', dtype=torch.float64)
  22063. predictions are: tensor([[ 0.5727, -0.3898, 1.5611, -0.2527, -0.5054, -0.5357, 0.4417, 0.4269],
  22064. [ 0.4821, -0.4974, 1.6563, -0.2330, -0.2927, -0.1848, 0.4000, 0.1484],
  22065. [ 0.5810, -0.4360, 1.5775, -0.0716, -0.4948, -0.5690, 0.5614, 0.1391],
  22066. [ 1.0774, -0.0504, 1.6881, -0.3434, -0.2318, 0.0752, 0.3415, 0.1202],
  22067. [ 0.6469, -0.3567, 1.6222, -0.0748, -0.4106, -0.3050, 0.4462, 0.2349],
  22068. [ 0.7940, -0.2222, 1.6271, -0.2506, -0.5119, -0.7262, 0.5196, 0.1868],
  22069. [ 0.4953, -0.4453, 1.4948, -1.0337, -0.5386, -0.4044, 0.5632, 0.2785],
  22070. [ 0.4612, -0.5075, 1.6961, -0.4348, -0.0683, -0.3957, 0.2832, 0.0984]],
  22071. device='cuda:0', grad_fn=<AddmmBackward>)
  22072. landmarks are: tensor([[[ 0.5713, -0.4105, 1.7557, 0.0467, -0.6520, -0.2690, 0.3988,
  22073. 0.5239],
  22074. [ 0.5668, -0.4306, 1.7730, 0.0697, -0.4037, 0.1390, 0.4450,
  22075. 0.0390],
  22076. [ 0.5786, -0.4463, 1.6655, 0.2216, -0.5115, -0.2675, 0.4236,
  22077. 0.2075],
  22078. [ 0.5645, -0.3797, 1.8249, -0.0688, -0.2882, 0.3854, 0.3789,
  22079. 0.0652],
  22080. [ 0.5603, -0.4319, 1.7788, 0.1775, -0.5538, -0.0611, 0.4797,
  22081. 0.1544],
  22082. [ 0.6066, -0.3632, 1.8018, 0.1082, -0.5480, -0.3691, 0.4970,
  22083. 0.2545],
  22084. [ 0.5631, -0.4008, 1.6113, -0.7309, -0.6693, -0.1304, 0.5836,
  22085. 0.2083],
  22086. [ 0.5292, -0.4520, 1.7268, -0.0842, -0.0413, -0.0324, 0.1116,
  22087. -0.0039]]], device='cuda:0')
  22088. loss_train_step before backward: tensor(0.0361, device='cuda:0', grad_fn=<MseLossBackward>)
  22089. loss_train_step after backward: tensor(0.0361, device='cuda:0', grad_fn=<MseLossBackward>)
  22090. loss_train: 2.227865911088884
  22091. step: 67
  22092. running loss: 0.033251730016252
  22093. Train Steps: 67/90 Loss: 0.0333 torch.Size([8, 600, 800])
  22094. torch.Size([8, 8])
  22095. tensor([[0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117],
  22096. [0.6257, 0.4034, 0.8287, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901],
  22097. [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834],
  22098. [0.6124, 0.4069, 0.8314, 0.5001, 0.3738, 0.4650, 0.5167, 0.5402],
  22099. [ nan, nan, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650],
  22100. [0.6222, 0.4169, 0.8638, 0.5650, 0.4313, 0.4783, 0.5637, 0.5633],
  22101. [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869],
  22102. [0.6208, 0.4082, 0.8538, 0.3067, 0.3588, 0.3717, 0.6112, 0.5517]],
  22103. device='cuda:0', dtype=torch.float64)
  22104. predictions are: tensor([[ 0.8848, -0.1729, 1.4798, 0.2416, -0.3977, -0.0930, 0.4602, 0.4335],
  22105. [ 0.7835, -0.2613, 1.4584, -1.2296, -0.3859, -1.1383, 0.6709, -0.0248],
  22106. [ 0.8811, -0.2185, 1.4146, -1.4647, -0.0089, -1.5862, 0.7598, 0.0176],
  22107. [ 0.7126, -0.2833, 1.5513, -0.0976, -0.4387, -0.1353, 0.2362, 0.2505],
  22108. [-2.0188, -2.1723, 1.0514, -1.4950, -0.3005, -1.1029, 0.2591, 0.3280],
  22109. [ 0.7695, -0.2286, 1.6170, 0.1916, -0.2849, -0.1186, 0.4334, 0.3344],
  22110. [ 0.7143, -0.2892, 1.6661, 0.1946, -0.4831, -0.2583, 0.2466, 0.0479],
  22111. [ 0.8132, -0.2083, 1.5800, -0.8766, -0.5991, -0.5577, 0.4929, 0.1575]],
  22112. device='cuda:0', grad_fn=<AddmmBackward>)
  22113. landmarks are: tensor([[[ 0.6118, -0.3983, 1.5824, 0.3469, -0.4268, -0.0688, 0.3469,
  22114. 0.5393],
  22115. [ 0.6042, -0.4225, 1.5420, -1.2082, -0.4730, -1.0311, 0.6380,
  22116. -0.0220],
  22117. [ 0.6126, -0.4057, 1.4439, -1.3159, -0.1150, -1.5777, 0.5537,
  22118. -0.0530],
  22119. [ 0.5425, -0.4067, 1.5543, 0.0241, -0.5596, -0.1381, 0.1005,
  22120. 0.2093],
  22121. [-2.2859, -2.2859, 0.9175, -1.3947, -0.3691, -1.2467, 0.2314,
  22122. 0.3238],
  22123. [ 0.5880, -0.3605, 1.7037, 0.3238, -0.2940, -0.0765, 0.3180,
  22124. 0.3161],
  22125. [ 0.5320, -0.4488, 1.6633, 0.3315, -0.5018, -0.2459, 0.0828,
  22126. -0.0370],
  22127. [ 0.5814, -0.4003, 1.6575, -0.8694, -0.6289, -0.5692, 0.5374,
  22128. 0.2622]]], device='cuda:0')
  22129. loss_train_step before backward: tensor(0.0155, device='cuda:0', grad_fn=<MseLossBackward>)
  22130. loss_train_step after backward: tensor(0.0155, device='cuda:0', grad_fn=<MseLossBackward>)
  22131. loss_train: 2.243375484831631
  22132. step: 68
  22133. running loss: 0.03299081595340634
  22134.  
  22135. Train Steps: 68/90 Loss: 0.0330 torch.Size([8, 600, 800])
  22136. torch.Size([8, 8])
  22137. tensor([[0.6229, 0.4198, 0.7662, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783],
  22138. [0.6222, 0.3957, 0.8838, 0.5017, 0.3937, 0.4600, 0.5900, 0.5017],
  22139. [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869],
  22140. [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841],
  22141. [0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6188, 0.5433],
  22142. [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609],
  22143. [0.6201, 0.4017, 0.8871, 0.4621, 0.3517, 0.4675, 0.5999, 0.5106],
  22144. [0.6142, 0.3982, 0.8650, 0.4883, 0.3912, 0.4317, 0.5315, 0.5350]],
  22145. device='cuda:0', dtype=torch.float64)
  22146. predictions are: tensor([[ 0.5492, -0.3537, 1.0510, -1.1307, -0.1063, -1.2669, 0.3406, 0.4151],
  22147. [ 0.5106, -0.4782, 1.6313, -0.0935, -0.4695, -0.1447, 0.2669, 0.0259],
  22148. [ 0.5940, -0.3389, 1.5285, -0.8920, -0.1874, -1.3273, 0.2889, 0.0553],
  22149. [ 0.6428, -0.3886, 1.6949, -0.4102, -0.3667, -0.6118, 0.9077, 0.2542],
  22150. [ 0.7256, -0.3228, 1.7196, -0.2445, -0.4457, 0.0972, 0.4459, 0.1771],
  22151. [ 0.4534, -0.4983, 1.6209, -0.9948, -0.3314, -0.8475, 0.8452, 0.2427],
  22152. [ 0.5449, -0.4638, 1.6773, -0.2547, -0.5393, -0.0724, 0.4429, 0.1515],
  22153. [ 0.4045, -0.5464, 1.6156, -0.0932, -0.4692, -0.2328, 0.0977, 0.1593]],
  22154. device='cuda:0', grad_fn=<AddmmBackward>)
  22155. landmarks are: tensor([[[ 0.5912, -0.3467, 1.2533, -1.0388, -0.1150, -1.3005, 0.3353,
  22156. 0.3854],
  22157. [ 0.5879, -0.4583, 1.7961, 0.0313, -0.4672, -0.1612, 0.4393,
  22158. 0.0313],
  22159. [ 0.6119, -0.3927, 1.6979, -0.7925, -0.2536, -1.3698, 0.4282,
  22160. -0.0368],
  22161. [ 0.6438, -0.3936, 2.0039, -0.2690, -0.2940, -0.6231, 1.0618,
  22162. 0.4119],
  22163. [ 0.5773, -0.4105, 1.8192, -0.1304, -0.5307, 0.0467, 0.5721,
  22164. 0.2237],
  22165. [ 0.6026, -0.4032, 1.7326, -0.7771, -0.3691, -0.8617, 0.9704,
  22166. 0.3050],
  22167. [ 0.5783, -0.4306, 1.8114, -0.1515, -0.6617, -0.1268, 0.4851,
  22168. 0.0727],
  22169. [ 0.5513, -0.4467, 1.7095, -0.0303, -0.4788, -0.2921, 0.1692,
  22170. 0.1854]]], device='cuda:0')
  22171. loss_train_step before backward: tensor(0.0106, device='cuda:0', grad_fn=<MseLossBackward>)
  22172. loss_train_step after backward: tensor(0.0106, device='cuda:0', grad_fn=<MseLossBackward>)
  22173. loss_train: 2.2540173502638936
  22174. step: 69
  22175. running loss: 0.032666918119766575
  22176. Train Steps: 69/90 Loss: 0.0327 torch.Size([8, 600, 800])
  22177. torch.Size([8, 8])
  22178. tensor([[0.6147, 0.4026, 0.6600, 0.2467, 0.4088, 0.2150, 0.5489, 0.5773],
  22179. [0.6265, 0.4088, 0.8025, 0.1850, 0.4163, 0.2500, 0.6290, 0.4947],
  22180. [0.6346, 0.4086, 0.7938, 0.5500, 0.3962, 0.4867, 0.7343, 0.5702],
  22181. [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927],
  22182. [0.6213, 0.4131, 0.8438, 0.3550, 0.3513, 0.4400, 0.5716, 0.5123],
  22183. [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050],
  22184. [0.6286, 0.4086, 0.8408, 0.2801, 0.4163, 0.2800, 0.6725, 0.5393],
  22185. [0.6346, 0.4144, 0.9088, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899]],
  22186. device='cuda:0', dtype=torch.float64)
  22187. predictions are: tensor([[ 0.2508, -0.5467, 0.7966, -1.1588, -0.2903, -1.3747, 0.0383, 0.3953],
  22188. [ 0.5413, -0.4319, 1.3898, -1.3634, -0.2613, -1.1370, 0.5943, 0.0810],
  22189. [ 0.6342, -0.3566, 1.4557, 0.1665, -0.3822, -0.0786, 0.7992, 0.3457],
  22190. [ 0.5728, -0.4382, 1.8137, -0.3345, -0.5790, -0.2428, 0.4343, 0.0768],
  22191. [ 0.9392, -0.1153, 1.6216, -0.5804, -0.5492, -0.3375, 0.1921, 0.1121],
  22192. [ 0.1582, -0.6907, 1.7376, -0.2148, -0.5145, -0.3778, 0.1450, 0.0500],
  22193. [ 0.3844, -0.5329, 1.5189, -0.9580, -0.1783, -1.0164, 0.7031, 0.2159],
  22194. [ 0.4851, -0.4767, 1.8267, -0.0919, -0.3050, -0.2376, 0.7236, 0.2378]],
  22195. device='cuda:0', grad_fn=<AddmmBackward>)
  22196. landmarks are: tensor([[[ 5.5318e-01, -4.2640e-01, 7.6259e-01, -1.1466e+00, -3.9792e-01,
  22197. -1.2928e+00, 2.4936e-01, 3.8081e-01],
  22198. [ 6.0785e-01, -3.9761e-01, 1.4208e+00, -1.4314e+00, -3.6328e-01,
  22199. -1.1312e+00, 6.1950e-01, -9.2270e-04],
  22200. [ 6.4542e-01, -3.9842e-01, 1.3804e+00, 2.5450e-01, -4.5566e-01,
  22201. -3.8029e-02, 1.1057e+00, 3.4780e-01],
  22202. [ 6.0095e-01, -4.3453e-01, 1.8480e+00, -3.5366e-01, -6.4619e-01,
  22203. -2.6128e-01, 6.5240e-01, -9.9401e-03],
  22204. [ 5.8406e-01, -3.7783e-01, 1.6113e+00, -6.4619e-01, -6.6351e-01,
  22205. -2.5358e-01, 3.5423e-01, 8.0233e-02],
  22206. [ 5.8435e-01, -4.4657e-01, 1.8423e+00, -1.9969e-01, -5.9423e-01,
  22207. -3.9985e-01, 4.2194e-01, 4.6651e-02],
  22208. [ 6.1742e-01, -3.9842e-01, 1.5975e+00, -9.9214e-01, -3.6328e-01,
  22209. -9.9261e-01, 8.2047e-01, 2.0505e-01],
  22210. [ 6.4542e-01, -3.7194e-01, 1.9115e+00, -1.3041e-01, -5.0762e-01,
  22211. -2.8437e-01, 1.0033e+00, 4.3864e-01]]], device='cuda:0')
  22212. loss_train_step before backward: tensor(0.0203, device='cuda:0', grad_fn=<MseLossBackward>)
  22213. loss_train_step after backward: tensor(0.0203, device='cuda:0', grad_fn=<MseLossBackward>)
  22214. loss_train: 2.2742685051634908
  22215. step: 70
  22216. running loss: 0.032489550073764153
  22217. Train Steps: 70/90 Loss: 0.0325 torch.Size([8, 600, 800])
  22218. torch.Size([8, 8])
  22219. tensor([[0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927],
  22220. [0.6179, 0.4008, 0.7505, 0.2678, 0.4368, 0.1891, 0.5831, 0.5263],
  22221. [0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407],
  22222. [0.6137, 0.4038, 0.8563, 0.4050, 0.3813, 0.2550, 0.5106, 0.4954],
  22223. [0.6135, 0.3994, 0.7913, 0.3050, 0.3625, 0.3050, 0.5837, 0.5050],
  22224. [0.6204, 0.4110, 0.7913, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367],
  22225. [0.6076, 0.3953, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954],
  22226. [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413]],
  22227. device='cuda:0', dtype=torch.float64)
  22228. predictions are: tensor([[ 0.5122, -0.4982, 1.8205, -0.3142, -0.5797, -0.0591, 0.6661, 0.1336],
  22229. [ 0.7927, -0.2415, 1.0857, -1.1362, -0.2013, -1.2924, 0.4452, 0.1942],
  22230. [ 0.3741, -0.4858, 1.5175, -0.3482, -0.5324, -0.6665, 0.0663, 0.2776],
  22231. [ 0.2747, -0.5748, 1.5718, -0.4627, -0.4041, -0.8664, 0.1605, 0.1826],
  22232. [ 0.2877, -0.5945, 1.3762, -0.9632, -0.5021, -0.6231, 0.5349, 0.1469],
  22233. [ 0.4169, -0.4920, 1.3689, -1.1018, -0.2625, -0.9907, 0.6163, 0.2166],
  22234. [ 0.5410, -0.4205, 1.5504, -0.5371, -0.5861, -0.2351, 0.3069, 0.0715],
  22235. [ 0.4781, -0.5026, 1.9161, -0.2707, -0.1580, -0.6887, 0.9722, 0.1813]],
  22236. device='cuda:0', grad_fn=<AddmmBackward>)
  22237. landmarks are: tensor([[[ 0.6010, -0.4345, 1.8480, -0.3537, -0.6462, -0.2613, 0.6524,
  22238. -0.0099],
  22239. [ 0.5680, -0.4345, 1.1806, -1.0490, -0.2683, -1.4127, 0.4074,
  22240. 0.1449],
  22241. [ 0.5470, -0.3952, 1.6377, -0.4253, -0.6289, -0.8079, 0.0249,
  22242. 0.2116],
  22243. [ 0.5485, -0.4209, 1.6691, -0.4152, -0.5249, -1.1081, 0.0725,
  22244. 0.0021],
  22245. [ 0.5477, -0.4413, 1.3688, -0.8771, -0.6115, -0.8771, 0.4104,
  22246. 0.0467],
  22247. [ 0.5796, -0.3878, 1.3688, -1.0542, -0.4095, -1.1312, 0.5894,
  22248. 0.1929],
  22249. [ 0.5206, -0.4603, 1.4670, -0.4768, -0.6577, -0.3998, 0.1836,
  22250. 0.0021],
  22251. [ 0.6224, -0.4345, 1.9404, -0.2921, -0.3171, -0.8771, 1.0655,
  22252. 0.2142]]], device='cuda:0')
  22253. loss_train_step before backward: tensor(0.0152, device='cuda:0', grad_fn=<MseLossBackward>)
  22254. loss_train_step after backward: tensor(0.0152, device='cuda:0', grad_fn=<MseLossBackward>)
  22255. loss_train: 2.289423213340342
  22256. step: 71
  22257. running loss: 0.03224539737099073
  22258. Train Steps: 71/90 Loss: 0.0322 torch.Size([8, 600, 800])
  22259. torch.Size([8, 8])
  22260. tensor([[0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617],
  22261. [0.6192, 0.4128, 0.8513, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633],
  22262. [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413],
  22263. [0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6063, 0.6083],
  22264. [0.6125, 0.4035, 0.7825, 0.3100, 0.3463, 0.4900, 0.5832, 0.5637],
  22265. [0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626],
  22266. [0.6161, 0.4024, 0.8662, 0.4683, 0.4935, 0.5364, 0.6063, 0.5567],
  22267. [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960]],
  22268. device='cuda:0', dtype=torch.float64)
  22269. predictions are: tensor([[ 0.4479, -0.4506, 1.6853, -0.1913, -0.3362, -0.1246, 0.2974, 0.2190],
  22270. [ 0.4831, -0.4276, 1.6233, 0.2123, -0.4058, -0.0915, 0.4055, 0.2812],
  22271. [ 0.6317, -0.3702, 1.9464, -0.4200, -0.3204, -1.1019, 0.9239, 0.0278],
  22272. [ 0.7079, -0.2363, 1.5972, -0.7340, -0.6275, -0.5334, 0.4059, 0.3556],
  22273. [ 0.3797, -0.4968, 1.3329, -0.9653, -0.6206, -0.1962, 0.4695, 0.2560],
  22274. [ 0.6649, -0.3512, 1.5966, 0.2881, -0.5217, -0.2935, 0.5200, 0.0120],
  22275. [ 0.5794, -0.3805, 1.7459, -0.2830, -0.0878, -0.1662, 0.4798, 0.1850],
  22276. [ 0.6714, -0.3518, 1.7962, -0.2419, -0.2278, -0.1438, 0.3156, 0.0287]],
  22277. device='cuda:0', grad_fn=<AddmmBackward>)
  22278. landmarks are: tensor([[[ 0.5637, -0.4143, 1.7519, -0.0787, -0.3055, -0.0149, 0.3758,
  22279. 0.3084],
  22280. [ 0.5742, -0.3792, 1.6460, 0.3084, -0.3460, 0.1467, 0.4162,
  22281. 0.3161],
  22282. [ 0.6224, -0.4345, 1.9404, -0.2921, -0.3171, -0.8771, 1.0655,
  22283. 0.2142],
  22284. [ 0.5774, -0.3868, 1.6286, -0.5692, -0.6462, -0.2767, 0.5143,
  22285. 0.5239],
  22286. [ 0.5432, -0.4224, 1.3284, -0.8540, -0.6866, -0.0226, 0.4077,
  22287. 0.3177],
  22288. [ 0.6105, -0.4293, 1.5824, 0.5239, -0.4730, -0.0380, 0.5025,
  22289. -0.1492],
  22290. [ 0.5598, -0.4273, 1.7152, -0.1227, -0.0065, 0.1917, 0.5143,
  22291. 0.2853],
  22292. [ 0.5205, -0.4577, 1.7326, -0.1304, -0.1862, 0.0447, 0.1301,
  22293. 0.0051]]], device='cuda:0')
  22294. loss_train_step before backward: tensor(0.0154, device='cuda:0', grad_fn=<MseLossBackward>)
  22295. loss_train_step after backward: tensor(0.0154, device='cuda:0', grad_fn=<MseLossBackward>)
  22296. loss_train: 2.304807144217193
  22297. step: 72
  22298. running loss: 0.032011210336349905
  22299.  
  22300. Train Steps: 72/90 Loss: 0.0320 torch.Size([8, 600, 800])
  22301. torch.Size([8, 8])
  22302. tensor([[0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750],
  22303. [0.6265, 0.4071, 0.8875, 0.3367, 0.3975, 0.3350, 0.6312, 0.5250],
  22304. [0.6207, 0.4081, 0.7662, 0.2067, 0.3962, 0.3200, 0.6312, 0.5300],
  22305. [0.6284, 0.4093, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167],
  22306. [0.6293, 0.4097, 0.8800, 0.2517, 0.5262, 0.2600, 0.7430, 0.5378],
  22307. [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896],
  22308. [ nan, nan, 0.7240, 0.2722, 0.3900, 0.2567, 0.5168, 0.5933],
  22309. [0.6263, 0.4065, 0.9038, 0.4317, 0.3588, 0.4550, 0.6325, 0.5250]],
  22310. device='cuda:0', dtype=torch.float64)
  22311. predictions are: tensor([[ 0.5122, -0.4145, 1.0718, -0.8660, -0.4378, -0.9175, 0.5208, 0.2975],
  22312. [ 0.6879, -0.2902, 1.8188, -0.6282, -0.4061, -0.6229, 0.5794, 0.0858],
  22313. [ 0.7078, -0.2872, 1.2850, -1.0183, -0.4095, -0.7490, 0.5825, 0.1415],
  22314. [ 0.7138, -0.3128, 1.8883, 0.1094, -0.5699, -0.3284, 0.4986, 0.0147],
  22315. [ 0.7631, -0.2945, 1.7644, -0.9269, 0.1532, -1.1395, 1.0988, 0.0832],
  22316. [ 0.6894, -0.2747, 1.4439, -0.3181, -0.5894, -0.7208, 0.0107, 0.2779],
  22317. [-2.3025, -2.3297, 1.0130, -1.0680, -0.5892, -0.9831, 0.0336, 0.3642],
  22318. [ 0.8791, -0.1854, 1.9073, -0.1095, -0.5426, -0.0348, 0.5527, 0.0630]],
  22319. device='cuda:0', grad_fn=<AddmmBackward>)
  22320. landmarks are: tensor([[[ 0.5805, -0.3818, 1.0282, -1.1774, -0.4903, -0.9310, 0.5894,
  22321. 0.3700],
  22322. [ 0.6078, -0.4057, 1.8134, -0.7309, -0.4499, -0.7386, 0.6298,
  22323. 0.1390],
  22324. [ 0.5809, -0.4011, 1.2533, -1.3313, -0.4557, -0.8079, 0.6298,
  22325. 0.1621],
  22326. [ 0.6164, -0.3956, 1.8249, -0.1150, -0.6000, -0.5076, 0.5836,
  22327. 0.1005],
  22328. [ 0.6207, -0.3936, 1.7788, -1.1235, 0.1448, -1.0850, 1.1459,
  22329. 0.1982],
  22330. [ 0.5485, -0.3997, 1.4445, -0.4895, -0.6000, -0.8309, 0.1878,
  22331. 0.4374],
  22332. [-2.2859, -2.2859, 1.0580, -1.0288, -0.4845, -1.1004, 0.1011,
  22333. 0.4543],
  22334. [ 0.6070, -0.4085, 1.8885, -0.2921, -0.6289, -0.1843, 0.6356,
  22335. 0.1390]]], device='cuda:0')
  22336. loss_train_step before backward: tensor(0.0139, device='cuda:0', grad_fn=<MseLossBackward>)
  22337. loss_train_step after backward: tensor(0.0139, device='cuda:0', grad_fn=<MseLossBackward>)
  22338. loss_train: 2.318676040507853
  22339. step: 73
  22340. running loss: 0.03176268548640895
  22341. Train Steps: 73/90 Loss: 0.0318 torch.Size([8, 600, 800])
  22342. torch.Size([8, 8])
  22343. tensor([[0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600],
  22344. [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402],
  22345. [0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118],
  22346. [0.6164, 0.4102, 0.8850, 0.4867, 0.4213, 0.5983, 0.5609, 0.5038],
  22347. [0.6109, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567],
  22348. [0.6230, 0.4152, 0.7588, 0.2283, 0.4012, 0.2883, 0.6200, 0.5767],
  22349. [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309],
  22350. [0.6078, 0.4033, 0.8019, 0.3055, 0.3450, 0.4200, 0.6025, 0.5550]],
  22351. device='cuda:0', dtype=torch.float64)
  22352. predictions are: tensor([[ 4.7113e-01, -4.8745e-01, 1.8539e+00, 3.2894e-01, -4.1775e-01,
  22353. -7.5153e-03, 5.0597e-01, 2.3463e-01],
  22354. [ 5.6107e-01, -4.3712e-01, 1.7635e+00, 2.1942e-01, -4.4146e-01,
  22355. -9.3449e-02, 4.3632e-01, 6.4832e-02],
  22356. [ 5.3550e-01, -3.7773e-01, 1.2046e+00, -7.6963e-01, -3.0734e-01,
  22357. -1.2585e+00, 3.2494e-01, 4.1616e-01],
  22358. [ 7.6130e-01, -2.6534e-01, 1.8936e+00, 1.0186e-03, -2.6857e-01,
  22359. 4.8667e-01, 3.9008e-01, 1.2859e-01],
  22360. [ 3.2546e-01, -5.6136e-01, 1.1409e+00, -1.3242e+00, -5.3331e-01,
  22361. -1.1989e+00, 3.7836e-01, 1.0678e-01],
  22362. [ 3.6444e-01, -5.2857e-01, 1.3846e+00, -1.0845e+00, -5.1766e-01,
  22363. -1.0753e+00, 6.7351e-01, 2.3173e-01],
  22364. [ 5.3250e-01, -4.3660e-01, 1.9654e+00, 2.5421e-03, -4.6362e-01,
  22365. -5.2328e-01, 7.3296e-01, 3.7710e-02],
  22366. [ 5.8066e-01, -4.1115e-01, 1.5945e+00, -8.2236e-01, -6.5597e-01,
  22367. -3.0854e-01, 5.7481e-01, 1.6315e-01]], device='cuda:0',
  22368. grad_fn=<AddmmBackward>)
  22369. landmarks are: tensor([[[ 0.5883, -0.3594, 1.7557, 0.2545, -0.4152, -0.0611, 0.3353,
  22370. 0.3007],
  22371. [ 0.5780, -0.4565, 1.6221, 0.2532, -0.3728, -0.1718, 0.4357,
  22372. 0.2091],
  22373. [ 0.6077, -0.3226, 0.9993, -0.6462, -0.2651, -1.3082, 0.2946,
  22374. 0.5401],
  22375. [ 0.5614, -0.3912, 1.8018, -0.0380, -0.3402, 0.4778, 0.3049,
  22376. 0.0412],
  22377. [ 0.5356, -0.4217, 1.0339, -1.4776, -0.5076, -1.1081, 0.4219,
  22378. 0.2853],
  22379. [ 0.5915, -0.3682, 1.2187, -1.2313, -0.4326, -0.9541, 0.5778,
  22380. 0.3777],
  22381. [ 0.6520, -0.3623, 1.8885, 0.0313, -0.5538, -0.5384, 0.6926,
  22382. 0.1661],
  22383. [ 0.5215, -0.4232, 1.4182, -0.8747, -0.6924, -0.3460, 0.4970,
  22384. 0.2776]]], device='cuda:0')
  22385. loss_train_step before backward: tensor(0.0116, device='cuda:0', grad_fn=<MseLossBackward>)
  22386. loss_train_step after backward: tensor(0.0116, device='cuda:0', grad_fn=<MseLossBackward>)
  22387. loss_train: 2.3302313135936856
  22388. step: 74
  22389. running loss: 0.03148961234586062
  22390. Train Steps: 74/90 Loss: 0.0315 torch.Size([8, 600, 800])
  22391. torch.Size([8, 8])
  22392. tensor([[0.6129, 0.4063, 0.8738, 0.5250, 0.4313, 0.4733, 0.5230, 0.5874],
  22393. [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466],
  22394. [0.6212, 0.4159, 0.8675, 0.5783, 0.4088, 0.4317, 0.5613, 0.5917],
  22395. [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837],
  22396. [0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183],
  22397. [0.6098, 0.3991, 0.8638, 0.4717, 0.4263, 0.4967, 0.5212, 0.5650],
  22398. [0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343],
  22399. [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466]],
  22400. device='cuda:0', dtype=torch.float64)
  22401. predictions are: tensor([[ 0.7608, -0.2662, 1.7774, 0.1798, -0.3807, -0.0089, 0.3839, 0.2495],
  22402. [ 0.3894, -0.5370, 1.4442, -0.7026, -0.7186, -0.3379, 0.3447, 0.1872],
  22403. [ 0.4860, -0.4275, 1.6876, 0.2732, -0.4168, -0.2713, 0.4909, 0.4333],
  22404. [ 0.5600, -0.3997, 1.0086, -1.0654, -0.6472, -1.0534, 0.2950, 0.2712],
  22405. [ 0.7111, -0.3574, 1.9038, -0.0752, -0.6639, -0.4627, 0.6405, 0.0408],
  22406. [ 0.5216, -0.4450, 1.8388, -0.1305, -0.3204, 0.0560, 0.3731, 0.1311],
  22407. [ 0.4902, -0.5181, 1.9518, -0.7413, -0.3985, -0.9655, 1.0865, 0.0022],
  22408. [ 0.5865, -0.3897, 1.7734, 0.0111, -0.2692, 0.1866, 0.3871, 0.1497]],
  22409. device='cuda:0', grad_fn=<AddmmBackward>)
  22410. landmarks are: tensor([[[ 0.5453, -0.4091, 1.7499, 0.1390, -0.2940, -0.0996, 0.1300,
  22411. 0.4272],
  22412. [ 0.5076, -0.4443, 1.2337, -0.5023, -0.6808, -0.3614, 0.0866,
  22413. 0.2386],
  22414. [ 0.5836, -0.3649, 1.7210, 0.3854, -0.3979, -0.2921, 0.3065,
  22415. 0.4470],
  22416. [ 0.5629, -0.3871, 0.7243, -0.9581, -0.5827, -0.9849, 0.1288,
  22417. 0.4103],
  22418. [ 0.5944, -0.4543, 1.8018, 0.0082, -0.6058, -0.4306, 0.4162,
  22419. 0.1082],
  22420. [ 0.5308, -0.4425, 1.7037, -0.1073, -0.3171, 0.0082, 0.1217,
  22421. 0.3238],
  22422. [ 0.6125, -0.4369, 1.9173, -0.5384, -0.2594, -0.8386, 0.9741,
  22423. 0.1821],
  22424. [ 0.5178, -0.4480, 1.6748, 0.0620, -0.2767, 0.2083, 0.1067,
  22425. 0.2386]]], device='cuda:0')
  22426. loss_train_step before backward: tensor(0.0163, device='cuda:0', grad_fn=<MseLossBackward>)
  22427. loss_train_step after backward: tensor(0.0163, device='cuda:0', grad_fn=<MseLossBackward>)
  22428. loss_train: 2.34652538318187
  22429. step: 75
  22430. running loss: 0.0312870051090916
  22431. Train Steps: 75/90 Loss: 0.0313 torch.Size([8, 600, 800])
  22432. torch.Size([8, 8])
  22433. tensor([[0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426],
  22434. [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402],
  22435. [0.6353, 0.4128, 0.8488, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540],
  22436. [0.6198, 0.4130, 0.8762, 0.4117, 0.3650, 0.4900, 0.5707, 0.5103],
  22437. [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426],
  22438. [0.6364, 0.4154, 0.8938, 0.3717, 0.4500, 0.2583, 0.6448, 0.5285],
  22439. [0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153],
  22440. [0.6122, 0.3993, 0.8738, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927]],
  22441. device='cuda:0', dtype=torch.float64)
  22442. predictions are: tensor([[ 0.8955, -0.1829, 1.1701, -1.0772, -0.3697, -1.3048, 0.5593, 0.1958],
  22443. [ 0.7101, -0.3358, 1.6189, 0.2893, -0.5472, -0.0433, 0.4882, 0.1745],
  22444. [-0.0980, -0.8081, 1.4499, -1.2712, -0.0374, -1.2084, 0.8850, 0.3341],
  22445. [ 0.5643, -0.3828, 1.7756, -0.2794, -0.7242, -0.0340, 0.4257, 0.1475],
  22446. [ 0.3468, -0.5400, 1.7338, -0.1671, -0.7472, -0.5945, 0.3265, 0.2328],
  22447. [ 0.3623, -0.5074, 1.7292, -0.6107, -0.3487, -1.0112, 0.6881, 0.2868],
  22448. [ 0.6102, -0.4003, 1.6497, 0.1532, -0.3328, 0.2402, 0.4762, 0.1845],
  22449. [ 0.4357, -0.5049, 1.7483, -0.0460, -0.4219, 0.1342, 0.1869, 0.0424]],
  22450. device='cuda:0', grad_fn=<AddmmBackward>)
  22451. landmarks are: tensor([[[ 0.5784, -0.4153, 1.2972, -1.2541, -0.2265, -1.4553, 0.4741,
  22452. 0.2203],
  22453. [ 0.5780, -0.4565, 1.6221, 0.2532, -0.3728, -0.1718, 0.4357,
  22454. 0.2091],
  22455. [ 0.6487, -0.3792, 1.6344, -1.0850, 0.2659, -1.5397, 0.8059,
  22456. 0.2730],
  22457. [ 0.5771, -0.3784, 1.7614, -0.3844, -0.6000, -0.0226, 0.3501,
  22458. 0.0712],
  22459. [ 0.5788, -0.4135, 1.8214, -0.2408, -0.6039, -0.7854, 0.4115,
  22460. 0.2203],
  22461. [ 0.6537, -0.3671, 1.8423, -0.5692, -0.2074, -1.0927, 0.6926,
  22462. 0.1554],
  22463. [ 0.5548, -0.4682, 1.7309, 0.0966, -0.1394, 0.1757, 0.5045,
  22464. 0.0942],
  22465. [ 0.5417, -0.4417, 1.7499, -0.1304, -0.1994, -0.0324, 0.0951,
  22466. -0.0099]]], device='cuda:0')
  22467. loss_train_step before backward: tensor(0.0295, device='cuda:0', grad_fn=<MseLossBackward>)
  22468. loss_train_step after backward: tensor(0.0295, device='cuda:0', grad_fn=<MseLossBackward>)
  22469. loss_train: 2.376007336191833
  22470. step: 76
  22471. running loss: 0.03126325442357675
  22472.  
  22473. Train Steps: 76/90 Loss: 0.0313 torch.Size([8, 600, 800])
  22474. torch.Size([8, 8])
  22475. tensor([[0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083],
  22476. [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153],
  22477. [0.6048, 0.3928, 0.8538, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719],
  22478. [0.6098, 0.3991, 0.8638, 0.4717, 0.4263, 0.4967, 0.5212, 0.5650],
  22479. [0.6170, 0.4102, 0.7468, 0.3695, 0.3463, 0.3767, 0.5238, 0.5823],
  22480. [0.6156, 0.4125, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084],
  22481. [0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748],
  22482. [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051]],
  22483. device='cuda:0', dtype=torch.float64)
  22484. predictions are: tensor([[ 6.2599e-01, -3.7787e-01, 1.7324e+00, -1.4953e-03, -5.6675e-01,
  22485. -2.6756e-01, 4.9518e-01, 4.3112e-01],
  22486. [ 6.0591e-01, -4.0288e-01, 1.4897e+00, -9.1640e-01, -6.3901e-01,
  22487. -7.7692e-01, 4.3591e-01, 9.2140e-02],
  22488. [ 3.4257e-01, -5.5844e-01, 1.6558e+00, 1.9338e-01, -4.5376e-01,
  22489. -9.7340e-03, 3.2920e-01, 2.5710e-01],
  22490. [ 4.5453e-01, -4.8078e-01, 1.7472e+00, -1.1293e-01, -2.6900e-01,
  22491. 2.1217e-03, 2.5967e-01, 2.2156e-01],
  22492. [ 6.1328e-01, -3.6578e-01, 1.3096e+00, -7.1024e-01, -6.7487e-01,
  22493. -5.3066e-01, 2.5872e-01, 3.9249e-01],
  22494. [ 5.6314e-01, -4.0142e-01, 1.8560e+00, -1.5741e-01, -1.9763e-01,
  22495. 2.6406e-01, 4.0827e-01, 1.1360e-01],
  22496. [ 5.6209e-01, -4.2365e-01, 1.7095e+00, 1.2593e-01, -4.6443e-01,
  22497. -4.5895e-01, 9.4878e-01, 2.4540e-01],
  22498. [ 7.4072e-01, -3.3453e-01, 1.8828e+00, -1.8974e-01, -5.4737e-01,
  22499. -6.2025e-01, 7.1610e-01, 1.2509e-01]], device='cuda:0',
  22500. grad_fn=<AddmmBackward>)
  22501. landmarks are: tensor([[[ 0.5713, -0.4105, 1.7557, 0.0467, -0.6520, -0.2690, 0.3988,
  22502. 0.5239],
  22503. [ 0.5614, -0.4514, 1.3252, -0.7240, -0.6237, -0.7215, 0.4299,
  22504. 0.0942],
  22505. [ 0.5079, -0.4714, 1.6575, 0.2237, -0.4961, 0.0774, 0.1465,
  22506. -0.1061],
  22507. [ 0.5308, -0.4425, 1.7037, -0.1073, -0.3171, 0.0082, 0.1217,
  22508. 0.3238],
  22509. [ 0.5639, -0.3911, 1.1634, -0.5794, -0.6866, -0.5461, 0.1334,
  22510. 0.4036],
  22511. [ 0.5573, -0.3808, 1.8018, -0.0534, -0.1843, 0.4008, 0.3769,
  22512. 0.0622],
  22513. [ 0.6421, -0.3647, 1.5940, 0.3084, -0.5192, -0.3691, 1.1057,
  22514. 0.3692],
  22515. [ 0.6204, -0.4336, 1.8654, -0.0688, -0.6058, -0.5230, 0.6503,
  22516. 0.0472]]], device='cuda:0')
  22517. loss_train_step before backward: tensor(0.0097, device='cuda:0', grad_fn=<MseLossBackward>)
  22518. loss_train_step after backward: tensor(0.0097, device='cuda:0', grad_fn=<MseLossBackward>)
  22519. loss_train: 2.3856922583654523
  22520. step: 77
  22521. running loss: 0.030983016342408472
  22522. Train Steps: 77/90 Loss: 0.0310 torch.Size([8, 600, 800])
  22523. torch.Size([8, 8])
  22524. tensor([[0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167],
  22525. [0.6201, 0.3970, 0.8413, 0.4950, 0.4413, 0.5183, 0.6088, 0.5400],
  22526. [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609],
  22527. [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212],
  22528. [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567],
  22529. [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550],
  22530. [0.6133, 0.4066, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358],
  22531. [0.6064, 0.3953, 0.8738, 0.4417, 0.3663, 0.4683, 0.5511, 0.5416]],
  22532. device='cuda:0', dtype=torch.float64)
  22533. predictions are: tensor([[ 0.4816, -0.4621, 1.4665, -0.6807, -0.7312, -0.5035, 0.4855, 0.4826],
  22534. [ 0.3986, -0.5600, 1.7154, 0.1935, -0.3803, 0.1976, 0.4821, 0.1684],
  22535. [ 0.6346, -0.3935, 1.7428, -0.9900, 0.1062, -1.3640, 1.2291, 0.2949],
  22536. [ 0.6019, -0.4224, 1.4777, -0.9274, -0.5007, -0.9511, 0.6558, 0.1184],
  22537. [ 0.4079, -0.5341, 1.7268, -0.0303, -0.6630, -0.3353, 0.3375, 0.2916],
  22538. [ 0.5138, -0.4476, 1.8194, -0.0117, -0.0904, 0.0841, 0.4128, 0.2011],
  22539. [ 0.3823, -0.5219, 0.9403, -0.9047, -0.5712, -1.1431, 0.1388, 0.1978],
  22540. [ 0.5692, -0.4489, 1.8115, -0.0553, -0.6830, -0.0256, 0.2668, 0.1434]],
  22541. device='cuda:0', grad_fn=<AddmmBackward>)
  22542. landmarks are: tensor([[[ 5.7685e-01, -3.8992e-01, 1.3861e+00, -7.7706e-01, -5.8845e-01,
  22543. -5.4611e-01, 5.0277e-01, 5.6243e-01],
  22544. [ 5.7812e-01, -4.5219e-01, 1.5998e+00, 4.6189e-04, -2.4781e-01,
  22545. 1.0824e-01, 5.2587e-01, 2.0831e-01],
  22546. [ 6.3718e-01, -3.5749e-01, 1.6633e+00, -1.2082e+00, 2.1986e-01,
  22547. -1.2467e+00, 1.1313e+00, 3.0505e-01],
  22548. [ 6.1248e-01, -4.3453e-01, 1.4308e+00, -1.1384e+00, -4.2133e-01,
  22549. -1.0031e+00, 7.1897e-01, 1.2136e-01],
  22550. [ 5.4324e-01, -4.3364e-01, 1.7095e+00, -1.7660e-01, -5.9423e-01,
  22551. -4.8453e-01, 3.0069e-01, 2.8530e-01],
  22552. [ 5.3508e-01, -4.3212e-01, 1.7326e+00, -1.3811e-01, 4.6329e-02,
  22553. 1.0491e-01, 4.1617e-01, 2.7760e-01],
  22554. [ 5.4700e-01, -4.0808e-01, 8.4919e-01, -1.0773e+00, -5.3072e-01,
  22555. -1.1620e+00, 9.1240e-02, 1.8903e-01],
  22556. [ 5.1490e-01, -4.6028e-01, 1.7499e+00, -2.4588e-01, -5.9423e-01,
  22557. -1.2271e-01, 2.5964e-01, 2.1549e-01]]], device='cuda:0')
  22558. loss_train_step before backward: tensor(0.0095, device='cuda:0', grad_fn=<MseLossBackward>)
  22559. loss_train_step after backward: tensor(0.0095, device='cuda:0', grad_fn=<MseLossBackward>)
  22560. loss_train: 2.3952217008918524
  22561. step: 78
  22562. running loss: 0.03070797052425452
  22563. Train Steps: 78/90 Loss: 0.0307 torch.Size([8, 600, 800])
  22564. torch.Size([8, 8])
  22565. tensor([[0.6127, 0.4115, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495],
  22566. [0.6179, 0.4118, 0.7278, 0.4237, 0.3588, 0.3400, 0.5675, 0.5917],
  22567. [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378],
  22568. [0.6339, 0.4123, 0.8638, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436],
  22569. [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103],
  22570. [0.6314, 0.4107, 0.8750, 0.5100, 0.3788, 0.4900, 0.7121, 0.5864],
  22571. [0.6115, 0.4005, 0.8838, 0.3867, 0.3763, 0.4700, 0.5800, 0.5550],
  22572. [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600]],
  22573. device='cuda:0', dtype=torch.float64)
  22574. predictions are: tensor([[ 0.5093, -0.4510, 1.1529, -0.9447, -0.5343, -0.9558, 0.1538, 0.2929],
  22575. [ 0.1427, -0.6651, 1.3357, -0.4493, -0.5945, -0.9460, 0.1704, 0.4914],
  22576. [ 0.6095, -0.4259, 1.8378, 0.1008, -0.3755, 0.2174, 0.8572, 0.2546],
  22577. [ 0.6198, -0.4002, 1.7901, 0.1931, -0.3497, 0.1815, 0.8804, 0.2716],
  22578. [ 0.6056, -0.3894, 1.8245, -0.1414, -0.4505, -0.2102, 0.1380, 0.1016],
  22579. [ 0.6638, -0.3807, 1.8774, 0.2831, -0.4401, -0.0791, 0.8332, 0.2436],
  22580. [ 0.5925, -0.3957, 1.8014, -0.3808, -0.5423, -0.1721, 0.1716, 0.2022],
  22581. [ 0.5354, -0.4287, 1.1695, -1.0155, -0.5159, -0.9060, 0.4118, 0.2849]],
  22582. device='cuda:0', grad_fn=<AddmmBackward>)
  22583. landmarks are: tensor([[[ 0.5442, -0.3855, 1.0224, -0.9541, -0.6115, -0.9233, 0.1745,
  22584. 0.2522],
  22585. [ 0.5680, -0.3840, 1.0756, -0.3290, -0.6289, -0.7155, 0.3353,
  22586. 0.4470],
  22587. [ 0.6421, -0.3864, 1.7961, 0.0543, -0.4383, 0.2237, 1.2007,
  22588. 0.1982],
  22589. [ 0.6421, -0.3816, 1.7037, 0.1929, -0.4037, 0.2391, 1.1861,
  22590. 0.2249],
  22591. [ 0.5253, -0.4392, 1.7730, -0.2305, -0.4268, -0.1381, 0.1651,
  22592. 0.0712],
  22593. [ 0.6306, -0.3888, 1.7557, 0.0697, -0.5365, -0.0226, 1.0033,
  22594. 0.4226],
  22595. [ 0.5384, -0.4361, 1.7961, -0.4999, -0.5480, -0.1150, 0.3931,
  22596. 0.2776],
  22597. [ 0.5466, -0.4080, 1.0668, -1.1764, -0.6289, -0.7617, 0.4855,
  22598. 0.3007]]], device='cuda:0')
  22599. loss_train_step before backward: tensor(0.0158, device='cuda:0', grad_fn=<MseLossBackward>)
  22600. loss_train_step after backward: tensor(0.0158, device='cuda:0', grad_fn=<MseLossBackward>)
  22601. loss_train: 2.411067308858037
  22602. step: 79
  22603. running loss: 0.03051983935263338
  22604. Train Steps: 79/90 Loss: 0.0305 torch.Size([8, 600, 800])
  22605. torch.Size([8, 8])
  22606. tensor([[0.6147, 0.4026, 0.6600, 0.2467, 0.4088, 0.2150, 0.5489, 0.5773],
  22607. [0.6076, 0.3953, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954],
  22608. [ nan, nan, 0.7097, 0.2346, 0.4250, 0.1850, 0.5175, 0.5583],
  22609. [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6088, 0.5583],
  22610. [0.6265, 0.4091, 0.8950, 0.3533, 0.3600, 0.3967, 0.6295, 0.4901],
  22611. [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012],
  22612. [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483],
  22613. [0.6262, 0.4085, 0.8438, 0.3150, 0.4025, 0.2633, 0.6339, 0.4810]],
  22614. device='cuda:0', dtype=torch.float64)
  22615. predictions are: tensor([[ 0.4540, -0.4703, 0.9067, -0.9209, -0.3107, -1.2274, 0.2563, 0.5059],
  22616. [ 0.5799, -0.4351, 1.6240, -0.3646, -0.6393, -0.2595, 0.2720, 0.1308],
  22617. [-1.5526, -1.8142, 0.9907, -1.0273, -0.3385, -1.2593, 0.1713, 0.4376],
  22618. [ 0.6030, -0.4002, 1.0383, -1.0469, -0.4132, -1.0222, 0.5741, 0.3941],
  22619. [ 0.8599, -0.2390, 1.9401, -0.4291, -0.5709, -0.3774, 0.7098, 0.1473],
  22620. [ 0.8415, -0.3189, 1.8656, 0.4869, -0.5526, -0.0591, 0.6731, 0.1598],
  22621. [ 0.6112, -0.4141, 1.7905, -0.2452, -0.4861, 0.1768, 0.4030, 0.2684],
  22622. [ 0.8313, -0.2715, 1.6857, -0.6763, -0.3634, -0.9058, 0.6089, 0.0587]],
  22623. device='cuda:0', grad_fn=<AddmmBackward>)
  22624. landmarks are: tensor([[[ 5.5318e-01, -4.2640e-01, 7.6259e-01, -1.1466e+00, -3.9792e-01,
  22625. -1.2928e+00, 2.4936e-01, 3.8081e-01],
  22626. [ 5.2061e-01, -4.6028e-01, 1.4670e+00, -4.7683e-01, -6.5774e-01,
  22627. -3.9985e-01, 1.8356e-01, 2.0831e-03],
  22628. [-2.2859e+00, -2.2859e+00, 9.9216e-01, -1.2021e+00, -3.2286e-01,
  22629. -1.4314e+00, 1.0439e-01, 2.9299e-01],
  22630. [ 5.9919e-01, -3.9684e-01, 9.3067e-01, -1.3497e+00, -4.7298e-01,
  22631. -1.0465e+00, 5.2587e-01, 2.9299e-01],
  22632. [ 6.0774e-01, -3.9646e-01, 1.8480e+00, -6.5389e-01, -6.2309e-01,
  22633. -4.5373e-01, 6.2155e-01, -2.1963e-02],
  22634. [ 6.2488e-01, -4.3518e-01, 1.8018e+00, 2.5450e-01, -6.1732e-01,
  22635. -1.9969e-01, 6.4006e-01, 2.9135e-02],
  22636. [ 5.5456e-01, -4.6205e-01, 1.7788e+00, -4.2294e-01, -5.1917e-01,
  22637. -2.2633e-02, 4.2771e-01, 2.4681e-01],
  22638. [ 6.0641e-01, -3.9900e-01, 1.6113e+00, -8.3095e-01, -4.2679e-01,
  22639. -1.0696e+00, 6.4212e-01, -6.4044e-02]]], device='cuda:0')
  22640. loss_train_step before backward: tensor(0.0272, device='cuda:0', grad_fn=<MseLossBackward>)
  22641. loss_train_step after backward: tensor(0.0272, device='cuda:0', grad_fn=<MseLossBackward>)
  22642. loss_train: 2.4382315184921026
  22643. step: 80
  22644. running loss: 0.030477893981151284
  22645.  
  22646. Train Steps: 80/90 Loss: 0.0305 torch.Size([8, 600, 800])
  22647. torch.Size([8, 8])
  22648. tensor([[0.6249, 0.4142, 0.8350, 0.3283, 0.3613, 0.3700, 0.6188, 0.5400],
  22649. [0.6185, 0.4098, 0.8838, 0.4900, 0.4537, 0.5800, 0.6288, 0.5400],
  22650. [0.6202, 0.4079, 0.8025, 0.2500, 0.3763, 0.3217, 0.6125, 0.5533],
  22651. [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209],
  22652. [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400],
  22653. [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879],
  22654. [0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5837, 0.5583],
  22655. [0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783]],
  22656. device='cuda:0', dtype=torch.float64)
  22657. predictions are: tensor([[ 0.6583, -0.3531, 1.5451, -0.6303, -0.7401, -0.6090, 0.5511, 0.2601],
  22658. [ 0.5577, -0.4344, 1.7469, 0.0396, -0.2320, 0.3542, 0.4915, 0.2582],
  22659. [ 0.5700, -0.4197, 1.3435, -1.0377, -0.5919, -0.7790, 0.5803, 0.3046],
  22660. [ 0.4306, -0.4598, 1.5850, 0.0190, -0.3076, -1.0861, 0.4304, 0.5001],
  22661. [ 0.6816, -0.3647, 1.1919, -1.0613, -0.2055, -1.3733, 0.5535, 0.2396],
  22662. [ 0.5564, -0.4769, 1.8245, -0.1644, -0.5580, 0.2945, 0.5149, 0.0401],
  22663. [ 0.4822, -0.5046, 1.7714, -0.1248, -0.1793, 0.1259, 0.3947, 0.1858],
  22664. [-0.4371, -1.0689, 0.9644, -0.8736, -0.3876, -1.3416, 0.3498, 0.4060]],
  22665. device='cuda:0', grad_fn=<AddmmBackward>)
  22666. landmarks are: tensor([[[ 0.6006, -0.3728, 1.5709, -0.7694, -0.6173, -0.5769, 0.5721,
  22667. 0.2083],
  22668. [ 0.5709, -0.3933, 1.7961, -0.0226, -0.1901, 0.3931, 0.6182,
  22669. 0.2083],
  22670. [ 0.5788, -0.4020, 1.4208, -1.1312, -0.5480, -0.8002, 0.5432,
  22671. 0.2699],
  22672. [ 0.6174, -0.3118, 1.6402, -0.2074, -0.1958, -1.0927, 0.2267,
  22673. 0.5822],
  22674. [ 0.5842, -0.3848, 1.2938, -1.2159, -0.2132, -1.4160, 0.5778,
  22675. 0.2083],
  22676. [ 0.5676, -0.4327, 1.8252, -0.2095, -0.4886, 0.3084, 0.4727,
  22677. -0.0322],
  22678. [ 0.5223, -0.4336, 1.7557, -0.2074, -0.0573, 0.3084, 0.4104,
  22679. 0.2930],
  22680. [ 0.5742, -0.4474, 0.9834, -1.0159, -0.3229, -1.3159, 0.2314,
  22681. 0.3854]]], device='cuda:0')
  22682. loss_train_step before backward: tensor(0.0289, device='cuda:0', grad_fn=<MseLossBackward>)
  22683. loss_train_step after backward: tensor(0.0289, device='cuda:0', grad_fn=<MseLossBackward>)
  22684. loss_train: 2.4671557769179344
  22685. step: 81
  22686. running loss: 0.03045871329528314
  22687. Train Steps: 81/90 Loss: 0.0305 torch.Size([8, 600, 800])
  22688. torch.Size([8, 8])
  22689. tensor([[0.6102, 0.4020, 0.8638, 0.3717, 0.3625, 0.5017, 0.6038, 0.5500],
  22690. [0.6339, 0.4118, 0.7988, 0.5800, 0.3912, 0.4583, 0.7343, 0.5760],
  22691. [0.6190, 0.4135, 0.8000, 0.4883, 0.3566, 0.3647, 0.5613, 0.5900],
  22692. [0.6246, 0.4028, 0.8738, 0.4867, 0.4088, 0.5667, 0.6362, 0.5200],
  22693. [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892],
  22694. [0.6243, 0.4128, 0.7762, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417],
  22695. [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609],
  22696. [0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869]],
  22697. device='cuda:0', dtype=torch.float64)
  22698. predictions are: tensor([[ 0.3561, -0.5817, 1.6280, -0.5200, -0.6140, 0.0832, 0.4891, 0.2815],
  22699. [ 0.6177, -0.4275, 1.4497, 0.2524, -0.5313, -0.1602, 0.7929, 0.3334],
  22700. [ 0.3869, -0.4878, 1.4759, 0.0193, -0.6202, -0.5727, 0.1075, 0.3970],
  22701. [ 0.6651, -0.3817, 1.7857, -0.0360, -0.4305, 0.4036, 0.5908, 0.1152],
  22702. [ 0.1623, -0.6708, 1.5985, -0.0590, -0.7143, -0.5345, 0.0244, 0.3532],
  22703. [ 0.5097, -0.4573, 1.2580, -1.0156, -0.4474, -0.7448, 0.4199, 0.3221],
  22704. [ 0.6839, -0.3833, 1.6283, -1.0961, 0.1755, -1.2620, 1.0173, 0.3348],
  22705. [ 0.6074, -0.4254, 1.3506, -1.1489, -0.2045, -1.2801, 0.4970, 0.0799]],
  22706. device='cuda:0', grad_fn=<AddmmBackward>)
  22707. landmarks are: tensor([[[ 0.5324, -0.4293, 1.7037, -0.5692, -0.6115, 0.0313, 0.5028,
  22708. 0.2545],
  22709. [ 0.6421, -0.3840, 1.4035, 0.3931, -0.4788, -0.1689, 1.1057,
  22710. 0.3745],
  22711. [ 0.5732, -0.3761, 1.4092, -0.0303, -0.6388, -0.6012, 0.3065,
  22712. 0.4393],
  22713. [ 0.5990, -0.4256, 1.7499, -0.0380, -0.3979, 0.3315, 0.6529,
  22714. 0.1159],
  22715. [ 0.5454, -0.4053, 1.6633, -0.1766, -0.6058, -0.5923, 0.1577,
  22716. 0.4357],
  22717. [ 0.5977, -0.3792, 1.2995, -1.0311, -0.5192, -0.8386, 0.5836,
  22718. 0.2160],
  22719. [ 0.6372, -0.3575, 1.6633, -1.2082, 0.2199, -1.2467, 1.1313,
  22720. 0.3050],
  22721. [ 0.6058, -0.3892, 1.4208, -1.0927, -0.1843, -1.4237, 0.6154,
  22722. -0.0370]]], device='cuda:0')
  22723. loss_train_step before backward: tensor(0.0116, device='cuda:0', grad_fn=<MseLossBackward>)
  22724. loss_train_step after backward: tensor(0.0116, device='cuda:0', grad_fn=<MseLossBackward>)
  22725. loss_train: 2.478730901144445
  22726. step: 82
  22727. running loss: 0.030228425623712744
  22728. Train Steps: 82/90 Loss: 0.0302 torch.Size([8, 600, 800])
  22729. torch.Size([8, 8])
  22730. tensor([[0.6205, 0.4016, 0.8350, 0.2717, 0.3987, 0.2550, 0.5787, 0.5133],
  22731. [0.6173, 0.4114, 0.7325, 0.2500, 0.4213, 0.1917, 0.5338, 0.5700],
  22732. [0.6364, 0.4165, 0.9088, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297],
  22733. [0.6229, 0.4198, 0.7662, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783],
  22734. [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402],
  22735. [0.6329, 0.4196, 0.9238, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748],
  22736. [0.6274, 0.4003, 0.8638, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661],
  22737. [0.6201, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044]],
  22738. device='cuda:0', dtype=torch.float64)
  22739. predictions are: tensor([[ 0.1982, -0.6441, 1.4626, -1.0844, -0.4183, -0.8629, 0.2960, 0.2098],
  22740. [ 0.1744, -0.6380, 0.9460, -1.1304, -0.3395, -1.1559, 0.1845, 0.3836],
  22741. [ 0.4829, -0.4581, 1.7923, -0.2788, -0.4446, -0.5581, 0.5883, 0.2422],
  22742. [ 0.5963, -0.3333, 1.1338, -1.0038, -0.1605, -1.0222, 0.3700, 0.4745],
  22743. [ 0.5174, -0.4872, 1.6344, 0.1660, -0.4626, 0.0136, 0.4201, 0.1727],
  22744. [ 0.4432, -0.5566, 1.8790, -0.1467, -0.3513, -0.7157, 0.9092, 0.3290],
  22745. [ 0.6365, -0.4482, 1.6620, 0.3447, -0.5882, 0.0983, 0.6642, 0.0769],
  22746. [ 0.6277, -0.3859, 1.1277, -1.2750, -0.3528, -1.0574, 0.4644, 0.2117]],
  22747. device='cuda:0', grad_fn=<AddmmBackward>)
  22748. landmarks are: tensor([[[ 0.5800, -0.4312, 1.5709, -1.0311, -0.4441, -1.1081, 0.3873,
  22749. 0.0851],
  22750. [ 0.5652, -0.3858, 1.0975, -1.1312, -0.3402, -1.4006, 0.1794,
  22751. 0.3469],
  22752. [ 0.6537, -0.3623, 1.9115, -0.2690, -0.4037, -0.8309, 0.6926,
  22753. 0.1608],
  22754. [ 0.5912, -0.3467, 1.2533, -1.0388, -0.1150, -1.3005, 0.3353,
  22755. 0.3854],
  22756. [ 0.5780, -0.4565, 1.6221, 0.2532, -0.3728, -0.1718, 0.4357,
  22757. 0.2091],
  22758. [ 0.6372, -0.3479, 1.9808, -0.1689, -0.3171, -0.9310, 1.1057,
  22759. 0.3692],
  22760. [ 0.6120, -0.4371, 1.7037, 0.4701, -0.5827, -0.0226, 0.5354,
  22761. -0.1331],
  22762. [ 0.5785, -0.4129, 1.2785, -1.3996, -0.3227, -1.3259, 0.4258,
  22763. 0.0438]]], device='cuda:0')
  22764. loss_train_step before backward: tensor(0.0205, device='cuda:0', grad_fn=<MseLossBackward>)
  22765. loss_train_step after backward: tensor(0.0205, device='cuda:0', grad_fn=<MseLossBackward>)
  22766. loss_train: 2.499280705116689
  22767. step: 83
  22768. running loss: 0.03011181572429746
  22769. Train Steps: 83/90 Loss: 0.0301 torch.Size([8, 600, 800])
  22770. torch.Size([8, 8])
  22771. tensor([[0.6200, 0.4118, 0.8287, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799],
  22772. [0.6201, 0.3970, 0.8413, 0.4950, 0.4413, 0.5183, 0.6088, 0.5400],
  22773. [0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400],
  22774. [0.6200, 0.3999, 0.8653, 0.5207, 0.4100, 0.5125, 0.5975, 0.5103],
  22775. [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220],
  22776. [ nan, nan, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533],
  22777. [0.6087, 0.3976, 0.8337, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300],
  22778. [0.6239, 0.4206, 0.8750, 0.5400, 0.3688, 0.4850, 0.5737, 0.5700]],
  22779. device='cuda:0', dtype=torch.float64)
  22780. predictions are: tensor([[ 0.6278, -0.3582, 1.4195, -0.5272, -0.5106, -0.8733, 0.2839, 0.3572],
  22781. [ 0.6381, -0.4075, 1.6691, -0.0616, -0.2797, 0.1486, 0.6430, 0.1670],
  22782. [ 0.5569, -0.3845, 1.3425, -0.9565, -0.0864, -1.1992, 0.4094, 0.2892],
  22783. [ 0.6495, -0.4110, 1.7332, 0.0540, -0.4006, 0.1699, 0.6011, 0.0674],
  22784. [ 0.7700, -0.2943, 0.8941, -1.3155, -0.2464, -1.4069, 0.3876, 0.2335],
  22785. [-1.6801, -1.9175, 1.2121, -0.9034, -0.5852, -0.9538, 0.2522, 0.3485],
  22786. [ 0.8054, -0.2963, 1.5600, -0.6453, -0.5434, -0.6607, 0.5905, 0.2499],
  22787. [ 0.9652, -0.1676, 1.7356, 0.2161, -0.5124, 0.1078, 0.6526, 0.2850]],
  22788. device='cuda:0', grad_fn=<AddmmBackward>)
  22789. landmarks are: tensor([[[ 5.7790e-01, -3.8397e-01, 1.5420e+00, -4.3064e-01, -5.4226e-01,
  22790. -9.7721e-01, 2.0412e-01, 3.9283e-01],
  22791. [ 5.7812e-01, -4.5219e-01, 1.5998e+00, 4.6189e-04, -2.4781e-01,
  22792. 1.0824e-01, 5.2587e-01, 2.0831e-01],
  22793. [ 5.9107e-01, -3.8879e-01, 1.4727e+00, -9.5412e-01, -9.1917e-02,
  22794. -1.4930e+00, 3.9885e-01, 2.0831e-01],
  22795. [ 5.7777e-01, -4.3888e-01, 1.7107e+00, 1.1921e-01, -3.9207e-01,
  22796. 8.1507e-02, 4.7413e-01, 7.1077e-02],
  22797. [ 5.5664e-01, -4.1601e-01, 9.9353e-01, -1.3313e+00, -2.8245e-01,
  22798. -1.5161e+00, 2.1441e-01, 1.2532e-01],
  22799. [-2.2859e+00, -2.2859e+00, 1.5074e+00, -1.0388e+00, -5.4226e-01,
  22800. -9.8491e-01, 2.1986e-01, 2.6990e-01],
  22801. [ 5.2546e-01, -4.4950e-01, 1.5651e+00, -4.9992e-01, -5.7113e-01,
  22802. -8.4634e-01, 4.5658e-01, 1.6212e-01],
  22803. [ 5.9601e-01, -3.4305e-01, 1.7557e+00, 2.0831e-01, -5.8268e-01,
  22804. -4.5727e-02, 3.6420e-01, 3.4688e-01]]], device='cuda:0')
  22805. loss_train_step before backward: tensor(0.0224, device='cuda:0', grad_fn=<MseLossBackward>)
  22806. loss_train_step after backward: tensor(0.0224, device='cuda:0', grad_fn=<MseLossBackward>)
  22807. loss_train: 2.5217115683481097
  22808. step: 84
  22809. running loss: 0.03002037581366797
  22810.  
  22811. Train Steps: 84/90 Loss: 0.0300 torch.Size([8, 600, 800])
  22812. torch.Size([8, 8])
  22813. tensor([[0.6109, 0.4041, 0.6975, 0.3167, 0.3513, 0.3383, 0.5153, 0.5319],
  22814. [0.6275, 0.4048, 0.8488, 0.2883, 0.4463, 0.2033, 0.6321, 0.5155],
  22815. [0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869],
  22816. [0.6112, 0.4029, 0.8638, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567],
  22817. [0.6115, 0.4081, 0.6725, 0.2433, 0.4088, 0.1933, 0.5167, 0.5544],
  22818. [0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133],
  22819. [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795],
  22820. [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600]],
  22821. device='cuda:0', dtype=torch.float64)
  22822. predictions are: tensor([[ 0.3755, -0.5651, 1.0948, -0.7817, -0.6757, -0.6152, 0.0656, 0.2687],
  22823. [ 0.5311, -0.4885, 1.6368, -0.8579, -0.2738, -1.1269, 0.7426, 0.1522],
  22824. [ 0.5372, -0.4697, 1.4447, -1.0901, -0.2341, -1.2558, 0.6001, 0.1076],
  22825. [ 0.4377, -0.5105, 1.8264, 0.0483, -0.0860, 0.1494, 0.4169, 0.2315],
  22826. [ 0.3859, -0.5386, 0.9109, -1.0271, -0.4228, -1.2534, 0.1069, 0.3340],
  22827. [ 0.7784, -0.3214, 1.5264, -1.0374, -0.1043, -1.2875, 0.6998, 0.2039],
  22828. [ 0.4173, -0.5761, 1.7450, 0.3063, -0.5354, 0.3130, 0.9737, 0.3100],
  22829. [ 0.4244, -0.5139, 1.1205, -0.9749, -0.6412, -0.6767, 0.4314, 0.2827]],
  22830. device='cuda:0', grad_fn=<AddmmBackward>)
  22831. landmarks are: tensor([[[ 0.5359, -0.4193, 0.9358, -0.8232, -0.6635, -0.7232, 0.0943,
  22832. 0.1710],
  22833. [ 0.6126, -0.4161, 1.6344, -0.9541, -0.2247, -1.3467, 0.6339,
  22834. 0.0953],
  22835. [ 0.6058, -0.3892, 1.4208, -1.0927, -0.1843, -1.4237, 0.6154,
  22836. -0.0370],
  22837. [ 0.5371, -0.4252, 1.7037, -0.0688, -0.0342, 0.0620, 0.3758,
  22838. 0.2853],
  22839. [ 0.5385, -0.4010, 0.8203, -1.1620, -0.3979, -1.3929, 0.1005,
  22840. 0.2747],
  22841. [ 0.6085, -0.4084, 1.5536, -1.1466, -0.0746, -1.4853, 0.6298,
  22842. 0.0851],
  22843. [ 0.6355, -0.4080, 1.6113, 0.1852, -0.4730, 0.1467, 0.9996,
  22844. 0.3905],
  22845. [ 0.5466, -0.4080, 1.0668, -1.1764, -0.6289, -0.7617, 0.4855,
  22846. 0.3007]]], device='cuda:0')
  22847. loss_train_step before backward: tensor(0.0115, device='cuda:0', grad_fn=<MseLossBackward>)
  22848. loss_train_step after backward: tensor(0.0115, device='cuda:0', grad_fn=<MseLossBackward>)
  22849. loss_train: 2.5331995971500874
  22850. step: 85
  22851. running loss: 0.029802348201765732
  22852. Train Steps: 85/90 Loss: 0.0298 torch.Size([8, 600, 800])
  22853. torch.Size([8, 8])
  22854. tensor([[0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133],
  22855. [0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400],
  22856. [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482],
  22857. [0.6055, 0.4015, 0.7425, 0.2033, 0.4113, 0.1883, 0.5217, 0.4823],
  22858. [0.6231, 0.3973, 0.8650, 0.3950, 0.3625, 0.3183, 0.5837, 0.5167],
  22859. [0.6163, 0.4114, 0.7650, 0.2017, 0.3763, 0.2867, 0.5631, 0.5071],
  22860. [0.6122, 0.4006, 0.8850, 0.4217, 0.4088, 0.5517, 0.6063, 0.5517],
  22861. [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5463, 0.5800]],
  22862. device='cuda:0', dtype=torch.float64)
  22863. predictions are: tensor([[ 0.3192, -0.5970, 1.7006, 0.2154, -0.0810, 0.0501, 0.5673, 0.2228],
  22864. [ 0.5665, -0.4718, 1.7598, -0.4422, -0.5859, -0.2406, 0.6804, 0.2694],
  22865. [ 0.7055, -0.3615, 1.8326, -0.3187, -0.5280, -0.6149, 0.5665, 0.2291],
  22866. [ 0.2587, -0.6218, 0.9886, -1.2755, -0.3026, -1.5091, 0.3010, 0.1494],
  22867. [ 0.6178, -0.4333, 1.6893, -0.4250, -0.4827, -0.8889, 0.5407, 0.1794],
  22868. [ 0.7328, -0.3638, 1.1332, -1.3629, -0.4475, -1.1089, 0.5656, 0.1385],
  22869. [ 0.4559, -0.5146, 1.7819, -0.2563, -0.3102, 0.2478, 0.6159, 0.2551],
  22870. [ 0.6827, -0.3224, 1.0977, -0.8458, -0.4696, -1.0328, 0.1970, 0.4110]],
  22871. device='cuda:0', grad_fn=<AddmmBackward>)
  22872. landmarks are: tensor([[[ 0.5786, -0.4141, 1.7037, 0.1544, -0.1862, 0.0736, 0.4393,
  22873. 0.0851],
  22874. [ 0.5778, -0.4393, 1.8018, -0.4614, -0.6693, -0.1381, 0.5490,
  22875. 0.2083],
  22876. [ 0.5791, -0.4127, 1.8442, -0.3985, -0.6031, -0.6154, 0.4473,
  22877. 0.2464],
  22878. [ 0.5109, -0.4316, 1.1436, -1.3467, -0.3864, -1.4160, 0.1239,
  22879. -0.0580],
  22880. [ 0.5924, -0.4507, 1.7095, -0.4614, -0.6115, -0.8156, 0.4104,
  22881. 0.1005],
  22882. [ 0.5606, -0.3856, 1.2476, -1.3544, -0.5480, -0.9618, 0.3152,
  22883. 0.0562],
  22884. [ 0.5417, -0.4355, 1.8018, -0.3383, -0.3979, 0.2622, 0.5143,
  22885. 0.2622],
  22886. [ 0.5548, -0.3936, 1.1634, -0.8105, -0.5192, -1.0696, 0.2372,
  22887. 0.3931]]], device='cuda:0')
  22888. loss_train_step before backward: tensor(0.0113, device='cuda:0', grad_fn=<MseLossBackward>)
  22889. loss_train_step after backward: tensor(0.0113, device='cuda:0', grad_fn=<MseLossBackward>)
  22890. loss_train: 2.5444738110527396
  22891. step: 86
  22892. running loss: 0.02958690477968302
  22893. Train Steps: 86/90 Loss: 0.0296 torch.Size([8, 600, 800])
  22894. torch.Size([8, 8])
  22895. tensor([[0.6053, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168],
  22896. [0.6137, 0.4038, 0.8563, 0.4050, 0.3813, 0.2550, 0.5106, 0.4954],
  22897. [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400],
  22898. [0.6270, 0.4267, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183],
  22899. [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092],
  22900. [0.6234, 0.4179, 0.7825, 0.3450, 0.3813, 0.2867, 0.5675, 0.5617],
  22901. [0.6122, 0.3993, 0.8738, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927],
  22902. [0.6179, 0.4118, 0.7278, 0.4237, 0.3588, 0.3400, 0.5675, 0.5917]],
  22903. device='cuda:0', dtype=torch.float64)
  22904. predictions are: tensor([[ 0.3100, -0.6210, 0.9047, -1.3336, -0.3877, -1.2771, 0.5174, 0.1538],
  22905. [ 0.5911, -0.4111, 1.6927, -0.4872, -0.4639, -0.9790, 0.3204, 0.0526],
  22906. [ 0.6158, -0.4260, 1.3063, -1.2585, -0.1382, -1.3125, 0.6621, 0.1757],
  22907. [ 0.7599, -0.2690, 1.0580, -1.0242, -0.0868, -1.2896, 0.4650, 0.4659],
  22908. [ 0.4308, -0.5882, 1.8699, 0.0753, -0.5445, 0.1465, 0.6465, -0.0091],
  22909. [ 0.4604, -0.4707, 1.4312, -0.7649, -0.4340, -0.8856, 0.4767, 0.3702],
  22910. [ 0.1774, -0.7147, 1.8432, -0.1140, -0.2642, 0.0558, 0.2860, -0.0067],
  22911. [ 0.5459, -0.4278, 1.3398, -0.5322, -0.5929, -0.7123, 0.4224, 0.4162]],
  22912. device='cuda:0', grad_fn=<AddmmBackward>)
  22913. landmarks are: tensor([[[ 0.5102, -0.4223, 0.8999, -1.4011, -0.4383, -1.3082, 0.2267,
  22914. 0.1013],
  22915. [ 0.5485, -0.4209, 1.6691, -0.4152, -0.5249, -1.1081, 0.0725,
  22916. 0.0021],
  22917. [ 0.5842, -0.3848, 1.2938, -1.2159, -0.2132, -1.4160, 0.5778,
  22918. 0.2083],
  22919. [ 0.6101, -0.3152, 1.0166, -0.7540, -0.0226, -1.4468, 0.2823,
  22920. 0.5702],
  22921. [ 0.5783, -0.4363, 1.7724, 0.0432, -0.5153, 0.0871, 0.4840,
  22922. 0.0663],
  22923. [ 0.5935, -0.3558, 1.3284, -0.6924, -0.5249, -0.9618, 0.3353,
  22924. 0.3084],
  22925. [ 0.5417, -0.4417, 1.7499, -0.1304, -0.1994, -0.0324, 0.0951,
  22926. -0.0099],
  22927. [ 0.5680, -0.3840, 1.0756, -0.3290, -0.6289, -0.7155, 0.3353,
  22928. 0.4470]]], device='cuda:0')
  22929. loss_train_step before backward: tensor(0.0161, device='cuda:0', grad_fn=<MseLossBackward>)
  22930. loss_train_step after backward: tensor(0.0161, device='cuda:0', grad_fn=<MseLossBackward>)
  22931. loss_train: 2.56061180960387
  22932. step: 87
  22933. running loss: 0.029432319650619196
  22934. Train Steps: 87/90 Loss: 0.0294 torch.Size([8, 600, 800])
  22935. torch.Size([8, 8])
  22936. tensor([[0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222],
  22937. [0.6200, 0.4071, 0.7338, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517],
  22938. [0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633],
  22939. [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167],
  22940. [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650],
  22941. [0.6128, 0.4115, 0.7934, 0.3778, 0.3450, 0.4033, 0.5337, 0.5456],
  22942. [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483],
  22943. [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167]],
  22944. device='cuda:0', dtype=torch.float64)
  22945. predictions are: tensor([[ 4.6039e-01, -4.2506e-01, 1.4384e+00, -4.5731e-01, -4.3452e-02,
  22946. -1.2264e+00, 3.2788e-01, 3.5845e-01],
  22947. [ 6.0257e-01, -4.3626e-01, 1.1690e+00, -1.2788e+00, -3.3650e-01,
  22948. -1.1651e+00, 6.8741e-01, 1.0520e-01],
  22949. [ 5.2720e-01, -4.8766e-01, 1.6576e+00, -1.2388e+00, -1.2994e-01,
  22950. -1.1635e+00, 1.0622e+00, 1.5645e-01],
  22951. [ 4.8725e-01, -4.7966e-01, 1.4731e+00, -7.5758e-01, -6.2636e-01,
  22952. -6.2725e-01, 5.6555e-01, 3.4666e-01],
  22953. [ 4.4267e-01, -4.8748e-01, 1.3048e+00, -9.8736e-01, -4.8932e-01,
  22954. -9.7766e-01, 2.2468e-01, 9.5480e-02],
  22955. [ 3.4075e-01, -5.4505e-01, 1.4408e+00, -4.8045e-01, -6.5704e-01,
  22956. -4.3574e-01, 2.1275e-01, 1.5534e-01],
  22957. [ 3.3860e-01, -5.9572e-01, 1.7547e+00, -3.4798e-01, -5.4307e-01,
  22958. -8.2032e-02, 4.6923e-01, 4.7698e-02],
  22959. [ 5.7668e-01, -3.5282e-01, 1.1982e+00, -6.6692e-01, -1.3449e-03,
  22960. -1.3082e+00, 3.4176e-01, 3.4211e-01]], device='cuda:0',
  22961. grad_fn=<AddmmBackward>)
  22962. landmarks are: tensor([[[ 0.6237, -0.2983, 1.3919, -0.4691, -0.0457, -1.2313, 0.2453,
  22963. 0.5882],
  22964. [ 0.5779, -0.4054, 1.1032, -1.4006, -0.3460, -1.1543, 0.5547,
  22965. 0.2622],
  22966. [ 0.6042, -0.4105, 1.5478, -1.2082, -0.1208, -1.0927, 0.9704,
  22967. 0.3157],
  22968. [ 0.5768, -0.3899, 1.3861, -0.7771, -0.5885, -0.5461, 0.5028,
  22969. 0.5624],
  22970. [ 0.5320, -0.4189, 1.3053, -1.0773, -0.5711, -0.9849, 0.2267,
  22971. 0.3237],
  22972. [ 0.5444, -0.3852, 1.3786, -0.5409, -0.6924, -0.4229, 0.1791,
  22973. 0.2341],
  22974. [ 0.5546, -0.4620, 1.7788, -0.4229, -0.5192, -0.0226, 0.4277,
  22975. 0.2468],
  22976. [ 0.6161, -0.3075, 1.1678, -0.6500, 0.0813, -1.4006, 0.2545,
  22977. 0.5624]]], device='cuda:0')
  22978. loss_train_step before backward: tensor(0.0105, device='cuda:0', grad_fn=<MseLossBackward>)
  22979. loss_train_step after backward: tensor(0.0105, device='cuda:0', grad_fn=<MseLossBackward>)
  22980. loss_train: 2.5710743190720677
  22981. step: 88
  22982. running loss: 0.02921675362581895
  22983.  
  22984. Train Steps: 88/90 Loss: 0.0292 torch.Size([8, 600, 800])
  22985. torch.Size([8, 8])
  22986. tensor([[0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389],
  22987. [0.6111, 0.3995, 0.8788, 0.4567, 0.3813, 0.4833, 0.5450, 0.5700],
  22988. [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524],
  22989. [0.6200, 0.3999, 0.8653, 0.5207, 0.4100, 0.5125, 0.5975, 0.5103],
  22990. [0.6229, 0.4066, 0.8513, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350],
  22991. [0.6110, 0.3984, 0.8750, 0.4933, 0.4625, 0.4950, 0.5578, 0.5676],
  22992. [0.6218, 0.4185, 0.7338, 0.2650, 0.4625, 0.1950, 0.5687, 0.5800],
  22993. [0.6268, 0.4052, 0.8175, 0.2250, 0.4688, 0.1917, 0.6375, 0.5267]],
  22994. device='cuda:0', dtype=torch.float64)
  22995. predictions are: tensor([[ 0.7000, -0.3330, 1.2081, -1.3917, -0.5066, -1.2099, 0.5612, 0.2131],
  22996. [ 0.5759, -0.4134, 1.6561, -0.3017, -0.5498, -0.2371, 0.2067, 0.2265],
  22997. [ 0.5207, -0.4113, 1.6869, -0.0436, -0.5771, -0.3776, 0.2774, 0.1871],
  22998. [ 0.5346, -0.4634, 1.6791, 0.0160, -0.4440, -0.0452, 0.5119, 0.0538],
  22999. [ 0.2992, -0.6279, 1.5847, 0.2346, -0.2637, -0.1236, 0.4955, 0.2122],
  23000. [ 0.4735, -0.4733, 1.6959, -0.1844, -0.1443, -0.1494, 0.2397, 0.2195],
  23001. [ 0.4997, -0.4006, 1.0715, -1.2295, -0.1961, -1.5172, 0.3869, 0.3413],
  23002. [ 0.9420, -0.1883, 1.4086, -1.4380, -0.1689, -1.5866, 0.7342, 0.0748]],
  23003. device='cuda:0', grad_fn=<AddmmBackward>)
  23004. landmarks are: tensor([[[ 5.6966e-01, -4.4656e-01, 1.1973e+00, -1.1871e+00, -4.5712e-01,
  23005. -9.9653e-01, 5.2186e-01, 2.0324e-01],
  23006. [ 5.3684e-01, -4.4057e-01, 1.7730e+00, -1.7660e-01, -5.2494e-01,
  23007. -5.3426e-02, 2.3141e-01, 3.4688e-01],
  23008. [ 5.5289e-01, -3.8106e-01, 1.7788e+00, -3.8029e-02, -5.3072e-01,
  23009. -2.0739e-01, 7.2734e-02, 2.6568e-01],
  23010. [ 5.7777e-01, -4.3888e-01, 1.7107e+00, 1.1921e-01, -3.9207e-01,
  23011. 8.1507e-02, 4.7413e-01, 7.1077e-02],
  23012. [ 5.9107e-01, -4.0805e-01, 1.6460e+00, 3.5458e-01, -2.0739e-01,
  23013. 4.6651e-02, 4.9700e-01, 1.8522e-01],
  23014. [ 5.3637e-01, -4.4573e-01, 1.7557e+00, -7.2363e-03, -1.4965e-01,
  23015. 4.6189e-04, 2.9049e-01, 3.3573e-01],
  23016. [ 5.8620e-01, -3.5296e-01, 1.1032e+00, -1.0619e+00, -1.4965e-01,
  23017. -1.3852e+00, 3.4111e-01, 3.9307e-01],
  23018. [ 6.0918e-01, -4.1432e-01, 1.4901e+00, -1.2467e+00, -1.2079e-01,
  23019. -1.4006e+00, 6.5866e-01, 1.4673e-01]]], device='cuda:0')
  23020. loss_train_step before backward: tensor(0.0142, device='cuda:0', grad_fn=<MseLossBackward>)
  23021. loss_train_step after backward: tensor(0.0142, device='cuda:0', grad_fn=<MseLossBackward>)
  23022. loss_train: 2.58522439096123
  23023. step: 89
  23024. running loss: 0.02904746506698011
  23025. Train Steps: 89/90 Loss: 0.0290 torch.Size([8, 600, 800])
  23026. torch.Size([8, 8])
  23027. tensor([[0.6079, 0.3964, 0.7420, 0.2958, 0.3563, 0.2917, 0.5351, 0.4980],
  23028. [0.6224, 0.4179, 0.8700, 0.5683, 0.4037, 0.4683, 0.5650, 0.5600],
  23029. [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285],
  23030. [0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494],
  23031. [0.6248, 0.4185, 0.8500, 0.5767, 0.4463, 0.4550, 0.5613, 0.5917],
  23032. [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167],
  23033. [0.6364, 0.4165, 0.9088, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297],
  23034. [0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385]],
  23035. device='cuda:0', dtype=torch.float64)
  23036. predictions are: tensor([[ 0.5832, -0.3993, 1.1275, -1.2637, -0.5853, -1.0007, -0.0508, -0.0505],
  23037. [ 0.3885, -0.4997, 1.5609, 0.1179, -0.3980, -0.2911, 0.2709, 0.3850],
  23038. [ 0.3156, -0.6031, 1.7021, -0.8399, -0.6414, -0.2453, 0.6654, 0.1443],
  23039. [ 0.7260, -0.3384, 1.6161, 0.0791, -0.5394, -0.3537, 0.7659, 0.1245],
  23040. [ 0.4992, -0.4325, 1.5218, 0.1549, -0.1555, -0.3437, 0.2934, 0.4448],
  23041. [ 0.6640, -0.2715, 1.1294, -0.9919, -0.0219, -1.3958, 0.1330, 0.4381],
  23042. [ 0.6496, -0.3338, 1.8106, -0.5189, -0.3958, -0.9181, 0.5184, 0.1384],
  23043. [ 0.6956, -0.3518, 1.4355, -1.4073, -0.2439, -1.3482, 0.7373, 0.1280]],
  23044. device='cuda:0', grad_fn=<AddmmBackward>)
  23045. landmarks are: tensor([[[ 0.5219, -0.4550, 1.1415, -0.9196, -0.6404, -0.9387, 0.1856,
  23046. 0.0141],
  23047. [ 0.5891, -0.3557, 1.7326, 0.3392, -0.4210, -0.1227, 0.3238,
  23048. 0.3007],
  23049. [ 0.6108, -0.4273, 1.8711, -0.6616, -0.5769, -0.1997, 0.9156,
  23050. 0.1554],
  23051. [ 0.6224, -0.4032, 1.5940, 0.2930, -0.5769, -0.2690, 0.8900,
  23052. 0.2516],
  23053. [ 0.5999, -0.3530, 1.6402, 0.3777, -0.2247, -0.1843, 0.3065,
  23054. 0.4470],
  23055. [ 0.6161, -0.3075, 1.1678, -0.6500, 0.0813, -1.4006, 0.2545,
  23056. 0.5624],
  23057. [ 0.6537, -0.3623, 1.9115, -0.2690, -0.4037, -0.8309, 0.6926,
  23058. 0.1608],
  23059. [ 0.6174, -0.3936, 1.4586, -1.1709, -0.2420, -1.1389, 0.8296,
  23060. 0.2012]]], device='cuda:0')
  23061. loss_train_step before backward: tensor(0.0192, device='cuda:0', grad_fn=<MseLossBackward>)
  23062. loss_train_step after backward: tensor(0.0192, device='cuda:0', grad_fn=<MseLossBackward>)
  23063. loss_train: 2.604389033280313
  23064. step: 90
  23065. running loss: 0.02893765592533681
  23066. Valid Steps: 10/10 Loss: nan 7.8799
  23067. --------------------------------------------------
  23068. Epoch: 6 Train Loss: 0.0289 Valid Loss: nan
  23069. --------------------------------------------------
  23070. size of train loader is: 90
  23071. torch.Size([8, 600, 800])
  23072. torch.Size([8, 8])
  23073. tensor([[0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717],
  23074. [0.6218, 0.4098, 0.7238, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350],
  23075. [0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533],
  23076. [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400],
  23077. [ nan, nan, 0.7612, 0.3250, 0.4037, 0.2533, 0.5438, 0.5767],
  23078. [0.6206, 0.4001, 0.8900, 0.3933, 0.3588, 0.3567, 0.5837, 0.5083],
  23079. [0.6165, 0.4106, 0.7575, 0.1733, 0.3838, 0.2650, 0.5680, 0.5116],
  23080. [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467]],
  23081. device='cuda:0', dtype=torch.float64)
  23082. predictions are: tensor([[ 0.6477, -0.3526, 0.9691, -0.9025, -0.4389, -1.1085, 0.2247, 0.2755],
  23083. [ 0.8920, -0.2164, 1.2343, -1.0870, -0.2121, -1.2321, 0.6169, 0.2335],
  23084. [ 0.5558, -0.4183, 1.1495, -1.0581, -0.4915, -0.8478, 0.4433, 0.2620],
  23085. [ 0.5084, -0.4509, 1.4207, -1.0557, -0.0454, -1.4207, 0.5016, 0.1969],
  23086. [-0.0192, -0.7355, 1.4270, -0.7559, -0.2804, -1.0867, 0.2119, 0.3873],
  23087. [ 0.6241, -0.4014, 1.9214, -0.2800, -0.5671, -0.6757, 0.3070, 0.0805],
  23088. [ 0.5517, -0.4502, 1.3219, -1.2073, -0.4492, -1.1858, 0.5526, 0.0171],
  23089. [ 0.4252, -0.5143, 1.6961, -0.5656, -0.6468, -0.3361, 0.3515, 0.2301]],
  23090. device='cuda:0', grad_fn=<AddmmBackward>)
  23091. landmarks are: tensor([[[ 0.5633, -0.3883, 0.8648, -1.0288, -0.5192, -1.1081, 0.2776,
  23092. 0.3546],
  23093. [ 0.5861, -0.3929, 1.0570, -1.4314, -0.3286, -1.1235, 0.6182,
  23094. 0.1852],
  23095. [ 0.5878, -0.4052, 1.0229, -1.2855, -0.5596, -0.8232, 0.5316,
  23096. 0.2699],
  23097. [ 0.5842, -0.3848, 1.2938, -1.2159, -0.2132, -1.4160, 0.5778,
  23098. 0.2083],
  23099. [-2.2859, -2.2859, 1.2303, -0.7848, -0.4210, -1.1158, 0.2256,
  23100. 0.3777],
  23101. [ 0.5807, -0.4378, 1.8249, -0.4691, -0.6289, -0.6385, 0.4104,
  23102. 0.0620],
  23103. [ 0.5618, -0.3895, 1.2129, -1.4853, -0.5134, -1.0619, 0.3378,
  23104. 0.0772],
  23105. [ 0.5484, -0.4591, 1.5940, -0.7771, -0.6693, -0.3460, 0.4219,
  23106. 0.2391]]], device='cuda:0')
  23107. loss_train_step before backward: tensor(0.1317, device='cuda:0', grad_fn=<MseLossBackward>)
  23108. loss_train_step after backward: tensor(0.1317, device='cuda:0', grad_fn=<MseLossBackward>)
  23109. loss_train: 0.13170002400875092
  23110. step: 1
  23111. running loss: 0.13170002400875092
  23112. Train Steps: 1/90 Loss: 0.1317 torch.Size([8, 600, 800])
  23113. torch.Size([8, 8])
  23114. tensor([[0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266],
  23115. [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683],
  23116. [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5687, 0.5567],
  23117. [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750],
  23118. [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869],
  23119. [0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118],
  23120. [0.6361, 0.4076, 0.8862, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297],
  23121. [0.6241, 0.4143, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550]],
  23122. device='cuda:0', dtype=torch.float64)
  23123. predictions are: tensor([[ 0.6647, -0.3062, 1.0944, -1.2538, -0.3591, -1.2520, 0.4543, 0.4699],
  23124. [ 0.3959, -0.5123, 1.6735, -0.1118, -0.0870, -0.0970, 0.0966, 0.1922],
  23125. [ 0.3720, -0.4999, 1.1677, -1.1540, -0.4044, -1.2134, 0.3068, 0.2836],
  23126. [ 0.6326, -0.3810, 1.6411, -0.0633, -0.3244, -0.1243, 0.2366, 0.2300],
  23127. [ 0.6179, -0.3312, 1.6827, -0.9184, -0.2375, -1.5012, 0.2988, -0.0049],
  23128. [ 0.7444, -0.2400, 1.0723, -0.8516, -0.3273, -1.3134, 0.1802, 0.4339],
  23129. [ 0.7338, -0.3491, 1.6987, 0.1139, -0.6360, -0.2006, 0.7904, 0.1141],
  23130. [ 0.5614, -0.3962, 1.7932, -0.2720, -0.4493, 0.1143, 0.6197, 0.2201]],
  23131. device='cuda:0', grad_fn=<AddmmBackward>)
  23132. landmarks are: tensor([[[ 0.5746, -0.4021, 1.0801, -1.1312, -0.3229, -1.1081, 0.4803,
  23133. 0.6084],
  23134. [ 0.5359, -0.4370, 1.7095, -0.0303, -0.0804, -0.0380, 0.1044,
  23135. 0.3392],
  23136. [ 0.5824, -0.3625, 1.0405, -0.9325, -0.4268, -1.1389, 0.3411,
  23137. 0.2853],
  23138. [ 0.5525, -0.4463, 1.7326, 0.0313, -0.2536, -0.0688, 0.1968,
  23139. 0.3700],
  23140. [ 0.6119, -0.3927, 1.6979, -0.7925, -0.2536, -1.3698, 0.4282,
  23141. -0.0368],
  23142. [ 0.6077, -0.3226, 0.9993, -0.6462, -0.2651, -1.3082, 0.2946,
  23143. 0.5401],
  23144. [ 0.6520, -0.4032, 1.8076, 0.1852, -0.5711, -0.1381, 0.7876,
  23145. 0.1608],
  23146. [ 0.5968, -0.3725, 1.8423, -0.1381, -0.4037, 0.1852, 0.6009,
  23147. 0.2776]]], device='cuda:0')
  23148. loss_train_step before backward: tensor(0.0086, device='cuda:0', grad_fn=<MseLossBackward>)
  23149. loss_train_step after backward: tensor(0.0086, device='cuda:0', grad_fn=<MseLossBackward>)
  23150. loss_train: 0.1403241939842701
  23151. step: 2
  23152. running loss: 0.07016209699213505
  23153.  
  23154. Train Steps: 2/90 Loss: 0.0702 torch.Size([8, 600, 800])
  23155. torch.Size([8, 8])
  23156. tensor([[0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446],
  23157. [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235],
  23158. [ nan, nan, 0.7335, 0.2569, 0.3788, 0.2667, 0.5066, 0.5578],
  23159. [0.6346, 0.4144, 0.9088, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899],
  23160. [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366],
  23161. [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578],
  23162. [0.6038, 0.3946, 0.8413, 0.4883, 0.3563, 0.4550, 0.5266, 0.4693],
  23163. [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377]],
  23164. device='cuda:0', dtype=torch.float64)
  23165. predictions are: tensor([[ 0.7319, -0.2171, 1.5618, -0.3020, -0.4572, -0.2303, 0.1329, 0.3098],
  23166. [ 0.8232, -0.2165, 1.7220, -0.2264, -0.2888, 0.0441, 0.4587, 0.1933],
  23167. [-1.7871, -1.9312, 1.1732, -1.2439, -0.4154, -1.3333, 0.0692, 0.2836],
  23168. [ 0.7511, -0.2776, 1.8293, -0.2960, -0.3076, -0.5175, 0.9069, 0.3537],
  23169. [ 0.9328, -0.1419, 1.4058, 0.0273, -0.4443, -0.1959, 0.7334, 0.2985],
  23170. [ 0.9208, -0.1094, 0.9170, -1.5732, -0.3551, -1.5826, 0.2315, 0.1841],
  23171. [ 0.7178, -0.2332, 1.5154, -0.1506, -0.5277, -0.4161, -0.0780, 0.2460],
  23172. [ 0.8859, -0.2005, 1.6186, 0.0542, -0.3882, -0.3113, 0.3717, 0.1910]],
  23173. device='cuda:0', grad_fn=<AddmmBackward>)
  23174. landmarks are: tensor([[[ 0.5444, -0.3846, 1.6171, -0.1689, -0.5885, -0.0380, 0.1791,
  23175. 0.2296],
  23176. [ 0.5762, -0.4706, 1.7754, -0.0984, -0.3680, 0.2380, 0.6277,
  23177. 0.1322],
  23178. [-2.2859, -2.2859, 1.1020, -1.0994, -0.5365, -1.0542, 0.0542,
  23179. 0.2905],
  23180. [ 0.6454, -0.3719, 1.9115, -0.1304, -0.5076, -0.2844, 1.0033,
  23181. 0.4386],
  23182. [ 0.6289, -0.4345, 1.3794, 0.3679, -0.4845, 0.0390, 0.9265,
  23183. 0.1928],
  23184. [ 0.5232, -0.4189, 0.9358, -1.4006, -0.3921, -1.3698, 0.2555,
  23185. 0.2906],
  23186. [ 0.5030, -0.4631, 1.5998, -0.0303, -0.6404, -0.1843, 0.1465,
  23187. -0.1181],
  23188. [ 0.5775, -0.4607, 1.6741, 0.1962, -0.4036, -0.1212, 0.4588,
  23189. 0.1979]]], device='cuda:0')
  23190. loss_train_step before backward: tensor(0.0370, device='cuda:0', grad_fn=<MseLossBackward>)
  23191. loss_train_step after backward: tensor(0.0370, device='cuda:0', grad_fn=<MseLossBackward>)
  23192. loss_train: 0.1773393712937832
  23193. step: 3
  23194. running loss: 0.0591131237645944
  23195. Train Steps: 3/90 Loss: 0.0591 torch.Size([8, 600, 800])
  23196. torch.Size([8, 8])
  23197. tensor([[0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366],
  23198. [ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600],
  23199. [0.6179, 0.4118, 0.7278, 0.4237, 0.3588, 0.3400, 0.5675, 0.5917],
  23200. [0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222],
  23201. [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356],
  23202. [0.6102, 0.3999, 0.8750, 0.5133, 0.3825, 0.4750, 0.5637, 0.5083],
  23203. [0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400],
  23204. [0.6250, 0.4106, 0.8700, 0.3717, 0.3588, 0.4967, 0.6038, 0.5167]],
  23205. device='cuda:0', dtype=torch.float64)
  23206. predictions are: tensor([[ 8.9299e-01, -1.9483e-01, 1.3833e+00, 1.2987e-01, -4.8383e-01,
  23207. -5.3583e-03, 7.4699e-01, 2.4503e-01],
  23208. [-1.8940e+00, -2.0092e+00, 1.3015e+00, -1.1059e+00, -3.3122e-01,
  23209. -1.2012e+00, 2.4633e-01, 2.7320e-01],
  23210. [ 6.9292e-01, -2.6543e-01, 1.1853e+00, -5.9782e-01, -6.0230e-01,
  23211. -8.2772e-01, 1.5278e-01, 4.5867e-01],
  23212. [ 6.5121e-01, -2.5269e-01, 1.3511e+00, -6.1461e-01, -7.2152e-02,
  23213. -1.2816e+00, 1.5690e-01, 5.1567e-01],
  23214. [ 7.7100e-01, -2.4929e-01, 1.3652e+00, -1.1704e+00, -3.3382e-01,
  23215. -1.2479e+00, 6.5034e-01, 1.7962e-01],
  23216. [ 8.6061e-01, -1.8263e-01, 1.6473e+00, -7.0001e-04, -5.0347e-01,
  23217. -9.7541e-02, 1.6001e-01, 6.7358e-02],
  23218. [ 5.2133e-01, -3.6087e-01, 1.4178e+00, -9.9686e-01, -1.1791e-01,
  23219. -1.4057e+00, 3.2734e-01, 2.0587e-01],
  23220. [ 8.5589e-01, -1.9249e-01, 1.6359e+00, -5.9213e-01, -6.1109e-01,
  23221. 4.0864e-02, 5.4792e-01, 2.4353e-01]], device='cuda:0',
  23222. grad_fn=<AddmmBackward>)
  23223. landmarks are: tensor([[[ 0.6289, -0.4345, 1.3794, 0.3679, -0.4845, 0.0390, 0.9265,
  23224. 0.1928],
  23225. [-2.2859, -2.2859, 1.4266, -1.1389, -0.4499, -1.1235, 0.2891,
  23226. 0.3007],
  23227. [ 0.5680, -0.3840, 1.0756, -0.3290, -0.6289, -0.7155, 0.3353,
  23228. 0.4470],
  23229. [ 0.6237, -0.2983, 1.3919, -0.4691, -0.0457, -1.2313, 0.2453,
  23230. 0.5882],
  23231. [ 0.6174, -0.4225, 1.4975, -1.1709, -0.3174, -1.1806, 0.6539,
  23232. 0.1879],
  23233. [ 0.5326, -0.4390, 1.7557, 0.0851, -0.5192, -0.0919, 0.3180,
  23234. 0.0620],
  23235. [ 0.5911, -0.3888, 1.4727, -0.9541, -0.0919, -1.4930, 0.3988,
  23236. 0.2083],
  23237. [ 0.6010, -0.3896, 1.7326, -0.5692, -0.6289, 0.0082, 0.5028,
  23238. 0.1005]]], device='cuda:0')
  23239. loss_train_step before backward: tensor(0.0179, device='cuda:0', grad_fn=<MseLossBackward>)
  23240. loss_train_step after backward: tensor(0.0179, device='cuda:0', grad_fn=<MseLossBackward>)
  23241. loss_train: 0.19527801871299744
  23242. step: 4
  23243. running loss: 0.04881950467824936
  23244. Train Steps: 4/90 Loss: 0.0488 torch.Size([8, 600, 800])
  23245. torch.Size([8, 8])
  23246. tensor([[0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933],
  23247. [0.6185, 0.4098, 0.8838, 0.4900, 0.4537, 0.5800, 0.6288, 0.5400],
  23248. [0.6231, 0.3973, 0.8650, 0.3950, 0.3625, 0.3183, 0.5837, 0.5167],
  23249. [0.6124, 0.4075, 0.7696, 0.4153, 0.3475, 0.3767, 0.5157, 0.5427],
  23250. [ nan, nan, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999],
  23251. [0.6216, 0.4100, 0.7350, 0.2067, 0.4325, 0.2050, 0.5950, 0.5333],
  23252. [0.6034, 0.4011, 0.7350, 0.2533, 0.3438, 0.3367, 0.5516, 0.5084],
  23253. [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5988, 0.5667]],
  23254. device='cuda:0', dtype=torch.float64)
  23255. predictions are: tensor([[ 0.6983, -0.2865, 1.5583, 0.4364, 0.0205, -0.2002, 0.3845, 0.5493],
  23256. [ 0.6020, -0.3588, 1.7794, -0.0123, -0.0761, 0.4327, 0.5758, 0.3702],
  23257. [ 0.8159, -0.2481, 1.7206, -0.3957, -0.6126, -0.8451, 0.4303, 0.1379],
  23258. [ 0.7543, -0.2461, 1.3220, -0.4106, -0.6687, -0.5468, 0.0829, 0.3590],
  23259. [-0.8455, -1.2990, 1.2776, -1.2102, -0.3380, -1.4762, 0.2634, 0.1421],
  23260. [ 0.4908, -0.4188, 1.0813, -1.3411, -0.3358, -1.4621, 0.4294, 0.2803],
  23261. [ 0.8225, -0.2386, 1.1020, -1.1353, -0.6921, -0.7803, 0.3172, 0.0501],
  23262. [ 0.5905, -0.3740, 1.6765, -0.6473, -0.5600, -0.7175, 0.5019, 0.3938]],
  23263. device='cuda:0', grad_fn=<AddmmBackward>)
  23264. landmarks are: tensor([[[ 0.5959, -0.3579, 1.6055, 0.3623, -0.0573, -0.2074, 0.3122,
  23265. 0.4547],
  23266. [ 0.5709, -0.3933, 1.7961, -0.0226, -0.1901, 0.3931, 0.6182,
  23267. 0.2083],
  23268. [ 0.5924, -0.4507, 1.7095, -0.4614, -0.6115, -0.8156, 0.4104,
  23269. 0.1005],
  23270. [ 0.5427, -0.4035, 1.2688, -0.3675, -0.6808, -0.5461, 0.0959,
  23271. 0.2206],
  23272. [-2.2859, -2.2859, 1.3400, -1.0388, -0.3055, -1.4930, 0.1157,
  23273. 0.0231],
  23274. [ 0.5853, -0.3920, 1.1090, -1.3313, -0.2882, -1.3390, 0.4624,
  23275. 0.1775],
  23276. [ 0.5009, -0.4333, 1.1090, -1.1158, -0.6982, -0.7309, 0.2617,
  23277. 0.0622],
  23278. [ 0.5783, -0.3933, 1.6748, -0.6154, -0.5769, -0.6462, 0.4797,
  23279. 0.3315]]], device='cuda:0')
  23280. loss_train_step before backward: tensor(0.0562, device='cuda:0', grad_fn=<MseLossBackward>)
  23281. loss_train_step after backward: tensor(0.0562, device='cuda:0', grad_fn=<MseLossBackward>)
  23282. loss_train: 0.25149907171726227
  23283. step: 5
  23284. running loss: 0.05029981434345245
  23285. Train Steps: 5/90 Loss: 0.0503 torch.Size([8, 600, 800])
  23286. torch.Size([8, 8])
  23287. tensor([[0.6179, 0.4040, 0.7412, 0.1850, 0.3825, 0.2783, 0.5837, 0.5600],
  23288. [0.6270, 0.4267, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183],
  23289. [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461],
  23290. [0.6243, 0.4128, 0.7762, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417],
  23291. [0.6241, 0.4143, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550],
  23292. [0.6143, 0.4040, 0.8237, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973],
  23293. [0.6111, 0.3995, 0.8788, 0.4567, 0.3813, 0.4833, 0.5450, 0.5700],
  23294. [0.6127, 0.4118, 0.8650, 0.5083, 0.4088, 0.5367, 0.5300, 0.5456]],
  23295. device='cuda:0', dtype=torch.float64)
  23296. predictions are: tensor([[ 0.7097, -0.2963, 1.1037, -1.2898, -0.6142, -1.1763, 0.4470, 0.2839],
  23297. [ 0.6372, -0.2858, 0.9849, -0.8955, -0.1584, -1.5070, 0.1930, 0.5510],
  23298. [ 0.6837, -0.3201, 1.6716, 0.2550, -0.1896, 0.0621, 0.3190, 0.2978],
  23299. [ 0.2536, -0.5944, 1.3756, -1.0492, -0.5398, -0.9605, 0.6340, 0.2841],
  23300. [ 0.4375, -0.4863, 1.8205, -0.1297, -0.4469, 0.1487, 0.6756, 0.3328],
  23301. [ 0.0804, -0.6800, 1.4471, -0.7608, -0.4763, -1.2891, 0.1751, 0.0604],
  23302. [ 0.7134, -0.3114, 1.6887, -0.1343, -0.5369, -0.1299, 0.2081, 0.2759],
  23303. [ 0.6029, -0.3714, 1.6532, 0.1362, -0.3379, 0.1308, 0.3406, 0.2828]],
  23304. device='cuda:0', grad_fn=<AddmmBackward>)
  23305. landmarks are: tensor([[[ 0.5680, -0.4201, 1.1379, -1.4314, -0.5192, -1.0003, 0.4104,
  23306. 0.3007],
  23307. [ 0.6101, -0.3152, 1.0166, -0.7540, -0.0226, -1.4468, 0.2823,
  23308. 0.5702],
  23309. [ 0.5443, -0.3831, 1.7095, 0.1621, -0.2016, 0.1390, 0.1437,
  23310. 0.2364],
  23311. [ 0.5977, -0.3792, 1.2995, -1.0311, -0.5192, -0.8386, 0.5836,
  23312. 0.2160],
  23313. [ 0.5968, -0.3725, 1.8423, -0.1381, -0.4037, 0.1852, 0.6009,
  23314. 0.2776],
  23315. [ 0.5515, -0.4201, 1.5189, -0.7463, -0.4037, -1.3082, 0.0869,
  23316. 0.0111],
  23317. [ 0.5368, -0.4406, 1.7730, -0.1766, -0.5249, -0.0534, 0.2314,
  23318. 0.3469],
  23319. [ 0.5443, -0.3840, 1.7095, 0.0620, -0.3979, 0.1929, 0.1622,
  23320. 0.2341]]], device='cuda:0')
  23321. loss_train_step before backward: tensor(0.0141, device='cuda:0', grad_fn=<MseLossBackward>)
  23322. loss_train_step after backward: tensor(0.0141, device='cuda:0', grad_fn=<MseLossBackward>)
  23323. loss_train: 0.2655968349426985
  23324. step: 6
  23325. running loss: 0.04426613915711641
  23326.  
  23327. Train Steps: 6/90 Loss: 0.0443 torch.Size([8, 600, 800])
  23328. torch.Size([8, 8])
  23329. tensor([[0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100],
  23330. [0.6124, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485],
  23331. [0.6133, 0.4066, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358],
  23332. [0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283],
  23333. [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285],
  23334. [ nan, nan, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702],
  23335. [0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862],
  23336. [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841]],
  23337. device='cuda:0', dtype=torch.float64)
  23338. predictions are: tensor([[ 0.9006, -0.2117, 1.6638, -0.3467, -0.6132, -0.4659, 0.5123, 0.1801],
  23339. [ 0.5317, -0.3793, 0.8797, -0.9900, -0.5496, -0.9261, 0.0243, 0.3312],
  23340. [ 0.6930, -0.3017, 0.7817, -1.0732, -0.4704, -1.2299, -0.0311, 0.2055],
  23341. [ 0.6957, -0.3161, 1.7998, -0.4552, -0.5499, -0.4094, 0.6199, 0.2619],
  23342. [ 0.6030, -0.3544, 1.2631, -0.3271, -0.5427, -0.2744, -0.1266, 0.2739],
  23343. [-1.6762, -1.8580, 1.8621, -0.7469, 0.1769, -1.1852, 0.8146, 0.5468],
  23344. [ 0.6555, -0.3252, 1.7625, -0.2648, -0.5922, -0.1266, 0.5597, 0.1009],
  23345. [ 0.7671, -0.2935, 1.6871, -0.2076, -0.3535, -0.5524, 0.7757, 0.3876]],
  23346. device='cuda:0', grad_fn=<AddmmBackward>)
  23347. landmarks are: tensor([[[ 0.6220, -0.4436, 1.8711, -0.3691, -0.6173, -0.4999, 0.6702,
  23348. 0.0697],
  23349. [ 0.5425, -0.3998, 0.9263, -0.8683, -0.6000, -1.0157, 0.0990,
  23350. 0.2476],
  23351. [ 0.5470, -0.4081, 0.8492, -1.0773, -0.5307, -1.1620, 0.0912,
  23352. 0.1890],
  23353. [ 0.6073, -0.4058, 1.8885, -0.4999, -0.5942, -0.4768, 0.6413,
  23354. 0.1544],
  23355. [ 0.5253, -0.4296, 1.3084, -0.3098, -0.6115, -0.2767, 0.0928,
  23356. 0.1552],
  23357. [-2.2859, -2.2859, 1.8942, -0.6693, 0.0871, -1.2236, 1.1130,
  23358. 0.3478],
  23359. [ 0.6072, -0.4075, 1.8942, -0.3537, -0.5423, -0.1612, 0.6277,
  23360. -0.0400],
  23361. [ 0.6438, -0.3936, 2.0039, -0.2690, -0.2940, -0.6231, 1.0618,
  23362. 0.4119]]], device='cuda:0')
  23363. loss_train_step before backward: tensor(0.0222, device='cuda:0', grad_fn=<MseLossBackward>)
  23364. loss_train_step after backward: tensor(0.0222, device='cuda:0', grad_fn=<MseLossBackward>)
  23365. loss_train: 0.2878271918743849
  23366. step: 7
  23367. running loss: 0.04111817026776927
  23368. Train Steps: 7/90 Loss: 0.0411 torch.Size([8, 600, 800])
  23369. torch.Size([8, 8])
  23370. tensor([[0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773],
  23371. [0.6200, 0.4118, 0.8287, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799],
  23372. [0.6179, 0.4008, 0.7505, 0.2678, 0.4368, 0.1891, 0.5831, 0.5263],
  23373. [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131],
  23374. [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649],
  23375. [0.6162, 0.4134, 0.6700, 0.2467, 0.3962, 0.2533, 0.5737, 0.5467],
  23376. [0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528],
  23377. [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220]],
  23378. device='cuda:0', dtype=torch.float64)
  23379. predictions are: tensor([[ 0.4279, -0.4700, 1.1087, -0.9343, -0.3377, -1.1035, 0.2972, 0.4786],
  23380. [-0.1724, -0.8663, 1.5414, -0.3400, -0.5710, -0.8034, 0.1809, 0.4591],
  23381. [ 0.6113, -0.3749, 1.2179, -1.1382, -0.3679, -1.2327, 0.2952, 0.1828],
  23382. [ 0.8218, -0.2730, 1.8150, -0.3814, -0.7125, -0.3080, 0.3443, 0.1518],
  23383. [ 0.6928, -0.3667, 1.6759, 0.5193, -0.4929, 0.2824, 0.4658, 0.1233],
  23384. [ 0.4151, -0.4984, 0.9299, -1.0047, -0.5191, -1.0346, 0.3536, 0.3316],
  23385. [ 0.2671, -0.6461, 2.0895, -0.5594, -0.3277, -0.3334, 1.0386, 0.3234],
  23386. [ 0.2336, -0.6060, 1.0417, -1.1955, -0.3274, -1.4027, 0.1155, 0.1706]],
  23387. device='cuda:0', grad_fn=<AddmmBackward>)
  23388. landmarks are: tensor([[[ 0.5841, -0.3574, 1.0859, -0.9541, -0.2824, -1.2851, 0.3460,
  23389. 0.3808],
  23390. [ 0.5779, -0.3840, 1.5420, -0.4306, -0.5423, -0.9772, 0.2041,
  23391. 0.3928],
  23392. [ 0.5680, -0.4345, 1.1806, -1.0490, -0.2683, -1.4127, 0.4074,
  23393. 0.1449],
  23394. [ 0.5783, -0.4216, 1.6847, -0.5078, -0.6732, -0.5377, 0.4752,
  23395. 0.0839],
  23396. [ 0.6105, -0.4360, 1.6171, 0.5162, -0.5076, 0.0159, 0.5171,
  23397. -0.1385],
  23398. [ 0.5603, -0.3764, 0.8088, -1.1466, -0.4557, -1.1158, 0.3642,
  23399. 0.2391],
  23400. [ 0.6388, -0.3623, 1.9173, -0.7386, -0.3517, -0.5846, 1.1495,
  23401. 0.2676],
  23402. [ 0.5566, -0.4160, 0.9935, -1.3313, -0.2824, -1.5161, 0.2144,
  23403. 0.1253]]], device='cuda:0')
  23404. loss_train_step before backward: tensor(0.0310, device='cuda:0', grad_fn=<MseLossBackward>)
  23405. loss_train_step after backward: tensor(0.0310, device='cuda:0', grad_fn=<MseLossBackward>)
  23406. loss_train: 0.3187991585582495
  23407. step: 8
  23408. running loss: 0.039849894819781184
  23409. Train Steps: 8/90 Loss: 0.0398 torch.Size([8, 600, 800])
  23410. torch.Size([8, 8])
  23411. tensor([[0.6236, 0.3967, 0.8675, 0.5400, 0.3862, 0.4517, 0.5825, 0.5200],
  23412. [0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862],
  23413. [0.6310, 0.4017, 0.8563, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006],
  23414. [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760],
  23415. [0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6188, 0.5433],
  23416. [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212],
  23417. [ nan, nan, 0.8463, 0.2550, 0.5850, 0.2133, 0.7129, 0.6072],
  23418. [0.6300, 0.4133, 0.8538, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413]],
  23419. device='cuda:0', dtype=torch.float64)
  23420. predictions are: tensor([[ 0.6302, -0.3617, 1.5214, 0.2184, -0.5714, -0.2713, 0.1169, 0.2587],
  23421. [ 0.6184, -0.3406, 1.6832, -0.3059, -0.7060, -0.2061, 0.3594, 0.0877],
  23422. [ 0.6739, -0.3559, 1.5060, 0.3060, -0.6391, -0.1109, 0.3257, 0.1549],
  23423. [ 0.6845, -0.3340, 1.4835, 0.2149, -0.5401, -0.0175, 0.7316, 0.3698],
  23424. [ 0.7103, -0.3030, 1.6602, -0.1140, -0.5570, 0.0523, 0.2161, 0.2177],
  23425. [ 0.6586, -0.3380, 1.3056, -1.1049, -0.5090, -0.9922, 0.3005, 0.1256],
  23426. [-1.3580, -1.6826, 1.5312, -1.0974, 0.1189, -1.1105, 0.7331, 0.5404],
  23427. [ 0.3991, -0.5250, 1.5950, -1.2209, 0.0730, -1.0968, 0.8225, 0.2854]],
  23428. device='cuda:0', grad_fn=<AddmmBackward>)
  23429. landmarks are: tensor([[[ 0.5944, -0.4538, 1.7210, 0.2083, -0.5018, -0.1997, 0.4046,
  23430. 0.1159],
  23431. [ 0.6072, -0.4075, 1.8942, -0.3537, -0.5423, -0.1612, 0.6277,
  23432. -0.0400],
  23433. [ 0.6286, -0.4303, 1.6691, 0.3931, -0.5827, -0.0919, 0.6359,
  23434. 0.0261],
  23435. [ 0.6421, -0.4008, 1.6055, 0.2160, -0.5076, -0.0534, 1.1020,
  23436. 0.3745],
  23437. [ 0.5773, -0.4105, 1.8192, -0.1304, -0.5307, 0.0467, 0.5721,
  23438. 0.2237],
  23439. [ 0.6125, -0.4345, 1.4308, -1.1384, -0.4213, -1.0031, 0.7190,
  23440. 0.1214],
  23441. [-2.2859, -2.2859, 1.6229, -1.1081, 0.4162, -1.3005, 1.0070,
  23442. 0.5188],
  23443. [ 0.6240, -0.3768, 1.6575, -1.2851, 0.2949, -1.2467, 1.1276,
  23444. 0.2142]]], device='cuda:0')
  23445. loss_train_step before backward: tensor(0.0426, device='cuda:0', grad_fn=<MseLossBackward>)
  23446. loss_train_step after backward: tensor(0.0426, device='cuda:0', grad_fn=<MseLossBackward>)
  23447. loss_train: 0.3614186178892851
  23448. step: 9
  23449. running loss: 0.04015762420992056
  23450. Train Steps: 9/90 Loss: 0.0402 torch.Size([8, 600, 800])
  23451. torch.Size([8, 8])
  23452. tensor([[0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183],
  23453. [0.6197, 0.4090, 0.7825, 0.2500, 0.4200, 0.2483, 0.5988, 0.5667],
  23454. [0.6200, 0.4070, 0.8938, 0.4183, 0.3538, 0.4567, 0.6175, 0.5400],
  23455. [0.6147, 0.4081, 0.8538, 0.3400, 0.3663, 0.3150, 0.5142, 0.4875],
  23456. [0.6219, 0.3934, 0.8688, 0.5267, 0.4313, 0.4967, 0.5988, 0.4983],
  23457. [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116],
  23458. [0.6197, 0.4091, 0.8800, 0.4783, 0.3538, 0.4767, 0.5950, 0.5550],
  23459. [0.6075, 0.4000, 0.8513, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280]],
  23460. device='cuda:0', dtype=torch.float64)
  23461. predictions are: tensor([[ 0.6971, -0.3688, 1.7350, -0.1732, -0.4029, 0.2110, 0.7087, 0.2211],
  23462. [ 0.4011, -0.5162, 1.1709, -1.1587, -0.3880, -1.3280, 0.4589, 0.3671],
  23463. [ 0.4933, -0.4978, 1.7170, -0.3445, -0.6054, -0.3812, 0.6053, 0.2377],
  23464. [ 0.3137, -0.5505, 1.5822, -0.6377, -0.6560, -1.0371, 0.2575, 0.1115],
  23465. [ 0.5390, -0.4642, 1.6548, 0.1634, -0.3035, -0.1852, 0.5739, 0.1073],
  23466. [ 0.2994, -0.5938, 1.6538, -0.3325, -0.4045, 0.1992, 0.3695, 0.1887],
  23467. [ 0.5082, -0.4512, 1.6077, -0.1040, -0.5408, -0.2585, 0.5207, 0.4175],
  23468. [ 0.4719, -0.4648, 1.5714, 0.0954, -0.1144, -0.0728, 0.1835, 0.2303]],
  23469. device='cuda:0', grad_fn=<AddmmBackward>)
  23470. landmarks are: tensor([[[ 0.5934, -0.4276, 1.8192, -0.1458, -0.4499, 0.3777, 0.6702,
  23471. 0.1082],
  23472. [ 0.5764, -0.3969, 1.3284, -1.1312, -0.3460, -1.1389, 0.4797,
  23473. 0.3315],
  23474. [ 0.5778, -0.4061, 1.8423, -0.3537, -0.6520, -0.1766, 0.5663,
  23475. 0.2083],
  23476. [ 0.5532, -0.4008, 1.6575, -0.7155, -0.5942, -0.8309, 0.0890,
  23477. -0.0340],
  23478. [ 0.5864, -0.4690, 1.7268, 0.1467, -0.2940, 0.0082, 0.4797,
  23479. 0.0159],
  23480. [ 0.5376, -0.3903, 1.7095, -0.3229, -0.4730, 0.4701, 0.3871,
  23481. 0.0772],
  23482. [ 0.5763, -0.3963, 1.7788, -0.0765, -0.6520, -0.0842, 0.4624,
  23483. 0.2776],
  23484. [ 0.5202, -0.4382, 1.6460, 0.1082, -0.2029, 0.1754, 0.1067,
  23485. 0.1530]]], device='cuda:0')
  23486. loss_train_step before backward: tensor(0.0139, device='cuda:0', grad_fn=<MseLossBackward>)
  23487. loss_train_step after backward: tensor(0.0139, device='cuda:0', grad_fn=<MseLossBackward>)
  23488. loss_train: 0.3753113318234682
  23489. step: 10
  23490. running loss: 0.03753113318234682
  23491.  
  23492. Train Steps: 10/90 Loss: 0.0375 torch.Size([8, 600, 800])
  23493. torch.Size([8, 8])
  23494. tensor([[0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6063, 0.5617],
  23495. [0.6176, 0.4030, 0.8850, 0.4850, 0.3688, 0.4050, 0.5312, 0.5783],
  23496. [0.6204, 0.4013, 0.8075, 0.2400, 0.4313, 0.2050, 0.5800, 0.5150],
  23497. [0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496],
  23498. [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334],
  23499. [0.6087, 0.3976, 0.8337, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300],
  23500. [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633],
  23501. [0.6200, 0.4049, 0.8638, 0.5617, 0.4125, 0.5100, 0.6013, 0.5317]],
  23502. device='cuda:0', dtype=torch.float64)
  23503. predictions are: tensor([[ 0.1967, -0.6591, 1.7363, -0.0536, -0.5547, -0.1339, 0.6452, 0.2842],
  23504. [ 0.5217, -0.4646, 1.6759, -0.0579, -0.5454, -0.3497, 0.2499, 0.1922],
  23505. [ 0.2723, -0.6024, 1.3251, -1.0565, -0.2461, -1.3451, 0.3772, 0.1343],
  23506. [ 0.2139, -0.6353, 1.6743, -0.9571, -0.2936, -0.9478, 0.5853, 0.1640],
  23507. [ 0.7073, -0.3514, 1.2029, -1.1984, -0.5418, -0.8277, 0.5706, 0.1203],
  23508. [ 0.6006, -0.4165, 1.5692, -0.5432, -0.5794, -0.7226, 0.4796, 0.1666],
  23509. [ 0.5078, -0.4796, 1.7146, 0.1441, -0.4945, -0.1384, 0.4881, 0.3228],
  23510. [ 0.2629, -0.6667, 1.6207, 0.2619, -0.2709, 0.1349, 0.5074, 0.1552]],
  23511. device='cuda:0', grad_fn=<AddmmBackward>)
  23512. landmarks are: tensor([[[ 0.5797, -0.3965, 1.8480, -0.0765, -0.6173, -0.1535, 0.5143,
  23513. 0.3084],
  23514. [ 0.5667, -0.4246, 1.8018, -0.0457, -0.5827, -0.4152, 0.1679,
  23515. 0.3854],
  23516. [ 0.5796, -0.4326, 1.4439, -1.1774, -0.2940, -1.3390, 0.3931,
  23517. 0.0928],
  23518. [ 0.5790, -0.4031, 1.6915, -0.9564, -0.4152, -1.1063, 0.4425,
  23519. 0.2528],
  23520. [ 0.5730, -0.4562, 1.2195, -1.2440, -0.5497, -0.7711, 0.5704,
  23521. 0.1779],
  23522. [ 0.5255, -0.4495, 1.5651, -0.4999, -0.5711, -0.8463, 0.4566,
  23523. 0.1621],
  23524. [ 0.5879, -0.3591, 1.8018, 0.1236, -0.5596, -0.1612, 0.3469,
  23525. 0.3161],
  23526. [ 0.5776, -0.4159, 1.7037, 0.3084, -0.3806, 0.0697, 0.4912,
  23527. 0.1698]]], device='cuda:0')
  23528. loss_train_step before backward: tensor(0.0162, device='cuda:0', grad_fn=<MseLossBackward>)
  23529. loss_train_step after backward: tensor(0.0162, device='cuda:0', grad_fn=<MseLossBackward>)
  23530. loss_train: 0.3914999272674322
  23531. step: 11
  23532. running loss: 0.035590902478857475
  23533. Train Steps: 11/90 Loss: 0.0356 torch.Size([8, 600, 800])
  23534. torch.Size([8, 8])
  23535. tensor([[0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938],
  23536. [0.6034, 0.4011, 0.7350, 0.2533, 0.3438, 0.3367, 0.5516, 0.5084],
  23537. [ nan, nan, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999],
  23538. [0.6250, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6088, 0.5183],
  23539. [0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901],
  23540. [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750],
  23541. [ nan, nan, 0.8300, 0.3150, 0.3588, 0.3383, 0.5208, 0.5194],
  23542. [0.6136, 0.4029, 0.8263, 0.3350, 0.3625, 0.3067, 0.5675, 0.5167]],
  23543. device='cuda:0', dtype=torch.float64)
  23544. predictions are: tensor([[ 0.9551, -0.1943, 1.7838, 0.1884, -0.5463, -0.5159, 0.7997, 0.0161],
  23545. [ 0.8863, -0.2218, 1.0800, -1.0480, -0.5751, -0.7298, 0.4776, 0.0688],
  23546. [-2.0307, -2.0780, 1.1745, -1.1032, -0.3354, -1.3458, 0.2500, 0.2283],
  23547. [ 0.9031, -0.1703, 1.7089, -0.1087, -0.1458, 0.2899, 0.7898, 0.2564],
  23548. [ 0.8976, -0.1821, 1.7356, -0.1796, -0.4199, -0.2579, 0.3009, 0.0515],
  23549. [ 0.9155, -0.2161, 1.6589, 0.0400, -0.2482, -0.0904, 0.4978, 0.2829],
  23550. [-1.8523, -1.9757, 1.3322, -0.7963, -0.5985, -0.8162, 0.2899, 0.2414],
  23551. [ 0.8009, -0.2398, 1.5121, -0.6899, -0.5752, -0.8632, 0.4632, 0.1816]],
  23552. device='cuda:0', grad_fn=<AddmmBackward>)
  23553. landmarks are: tensor([[[ 0.6273, -0.4105, 1.8480, 0.1082, -0.5538, -0.5076, 0.6414,
  23554. -0.0049],
  23555. [ 0.5009, -0.4333, 1.1090, -1.1158, -0.6982, -0.7309, 0.2617,
  23556. 0.0622],
  23557. [-2.2859, -2.2859, 1.3400, -1.0388, -0.3055, -1.4930, 0.1157,
  23558. 0.0231],
  23559. [ 0.6012, -0.3846, 1.7326, -0.0457, -0.1381, 0.2853, 0.5259,
  23560. 0.1082],
  23561. [ 0.5663, -0.3955, 1.7788, -0.2382, -0.4037, -0.2690, 0.0828,
  23562. -0.0220],
  23563. [ 0.5525, -0.4463, 1.7326, 0.0313, -0.2536, -0.0688, 0.1968,
  23564. 0.3700],
  23565. [-2.2859, -2.2859, 1.5478, -0.8309, -0.6289, -0.7232, 0.1198,
  23566. 0.1133],
  23567. [ 0.5483, -0.4249, 1.5305, -0.7386, -0.6115, -0.8694, 0.3353,
  23568. 0.1005]]], device='cuda:0')
  23569. loss_train_step before backward: tensor(0.0300, device='cuda:0', grad_fn=<MseLossBackward>)
  23570. loss_train_step after backward: tensor(0.0300, device='cuda:0', grad_fn=<MseLossBackward>)
  23571. loss_train: 0.42152343317866325
  23572. step: 12
  23573. running loss: 0.03512695276488861
  23574. Train Steps: 12/90 Loss: 0.0351 torch.Size([8, 600, 800])
  23575. torch.Size([8, 8])
  23576. tensor([[0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367],
  23577. [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167],
  23578. [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207],
  23579. [0.6147, 0.4112, 0.7988, 0.3200, 0.3775, 0.2767, 0.5150, 0.5550],
  23580. [0.6055, 0.4015, 0.7425, 0.2033, 0.4113, 0.1883, 0.5217, 0.4823],
  23581. [0.6109, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567],
  23582. [0.6125, 0.3974, 0.7725, 0.2517, 0.3538, 0.3317, 0.5887, 0.5500],
  23583. [0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320]],
  23584. device='cuda:0', dtype=torch.float64)
  23585. predictions are: tensor([[-0.0789, -0.8634, 1.9075, -0.1872, -0.4406, -0.8748, 0.5152, 0.2647],
  23586. [ 0.1872, -0.6352, 1.2463, -0.5061, 0.0230, -1.0992, 0.3518, 0.4947],
  23587. [ 0.6524, -0.4729, 1.6001, -1.0348, -0.2872, -1.0288, 0.8581, -0.0065],
  23588. [-0.0280, -0.7967, 1.5074, -0.6335, -0.5794, -0.7054, 0.1318, 0.1660],
  23589. [-0.4311, -1.0920, 1.1659, -1.0859, -0.4291, -1.2241, 0.2571, -0.0169],
  23590. [ 0.3784, -0.5783, 1.1265, -1.1726, -0.5293, -0.8971, 0.4713, 0.1641],
  23591. [ 0.4242, -0.5747, 1.4232, -0.9487, -0.6321, -0.5319, 0.5436, 0.0818],
  23592. [ 0.7843, -0.3753, 1.9808, -0.0349, -0.5629, -0.0903, 0.9297, -0.0148]],
  23593. device='cuda:0', grad_fn=<AddmmBackward>)
  23594. landmarks are: tensor([[[ 0.5900, -0.3932, 1.8307, -0.3921, -0.4268, -1.1851, 0.3758,
  23595. 0.1929],
  23596. [ 0.6161, -0.3075, 1.1678, -0.6500, 0.0813, -1.4006, 0.2545,
  23597. 0.5624],
  23598. [ 0.6132, -0.4248, 1.5247, -1.1620, -0.2882, -1.3159, 0.6545,
  23599. 0.1193],
  23600. [ 0.5532, -0.3864, 1.4035, -0.8079, -0.5423, -1.0080, 0.0928,
  23601. 0.2776],
  23602. [ 0.5109, -0.4316, 1.1436, -1.3467, -0.3864, -1.4160, 0.1239,
  23603. -0.0580],
  23604. [ 0.5356, -0.4217, 1.0339, -1.4776, -0.5076, -1.1081, 0.4219,
  23605. 0.2853],
  23606. [ 0.5430, -0.4503, 1.2822, -1.1235, -0.6520, -0.7540, 0.4335,
  23607. 0.2545],
  23608. [ 0.6174, -0.4490, 1.8885, -0.0996, -0.4845, -0.3691, 0.9814,
  23609. 0.1715]]], device='cuda:0')
  23610. loss_train_step before backward: tensor(0.0638, device='cuda:0', grad_fn=<MseLossBackward>)
  23611. loss_train_step after backward: tensor(0.0638, device='cuda:0', grad_fn=<MseLossBackward>)
  23612. loss_train: 0.4853547476232052
  23613. step: 13
  23614. running loss: 0.0373349805864004
  23615. Train Steps: 13/90 Loss: 0.0373 torch.Size([8, 600, 800])
  23616. torch.Size([8, 8])
  23617. tensor([[0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750],
  23618. [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567],
  23619. [0.6339, 0.4118, 0.7988, 0.5800, 0.3912, 0.4583, 0.7343, 0.5760],
  23620. [0.6261, 0.3987, 0.8688, 0.4917, 0.4300, 0.5333, 0.7010, 0.5309],
  23621. [0.6278, 0.4253, 0.8875, 0.5017, 0.4113, 0.2750, 0.5413, 0.6196],
  23622. [0.6153, 0.4119, 0.8463, 0.3833, 0.3600, 0.3200, 0.5106, 0.5563],
  23623. [0.6263, 0.4039, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804],
  23624. [0.6357, 0.4118, 0.8400, 0.2500, 0.5413, 0.1633, 0.6725, 0.5586]],
  23625. device='cuda:0', dtype=torch.float64)
  23626. predictions are: tensor([[ 0.3884, -0.5688, 1.7313, -0.1300, -0.3147, -0.0134, 0.2209, 0.1419],
  23627. [ 0.3916, -0.5727, 1.6822, -0.3438, -0.6272, -0.4326, 0.3543, 0.1778],
  23628. [ 0.5572, -0.4856, 1.4594, 0.1052, -0.5221, -0.1518, 0.9008, 0.2143],
  23629. [ 0.1300, -0.7581, 1.7953, -0.2225, -0.3607, 0.2561, 0.8635, 0.0762],
  23630. [ 0.4244, -0.4891, 1.6533, -0.0386, -0.3964, -0.9195, 0.3851, 0.3819],
  23631. [ 0.3393, -0.5632, 1.5823, -0.5397, -0.6619, -0.7676, 0.0628, 0.0785],
  23632. [ 0.5327, -0.5028, 1.9065, -0.4265, -0.6527, -0.1407, 0.5415, -0.1797],
  23633. [ 0.2941, -0.6158, 1.6339, -1.2622, 0.2046, -1.5005, 0.8122, 0.0909]],
  23634. device='cuda:0', grad_fn=<AddmmBackward>)
  23635. landmarks are: tensor([[[ 0.5525, -0.4463, 1.7326, 0.0313, -0.2536, -0.0688, 0.1968,
  23636. 0.3700],
  23637. [ 0.5432, -0.4336, 1.7095, -0.1766, -0.5942, -0.4845, 0.3007,
  23638. 0.2853],
  23639. [ 0.6421, -0.3840, 1.4035, 0.3931, -0.4788, -0.1689, 1.1057,
  23640. 0.3745],
  23641. [ 0.6059, -0.4442, 1.7268, -0.0149, -0.2998, 0.1775, 0.9521,
  23642. 0.1661],
  23643. [ 0.6139, -0.3216, 1.8134, 0.0313, -0.3864, -1.0157, 0.2144,
  23644. 0.5762],
  23645. [ 0.5561, -0.3834, 1.6229, -0.5153, -0.6231, -0.8079, 0.0727,
  23646. 0.2837],
  23647. [ 0.6072, -0.4206, 1.8711, -0.2536, -0.6115, -0.1304, 0.6812,
  23648. -0.0670],
  23649. [ 0.6504, -0.3840, 1.5940, -1.1312, 0.2141, -1.5315, 0.8205,
  23650. 0.2944]]], device='cuda:0')
  23651. loss_train_step before backward: tensor(0.0220, device='cuda:0', grad_fn=<MseLossBackward>)
  23652. loss_train_step after backward: tensor(0.0220, device='cuda:0', grad_fn=<MseLossBackward>)
  23653. loss_train: 0.507318003103137
  23654. step: 14
  23655. running loss: 0.036237000221652646
  23656.  
  23657. Train Steps: 14/90 Loss: 0.0362 torch.Size([8, 600, 800])
  23658. torch.Size([8, 8])
  23659. tensor([[0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133],
  23660. [0.6286, 0.4078, 0.8063, 0.2267, 0.4788, 0.1533, 0.5953, 0.4913],
  23661. [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690],
  23662. [0.6213, 0.4131, 0.8438, 0.3550, 0.3513, 0.4400, 0.5716, 0.5123],
  23663. [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332],
  23664. [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6063, 0.5617],
  23665. [0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297],
  23666. [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333]],
  23667. device='cuda:0', dtype=torch.float64)
  23668. predictions are: tensor([[ 0.5010, -0.5033, 1.5258, -1.1736, -0.0400, -1.5058, 0.5740, 0.0958],
  23669. [-0.6502, -1.2076, 1.3271, -1.1346, -0.1043, -1.5780, 0.2231, 0.0820],
  23670. [ 0.1329, -0.7390, 1.6516, 0.0745, -0.3668, 0.2067, 0.9566, 0.2391],
  23671. [ 0.5369, -0.4364, 1.6081, -0.6326, -0.6196, -0.3833, 0.2194, 0.1052],
  23672. [ 0.6469, -0.4278, 1.8764, -0.1909, -0.5821, -0.4823, 0.7569, 0.0944],
  23673. [ 0.5482, -0.4222, 1.8058, -0.0620, -0.5755, -0.1781, 0.4873, 0.2051],
  23674. [ 0.6252, -0.3851, 1.7494, 0.2343, -0.5574, -0.2445, 0.6326, 0.0882],
  23675. [ 0.5591, -0.4175, 1.8093, -0.2406, -0.5084, -0.0182, 0.4032, 0.1271]],
  23676. device='cuda:0', grad_fn=<AddmmBackward>)
  23677. landmarks are: tensor([[[ 0.6085, -0.4084, 1.5536, -1.1466, -0.0746, -1.4853, 0.6298,
  23678. 0.0851],
  23679. [ 0.6177, -0.4022, 1.4381, -1.2390, -0.0746, -1.5777, 0.4639,
  23680. -0.0168],
  23681. [ 0.6306, -0.4153, 1.5141, 0.2224, -0.3691, 0.2622, 1.0033,
  23682. 0.3425],
  23683. [ 0.5841, -0.3778, 1.6113, -0.6462, -0.6635, -0.2536, 0.3542,
  23684. 0.0802],
  23685. [ 0.6520, -0.4056, 1.9173, -0.0765, -0.5596, -0.4537, 0.7949,
  23686. 0.1768],
  23687. [ 0.5797, -0.3965, 1.8480, -0.0765, -0.6173, -0.1535, 0.5143,
  23688. 0.3084],
  23689. [ 0.6504, -0.3647, 1.7730, 0.2930, -0.6058, -0.2382, 0.7109,
  23690. 0.1608],
  23691. [ 0.6003, -0.3747, 1.8885, -0.1997, -0.5018, -0.0149, 0.5490,
  23692. 0.1775]]], device='cuda:0')
  23693. loss_train_step before backward: tensor(0.0457, device='cuda:0', grad_fn=<MseLossBackward>)
  23694. loss_train_step after backward: tensor(0.0457, device='cuda:0', grad_fn=<MseLossBackward>)
  23695. loss_train: 0.5530113819986582
  23696. step: 15
  23697. running loss: 0.03686742546657721
  23698. Train Steps: 15/90 Loss: 0.0369 torch.Size([8, 600, 800])
  23699. torch.Size([8, 8])
  23700. tensor([[0.6260, 0.4214, 0.8538, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983],
  23701. [0.6078, 0.4033, 0.8019, 0.3055, 0.3450, 0.4200, 0.6025, 0.5550],
  23702. [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378],
  23703. [0.6133, 0.4066, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358],
  23704. [0.6069, 0.3975, 0.8625, 0.5083, 0.4388, 0.5483, 0.5650, 0.4967],
  23705. [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133],
  23706. [0.6364, 0.4165, 0.9088, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297],
  23707. [0.6265, 0.4091, 0.8950, 0.3533, 0.3600, 0.3967, 0.6295, 0.4901]],
  23708. device='cuda:0', dtype=torch.float64)
  23709. predictions are: tensor([[ 5.0703e-01, -4.5975e-01, 1.6656e+00, 3.0593e-01, -4.7732e-01,
  23710. -6.5627e-01, 3.7422e-01, 3.8503e-01],
  23711. [ 2.5524e-01, -6.5057e-01, 1.5289e+00, -9.4455e-01, -6.3493e-01,
  23712. -4.8177e-01, 5.8337e-01, 1.9329e-01],
  23713. [ 2.5623e-01, -6.6214e-01, 1.9205e+00, -1.1357e-01, -4.2664e-01,
  23714. 1.1833e-01, 1.0859e+00, 1.5132e-01],
  23715. [ 5.9606e-01, -4.3202e-01, 9.1692e-01, -1.0829e+00, -4.4209e-01,
  23716. -1.2715e+00, 4.4589e-02, 1.2495e-01],
  23717. [ 2.9305e-01, -6.0549e-01, 1.8639e+00, -2.6410e-02, -2.2537e-01,
  23718. 6.4895e-02, 2.5880e-01, -2.0055e-02],
  23719. [ 2.7239e-01, -6.2210e-01, 1.7858e+00, 4.2429e-02, -1.7373e-01,
  23720. -4.4527e-02, 4.1536e-01, 8.2200e-02],
  23721. [ 4.3774e-01, -5.2086e-01, 1.9156e+00, -2.9562e-01, -3.8331e-01,
  23722. -9.2735e-01, 6.8998e-01, 1.6443e-01],
  23723. [ 6.5677e-01, -3.7615e-01, 1.9344e+00, -6.7801e-01, -5.9268e-01,
  23724. -6.3210e-01, 6.6573e-01, -4.8861e-04]], device='cuda:0',
  23725. grad_fn=<AddmmBackward>)
  23726. landmarks are: tensor([[[ 0.6055, -0.3393, 1.6575, 0.2545, -0.5942, -0.5461, 0.2949,
  23727. 0.4778],
  23728. [ 0.5215, -0.4232, 1.4182, -0.8747, -0.6924, -0.3460, 0.4970,
  23729. 0.2776],
  23730. [ 0.6421, -0.3864, 1.7961, 0.0543, -0.4383, 0.2237, 1.2007,
  23731. 0.1982],
  23732. [ 0.5470, -0.4081, 0.8492, -1.0773, -0.5307, -1.1620, 0.0912,
  23733. 0.1890],
  23734. [ 0.5174, -0.4497, 1.6979, 0.0620, -0.2594, 0.2468, 0.3238,
  23735. 0.0082],
  23736. [ 0.5786, -0.4141, 1.7037, 0.1544, -0.1862, 0.0736, 0.4393,
  23737. 0.0851],
  23738. [ 0.6537, -0.3623, 1.9115, -0.2690, -0.4037, -0.8309, 0.6926,
  23739. 0.1608],
  23740. [ 0.6077, -0.3965, 1.8480, -0.6539, -0.6231, -0.4537, 0.6216,
  23741. -0.0220]]], device='cuda:0')
  23742. loss_train_step before backward: tensor(0.0160, device='cuda:0', grad_fn=<MseLossBackward>)
  23743. loss_train_step after backward: tensor(0.0160, device='cuda:0', grad_fn=<MseLossBackward>)
  23744. loss_train: 0.5689659360796213
  23745. step: 16
  23746. running loss: 0.03556037100497633
  23747. Train Steps: 16/90 Loss: 0.0356 torch.Size([8, 600, 800])
  23748. torch.Size([8, 8])
  23749. tensor([[0.6230, 0.4113, 0.7213, 0.1983, 0.4325, 0.2367, 0.6262, 0.5400],
  23750. [0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320],
  23751. [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398],
  23752. [0.6102, 0.3999, 0.8750, 0.5133, 0.3825, 0.4750, 0.5637, 0.5083],
  23753. [0.6250, 0.4110, 0.7238, 0.2067, 0.4263, 0.1883, 0.5625, 0.5633],
  23754. [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317],
  23755. [0.6248, 0.4185, 0.8500, 0.5767, 0.4463, 0.4550, 0.5613, 0.5917],
  23756. [0.6115, 0.4081, 0.6725, 0.2433, 0.4088, 0.1933, 0.5167, 0.5544]],
  23757. device='cuda:0', dtype=torch.float64)
  23758. predictions are: tensor([[-0.0935, -0.8722, 1.3125, -1.3468, -0.2899, -1.2047, 0.7003, 0.2224],
  23759. [ 0.2963, -0.6498, 2.0224, -0.2893, -0.5738, -0.3291, 0.8886, 0.0073],
  23760. [ 0.6682, -0.4188, 1.9193, 0.1636, -0.6020, -0.1303, 0.5732, 0.0660],
  23761. [ 0.6721, -0.3713, 1.9908, -0.0065, -0.5589, 0.0488, 0.3812, -0.0499],
  23762. [ 0.1340, -0.7117, 1.1896, -1.2837, -0.2576, -1.3935, 0.3966, 0.1664],
  23763. [ 0.5357, -0.4752, 1.9282, 0.1077, -0.5723, -0.4885, 0.5420, 0.1317],
  23764. [ 0.7458, -0.3102, 1.8443, 0.3719, -0.2171, -0.0943, 0.5615, 0.3431],
  23765. [ 0.4249, -0.5251, 1.0434, -1.1636, -0.3758, -1.2958, 0.1229, 0.1956]],
  23766. device='cuda:0', grad_fn=<AddmmBackward>)
  23767. landmarks are: tensor([[[ 0.5917, -0.3861, 1.0455, -1.3698, -0.2882, -1.1928, 0.6067,
  23768. 0.2083],
  23769. [ 0.6174, -0.4490, 1.8885, -0.0996, -0.4845, -0.3691, 0.9814,
  23770. 0.1715],
  23771. [ 0.5786, -0.4463, 1.6655, 0.2216, -0.5115, -0.2675, 0.4236,
  23772. 0.2075],
  23773. [ 0.5326, -0.4390, 1.7557, 0.0851, -0.5192, -0.0919, 0.3180,
  23774. 0.0620],
  23775. [ 0.6010, -0.3875, 1.0570, -1.3313, -0.3171, -1.4160, 0.3122,
  23776. 0.3161],
  23777. [ 0.5901, -0.4157, 1.7557, 0.1929, -0.5423, -0.5923, 0.3584,
  23778. 0.1698],
  23779. [ 0.5999, -0.3530, 1.6402, 0.3777, -0.2247, -0.1843, 0.3065,
  23780. 0.4470],
  23781. [ 0.5385, -0.4010, 0.8203, -1.1620, -0.3979, -1.3929, 0.1005,
  23782. 0.2747]]], device='cuda:0')
  23783. loss_train_step before backward: tensor(0.0314, device='cuda:0', grad_fn=<MseLossBackward>)
  23784. loss_train_step after backward: tensor(0.0314, device='cuda:0', grad_fn=<MseLossBackward>)
  23785. loss_train: 0.6003932412713766
  23786. step: 17
  23787. running loss: 0.035317249486551565
  23788. Train Steps: 17/90 Loss: 0.0353 torch.Size([8, 600, 800])
  23789. torch.Size([8, 8])
  23790. tensor([[0.6069, 0.3975, 0.8625, 0.5083, 0.4388, 0.5483, 0.5650, 0.4967],
  23791. [0.6179, 0.4008, 0.8600, 0.4015, 0.3932, 0.2515, 0.5711, 0.5438],
  23792. [0.6197, 0.3986, 0.8800, 0.4617, 0.4188, 0.4783, 0.5687, 0.5550],
  23793. [0.6304, 0.4024, 0.8925, 0.4800, 0.3937, 0.4817, 0.7485, 0.5297],
  23794. [0.6187, 0.4104, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683],
  23795. [0.6090, 0.4010, 0.7838, 0.3483, 0.3538, 0.3783, 0.5462, 0.5077],
  23796. [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351],
  23797. [0.6147, 0.4081, 0.8538, 0.3400, 0.3663, 0.3150, 0.5142, 0.4875]],
  23798. device='cuda:0', dtype=torch.float64)
  23799. predictions are: tensor([[ 0.3610, -0.5678, 1.9080, 0.1425, -0.2585, 0.2075, 0.4586, 0.0328],
  23800. [ 0.5887, -0.4187, 1.6380, -0.3694, -0.3123, -1.1082, 0.3524, 0.3063],
  23801. [ 0.3034, -0.5894, 1.8729, -0.0846, -0.3672, 0.0066, 0.5765, 0.1857],
  23802. [ 0.2782, -0.6576, 1.9356, -0.0492, -0.4911, -0.0651, 1.0794, 0.1156],
  23803. [ 0.3067, -0.6011, 1.0080, -1.2048, -0.4256, -1.1716, 0.5215, 0.3087],
  23804. [ 0.1235, -0.7251, 1.5411, -0.6893, -0.6735, -0.5725, 0.3647, 0.0733],
  23805. [ 0.6070, -0.4165, 1.3449, -0.9162, -0.1758, -1.4343, 0.4269, 0.1630],
  23806. [ 0.7717, -0.3102, 1.7616, -0.6037, -0.6002, -0.8472, 0.3638, 0.0585]],
  23807. device='cuda:0', grad_fn=<AddmmBackward>)
  23808. landmarks are: tensor([[[ 0.5174, -0.4497, 1.6979, 0.0620, -0.2594, 0.2468, 0.3238,
  23809. 0.0082],
  23810. [ 0.5680, -0.4345, 1.6864, -0.4315, -0.4698, -1.1241, 0.3518,
  23811. 0.2261],
  23812. [ 0.5763, -0.4450, 1.7788, -0.1535, -0.3517, -0.0765, 0.3411,
  23813. 0.2776],
  23814. [ 0.6257, -0.4273, 1.8365, -0.0688, -0.4672, -0.0611, 1.1715,
  23815. 0.1608],
  23816. [ 0.5718, -0.3905, 1.0053, -1.3305, -0.4614, -1.1235, 0.4450,
  23817. 0.3392],
  23818. [ 0.5270, -0.4336, 1.3342, -0.6770, -0.6520, -0.5384, 0.2370,
  23819. 0.0592],
  23820. [ 0.5664, -0.4321, 1.2862, -1.0003, -0.2189, -1.4608, 0.3883,
  23821. 0.1855],
  23822. [ 0.5532, -0.4008, 1.6575, -0.7155, -0.5942, -0.8309, 0.0890,
  23823. -0.0340]]], device='cuda:0')
  23824. loss_train_step before backward: tensor(0.0180, device='cuda:0', grad_fn=<MseLossBackward>)
  23825. loss_train_step after backward: tensor(0.0180, device='cuda:0', grad_fn=<MseLossBackward>)
  23826. loss_train: 0.618410861119628
  23827. step: 18
  23828. running loss: 0.03435615895109044
  23829.  
  23830. Train Steps: 18/90 Loss: 0.0344 torch.Size([8, 600, 800])
  23831. torch.Size([8, 8])
  23832. tensor([[0.6275, 0.4024, 0.8600, 0.2283, 0.5350, 0.1800, 0.7074, 0.5413],
  23833. [0.6185, 0.4098, 0.8838, 0.4900, 0.4537, 0.5800, 0.6288, 0.5400],
  23834. [0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118],
  23835. [0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945],
  23836. [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795],
  23837. [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377],
  23838. [0.6346, 0.4086, 0.7938, 0.5500, 0.3962, 0.4867, 0.7343, 0.5702],
  23839. [0.6092, 0.4001, 0.8638, 0.4867, 0.4288, 0.5367, 0.5484, 0.5064]],
  23840. device='cuda:0', dtype=torch.float64)
  23841. predictions are: tensor([[ 0.1828, -0.6863, 1.7455, -1.3498, 0.0840, -1.6401, 0.8802, 0.0811],
  23842. [ 0.5804, -0.3763, 1.8545, -0.1124, -0.2649, 0.2308, 0.4292, 0.1976],
  23843. [ 0.4637, -0.4524, 1.1475, -0.7110, -0.3402, -1.3154, 0.0892, 0.4226],
  23844. [ 0.5057, -0.4969, 1.9738, -0.3320, -0.3745, -0.6609, 0.8750, 0.2639],
  23845. [ 0.7743, -0.3068, 1.7448, 0.1914, -0.5778, 0.1589, 0.8097, 0.1874],
  23846. [ 0.4554, -0.5175, 1.7897, 0.2008, -0.5382, -0.1848, 0.2942, 0.0439],
  23847. [ 0.6566, -0.3793, 1.5341, 0.1815, -0.5751, -0.1516, 0.8652, 0.2552],
  23848. [ 0.3685, -0.5335, 1.7676, -0.0991, -0.3964, 0.0852, 0.0582, 0.0316]],
  23849. device='cuda:0', grad_fn=<AddmmBackward>)
  23850. landmarks are: tensor([[[ 0.6125, -0.4273, 1.6864, -1.2313, 0.1852, -1.4545, 0.9814,
  23851. 0.2142],
  23852. [ 0.5709, -0.3933, 1.7961, -0.0226, -0.1901, 0.3931, 0.6182,
  23853. 0.2083],
  23854. [ 0.6077, -0.3226, 0.9993, -0.6462, -0.2651, -1.3082, 0.2946,
  23855. 0.5401],
  23856. [ 0.6471, -0.3719, 1.9866, -0.3921, -0.3806, -0.5538, 1.0070,
  23857. 0.4600],
  23858. [ 0.6355, -0.4080, 1.6113, 0.1852, -0.4730, 0.1467, 0.9996,
  23859. 0.3905],
  23860. [ 0.5775, -0.4607, 1.6741, 0.1962, -0.4036, -0.1212, 0.4588,
  23861. 0.1979],
  23862. [ 0.6454, -0.3984, 1.3804, 0.2545, -0.4557, -0.0380, 1.1057,
  23863. 0.3478],
  23864. [ 0.5278, -0.4377, 1.7037, -0.0380, -0.3055, 0.1929, 0.2473,
  23865. 0.0532]]], device='cuda:0')
  23866. loss_train_step before backward: tensor(0.0174, device='cuda:0', grad_fn=<MseLossBackward>)
  23867. loss_train_step after backward: tensor(0.0174, device='cuda:0', grad_fn=<MseLossBackward>)
  23868. loss_train: 0.6358400098979473
  23869. step: 19
  23870. running loss: 0.033465263678839334
  23871. Train Steps: 19/90 Loss: 0.0335 torch.Size([8, 600, 800])
  23872. torch.Size([8, 8])
  23873. tensor([[0.6100, 0.4071, 0.7601, 0.3444, 0.3400, 0.4117, 0.5625, 0.5617],
  23874. [0.6201, 0.3970, 0.8413, 0.4950, 0.4413, 0.5183, 0.6088, 0.5400],
  23875. [0.6265, 0.4091, 0.8950, 0.3533, 0.3600, 0.3967, 0.6295, 0.4901],
  23876. [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750],
  23877. [0.6122, 0.4006, 0.8850, 0.4217, 0.4088, 0.5517, 0.6063, 0.5517],
  23878. [0.6339, 0.4149, 0.8800, 0.5000, 0.3900, 0.5283, 0.7541, 0.5424],
  23879. [ nan, nan, 0.7192, 0.2346, 0.4037, 0.2050, 0.5138, 0.5650],
  23880. [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378]],
  23881. device='cuda:0', dtype=torch.float64)
  23882. predictions are: tensor([[ 0.7286, -0.3021, 1.2757, -0.6294, -0.6770, -0.5504, 0.3182, 0.2894],
  23883. [ 0.6831, -0.3413, 1.7177, 0.1549, -0.2115, -0.1291, 0.4543, 0.2399],
  23884. [ 1.0485, -0.1073, 1.8392, -0.5684, -0.5598, -0.7251, 0.6200, 0.0672],
  23885. [ 0.6044, -0.4001, 1.7955, 0.1472, -0.2306, -0.2585, 0.1774, 0.2846],
  23886. [ 0.6464, -0.3412, 1.8233, -0.2097, -0.3579, 0.0401, 0.5192, 0.2039],
  23887. [ 0.8227, -0.2677, 1.8455, 0.1839, -0.4548, 0.0273, 0.8990, 0.1851],
  23888. [-1.6205, -1.8727, 1.1033, -1.1597, -0.3460, -1.5958, 0.1061, 0.0953],
  23889. [ 0.7472, -0.3128, 1.8387, 0.1208, -0.4073, 0.0710, 1.0212, 0.2521]],
  23890. device='cuda:0', grad_fn=<AddmmBackward>)
  23891. landmarks are: tensor([[[ 5.3178e-01, -4.0564e-01, 1.2249e+00, -6.9494e-01, -7.1547e-01,
  23892. -3.8445e-01, 3.1224e-01, 3.0839e-01],
  23893. [ 5.7812e-01, -4.5219e-01, 1.5998e+00, 4.6189e-04, -2.4781e-01,
  23894. 1.0824e-01, 5.2587e-01, 2.0831e-01],
  23895. [ 6.0774e-01, -3.9646e-01, 1.8480e+00, -6.5389e-01, -6.2309e-01,
  23896. -4.5373e-01, 6.2155e-01, -2.1963e-02],
  23897. [ 5.5254e-01, -4.4627e-01, 1.7326e+00, 3.1255e-02, -2.5358e-01,
  23898. -6.8822e-02, 1.9677e-01, 3.6998e-01],
  23899. [ 5.4169e-01, -4.3549e-01, 1.8018e+00, -3.3826e-01, -3.9792e-01,
  23900. 2.6220e-01, 5.1432e-01, 2.6220e-01],
  23901. [ 6.4212e-01, -3.6953e-01, 1.7788e+00, 2.3557e-02, -4.8453e-01,
  23902. 1.5443e-01, 1.1971e+00, 2.1955e-01],
  23903. [-2.2859e+00, -2.2859e+00, 1.0361e+00, -1.2021e+00, -4.2102e-01,
  23904. -1.3390e+00, 8.7067e-02, 3.2379e-01],
  23905. [ 6.4212e-01, -3.8638e-01, 1.7961e+00, 5.4350e-02, -4.3834e-01,
  23906. 2.2371e-01, 1.2007e+00, 1.9818e-01]]], device='cuda:0')
  23907. loss_train_step before backward: tensor(0.0273, device='cuda:0', grad_fn=<MseLossBackward>)
  23908. loss_train_step after backward: tensor(0.0273, device='cuda:0', grad_fn=<MseLossBackward>)
  23909. loss_train: 0.6631092224270105
  23910. step: 20
  23911. running loss: 0.033155461121350524
  23912. Train Steps: 20/90 Loss: 0.0332 torch.Size([8, 600, 800])
  23913. torch.Size([8, 8])
  23914. tensor([[0.6198, 0.4101, 0.8838, 0.5283, 0.3763, 0.5267, 0.5913, 0.5567],
  23915. [0.6339, 0.4102, 0.9088, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390],
  23916. [0.6265, 0.4088, 0.8025, 0.1850, 0.4163, 0.2500, 0.6290, 0.4947],
  23917. [0.6371, 0.4092, 0.8337, 0.5850, 0.3950, 0.5117, 0.6559, 0.5262],
  23918. [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082],
  23919. [0.6129, 0.4069, 0.8750, 0.5067, 0.3875, 0.4233, 0.5235, 0.5881],
  23920. [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098],
  23921. [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483]],
  23922. device='cuda:0', dtype=torch.float64)
  23923. predictions are: tensor([[ 0.5992, -0.3980, 1.7403, 0.2467, -0.4589, -0.0764, 0.5850, 0.3452],
  23924. [ 0.3023, -0.6008, 1.8628, -0.1265, -0.4278, -0.0498, 1.0072, 0.2887],
  23925. [ 0.8948, -0.2468, 1.3591, -1.4355, -0.3035, -1.3032, 0.6622, 0.1346],
  23926. [ 0.8589, -0.2448, 1.5960, 0.3670, -0.4593, -0.1402, 0.7276, 0.1638],
  23927. [ 0.5828, -0.4151, 1.7459, 0.1039, -0.2371, 0.0888, 0.3880, 0.1641],
  23928. [ 0.5667, -0.4005, 1.7422, 0.1350, -0.4795, -0.4804, 0.1186, 0.3531],
  23929. [ 0.5436, -0.4430, 1.7085, -0.0783, -0.4442, -0.0420, 0.4933, 0.1895],
  23930. [ 0.4270, -0.5031, 1.7202, -0.4630, -0.4792, -0.2533, 0.3859, 0.2487]],
  23931. device='cuda:0', grad_fn=<AddmmBackward>)
  23932. landmarks are: tensor([[[ 5.7696e-01, -3.9176e-01, 1.7961e+00, 1.5443e-01, -5.4804e-01,
  23933. 1.4673e-01, 4.4503e-01, 2.8530e-01],
  23934. [ 6.4212e-01, -3.9120e-01, 1.9115e+00, -8.4219e-02, -4.7298e-01,
  23935. 1.5443e-01, 1.1824e+00, 2.0352e-01],
  23936. [ 6.0785e-01, -3.9761e-01, 1.4208e+00, -1.4314e+00, -3.6328e-01,
  23937. -1.1312e+00, 6.1950e-01, -9.2270e-04],
  23938. [ 6.5695e-01, -3.9601e-01, 1.5651e+00, 4.1617e-01, -4.6143e-01,
  23939. 7.7444e-02, 7.4375e-01, 1.4474e-01],
  23940. [ 5.4496e-01, -4.7064e-01, 1.7643e+00, 7.2204e-02, -3.7076e-01,
  23941. 3.2001e-01, 4.8543e-01, 6.1219e-02],
  23942. [ 5.4515e-01, -4.0670e-01, 1.7557e+00, 5.4350e-02, -4.9607e-01,
  23943. -3.3056e-01, 1.3228e-01, 4.3063e-01],
  23944. [ 5.4660e-01, -4.7064e-01, 1.7198e+00, -9.0292e-02, -5.7125e-01,
  23945. 1.2613e-01, 4.7328e-01, 6.8827e-02],
  23946. [ 5.5456e-01, -4.6205e-01, 1.7788e+00, -4.2294e-01, -5.1917e-01,
  23947. -2.2633e-02, 4.2771e-01, 2.4681e-01]]], device='cuda:0')
  23948. loss_train_step before backward: tensor(0.0139, device='cuda:0', grad_fn=<MseLossBackward>)
  23949. loss_train_step after backward: tensor(0.0139, device='cuda:0', grad_fn=<MseLossBackward>)
  23950. loss_train: 0.6769989067688584
  23951. step: 21
  23952. running loss: 0.03223804317946945
  23953. Train Steps: 21/90 Loss: 0.0322 torch.Size([8, 600, 800])
  23954. torch.Size([8, 8])
  23955. tensor([[0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895],
  23956. [0.6152, 0.4131, 0.6863, 0.2567, 0.3625, 0.3300, 0.5765, 0.5305],
  23957. [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517],
  23958. [0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626],
  23959. [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633],
  23960. [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116],
  23961. [0.6165, 0.4106, 0.7575, 0.1733, 0.3838, 0.2650, 0.5680, 0.5116],
  23962. [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400]],
  23963. device='cuda:0', dtype=torch.float64)
  23964. predictions are: tensor([[ 0.6423, -0.4058, 1.3525, -1.2299, -0.0814, -1.5329, 0.4613, 0.1552],
  23965. [ 0.3405, -0.5563, 1.0578, -0.8936, -0.5718, -0.7640, 0.4300, 0.3576],
  23966. [ 0.5428, -0.4354, 1.0600, -0.8942, -0.3129, -1.1778, 0.3834, 0.4182],
  23967. [ 0.7124, -0.3692, 1.7740, 0.6887, -0.4695, -0.0495, 0.6066, 0.1195],
  23968. [ 0.4637, -0.4929, 1.9336, 0.4006, -0.5610, -0.1530, 0.5893, 0.3991],
  23969. [ 0.6304, -0.3523, 1.9097, -0.1257, -0.4233, 0.4380, 0.6059, 0.1707],
  23970. [ 0.7436, -0.3395, 1.3049, -1.1675, -0.4783, -1.1101, 0.5469, 0.1452],
  23971. [ 0.1071, -0.7279, 1.7459, -0.5140, -0.5872, -0.1503, 0.7996, 0.2985]],
  23972. device='cuda:0', grad_fn=<AddmmBackward>)
  23973. landmarks are: tensor([[[ 0.5828, -0.4417, 1.2476, -1.3929, -0.1727, -1.5700, 0.4694,
  23974. -0.0248],
  23975. [ 0.5557, -0.3779, 0.8838, -1.1004, -0.6115, -0.7617, 0.3769,
  23976. 0.1644],
  23977. [ 0.5713, -0.3671, 0.8665, -1.0696, -0.3691, -1.2236, 0.3527,
  23978. 0.2622],
  23979. [ 0.6105, -0.4293, 1.5824, 0.5239, -0.4730, -0.0380, 0.5025,
  23980. -0.1492],
  23981. [ 0.5879, -0.3591, 1.8018, 0.1236, -0.5596, -0.1612, 0.3469,
  23982. 0.3161],
  23983. [ 0.5376, -0.3903, 1.7095, -0.3229, -0.4730, 0.4701, 0.3871,
  23984. 0.0772],
  23985. [ 0.5618, -0.3895, 1.2129, -1.4853, -0.5134, -1.0619, 0.3378,
  23986. 0.0772],
  23987. [ 0.5631, -0.4008, 1.6113, -0.7309, -0.6693, -0.1304, 0.5836,
  23988. 0.2083]]], device='cuda:0')
  23989. loss_train_step before backward: tensor(0.0238, device='cuda:0', grad_fn=<MseLossBackward>)
  23990. loss_train_step after backward: tensor(0.0238, device='cuda:0', grad_fn=<MseLossBackward>)
  23991. loss_train: 0.7008060766384006
  23992. step: 22
  23993. running loss: 0.031854821665381845
  23994.  
  23995. Train Steps: 22/90 Loss: 0.0319 torch.Size([8, 600, 800])
  23996. torch.Size([8, 8])
  23997. tensor([[0.6097, 0.4024, 0.8488, 0.3717, 0.3875, 0.5517, 0.5836, 0.5591],
  23998. [0.6251, 0.4163, 0.8662, 0.4467, 0.3625, 0.3567, 0.6038, 0.5533],
  23999. [0.6161, 0.4024, 0.8662, 0.4683, 0.4935, 0.5364, 0.6063, 0.5567],
  24000. [ nan, nan, 0.7225, 0.2167, 0.3987, 0.2283, 0.5427, 0.5181],
  24001. [0.6266, 0.4070, 0.8712, 0.5600, 0.3713, 0.4783, 0.5775, 0.6100],
  24002. [0.6100, 0.4016, 0.8600, 0.5067, 0.4612, 0.5233, 0.5086, 0.5519],
  24003. [0.6185, 0.4098, 0.8838, 0.4900, 0.4537, 0.5800, 0.6288, 0.5400],
  24004. [0.6271, 0.4040, 0.9138, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413]],
  24005. device='cuda:0', dtype=torch.float64)
  24006. predictions are: tensor([[ 0.5192, -0.4303, 1.5533, -0.5186, -0.4992, 0.2291, 0.8167, 0.2161],
  24007. [ 0.8398, -0.2164, 1.5734, -0.2260, -0.6431, -0.7362, 0.4513, 0.3491],
  24008. [ 0.7983, -0.2492, 1.6775, -0.1223, -0.1407, 0.0391, 0.6358, 0.2704],
  24009. [-0.8730, -1.3414, 0.9229, -1.2942, -0.4417, -1.3333, 0.1050, 0.1558],
  24010. [ 0.7856, -0.2575, 1.5907, 0.4904, -0.6020, -0.0994, 0.5406, 0.3689],
  24011. [ 0.7232, -0.2932, 1.6110, 0.1631, -0.2535, 0.0306, 0.2218, 0.2326],
  24012. [ 0.7808, -0.2383, 1.7002, 0.0022, -0.2485, 0.3157, 0.6546, 0.3287],
  24013. [ 0.6408, -0.3928, 1.9235, -0.6045, -0.2477, -1.1302, 0.9770, 0.1677]],
  24014. device='cuda:0', grad_fn=<AddmmBackward>)
  24015. landmarks are: tensor([[[ 0.5301, -0.4273, 1.6344, -0.5692, -0.4961, 0.2622, 0.4098,
  24016. 0.2966],
  24017. [ 0.6016, -0.3633, 1.7152, -0.2228, -0.6115, -0.6385, 0.5028,
  24018. 0.2699],
  24019. [ 0.5598, -0.4273, 1.7152, -0.1227, -0.0065, 0.1917, 0.5143,
  24020. 0.2853],
  24021. [-2.2859, -2.2859, 1.0513, -1.2851, -0.4441, -1.2313, 0.2206,
  24022. 0.1073],
  24023. [ 0.6081, -0.4059, 1.7383, 0.3007, -0.5711, -0.0765, 0.3815,
  24024. 0.5316],
  24025. [ 0.5318, -0.4310, 1.6864, 0.0543, -0.1554, 0.1313, 0.0635,
  24026. 0.2634],
  24027. [ 0.5709, -0.3933, 1.7961, -0.0226, -0.1901, 0.3931, 0.6182,
  24028. 0.2083],
  24029. [ 0.6108, -0.4201, 1.9346, -0.5538, -0.1497, -1.0773, 1.0545,
  24030. 0.2142]]], device='cuda:0')
  24031. loss_train_step before backward: tensor(0.0592, device='cuda:0', grad_fn=<MseLossBackward>)
  24032. loss_train_step after backward: tensor(0.0592, device='cuda:0', grad_fn=<MseLossBackward>)
  24033. loss_train: 0.7600116161629558
  24034. step: 23
  24035. running loss: 0.03304398331143286
  24036. Train Steps: 23/90 Loss: 0.0330 torch.Size([8, 600, 800])
  24037. torch.Size([8, 8])
  24038. tensor([[0.6147, 0.4107, 0.8137, 0.3333, 0.3750, 0.2683, 0.5006, 0.5412],
  24039. [0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431],
  24040. [0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869],
  24041. [0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950],
  24042. [0.6250, 0.4110, 0.7238, 0.2067, 0.4263, 0.1883, 0.5625, 0.5633],
  24043. [0.6193, 0.3930, 0.8949, 0.4437, 0.3852, 0.5435, 0.6263, 0.5263],
  24044. [ nan, nan, 0.6412, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700],
  24045. [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267]],
  24046. device='cuda:0', dtype=torch.float64)
  24047. predictions are: tensor([[ 0.6867, -0.2904, 1.3013, -0.6740, -0.6040, -0.8209, 0.2258, 0.2238],
  24048. [ 0.7138, -0.2789, 1.6832, 0.0185, -0.3824, 0.1739, 0.3402, 0.2216],
  24049. [ 0.8082, -0.2157, 1.4416, -1.2298, -0.1667, -1.0268, 0.7925, 0.1555],
  24050. [ 0.6891, -0.3364, 1.7066, 0.2175, -0.3908, 0.2481, 0.6137, 0.1848],
  24051. [ 0.6694, -0.3064, 1.0297, -1.0955, -0.3065, -1.2666, 0.5156, 0.3473],
  24052. [ 0.6522, -0.3592, 1.7807, -0.1508, -0.5552, 0.2680, 0.8259, 0.2385],
  24053. [-1.8048, -1.9724, 0.9865, -1.1424, -0.2959, -1.4394, 0.2511, 0.3223],
  24054. [ 0.8061, -0.2563, 1.6035, 0.4277, -0.5191, -0.0406, 0.5202, 0.2413]],
  24055. device='cuda:0', grad_fn=<AddmmBackward>)
  24056. landmarks are: tensor([[[ 5.5319e-01, -3.8879e-01, 1.4727e+00, -7.4627e-01, -5.5381e-01,
  24057. -1.0465e+00, 2.6467e-02, 2.1383e-01],
  24058. [ 5.3508e-01, -4.1527e-01, 1.7326e+00, -4.5727e-02, -2.2139e-01,
  24059. -4.6642e-02, 4.3431e-02, 2.2284e-01],
  24060. [ 6.1270e-01, -3.9438e-01, 1.5189e+00, -1.2467e+00, -1.3233e-01,
  24061. -1.4622e+00, 5.6463e-01, -3.6943e-02],
  24062. [ 5.8199e-01, -4.7544e-01, 1.7095e+00, 1.9292e-01, -2.5358e-01,
  24063. 5.4350e-02, 4.9700e-01, 4.6189e-04],
  24064. [ 6.0098e-01, -3.8745e-01, 1.0570e+00, -1.3313e+00, -3.1709e-01,
  24065. -1.4160e+00, 3.1224e-01, 3.1609e-01],
  24066. [ 5.7460e-01, -4.7064e-01, 1.8476e+00, -2.3654e-01, -5.0683e-01,
  24067. 2.2450e-01, 6.0688e-01, 1.4491e-01],
  24068. [-2.2859e+00, -2.2859e+00, 6.7598e-01, -1.4083e+00, -3.2864e-01,
  24069. -1.4160e+00, 2.4873e-01, 3.4688e-01],
  24070. [ 5.8915e-01, -4.5504e-01, 1.5132e+00, 3.5458e-01, -3.6905e-01,
  24071. -1.5350e-01, 3.8152e-01, 1.4673e-01]]], device='cuda:0')
  24072. loss_train_step before backward: tensor(0.0274, device='cuda:0', grad_fn=<MseLossBackward>)
  24073. loss_train_step after backward: tensor(0.0274, device='cuda:0', grad_fn=<MseLossBackward>)
  24074. loss_train: 0.7874013772234321
  24075. step: 24
  24076. running loss: 0.032808390717643
  24077. Train Steps: 24/90 Loss: 0.0328 torch.Size([8, 600, 800])
  24078. torch.Size([8, 8])
  24079. tensor([[0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760],
  24080. [0.6277, 0.4057, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823],
  24081. [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320],
  24082. [0.6276, 0.4120, 0.8738, 0.3133, 0.4225, 0.2217, 0.6203, 0.4892],
  24083. [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466],
  24084. [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395],
  24085. [0.6230, 0.4152, 0.7588, 0.2283, 0.4012, 0.2883, 0.6200, 0.5767],
  24086. [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150]],
  24087. device='cuda:0', dtype=torch.float64)
  24088. predictions are: tensor([[ 0.5090, -0.4777, 1.5581, 0.3141, -0.4932, 0.0368, 0.8771, 0.3633],
  24089. [ 0.2972, -0.5898, 1.5307, -1.0487, -0.2588, -1.2857, 0.4801, 0.0661],
  24090. [ 0.3602, -0.5758, 1.5941, 0.1848, -0.5014, 0.1306, 0.7942, 0.2363],
  24091. [ 0.6912, -0.3386, 1.7118, -0.7506, -0.3744, -1.1410, 0.4392, 0.1324],
  24092. [ 0.4615, -0.4464, 1.5727, 0.1339, -0.2828, 0.2632, 0.0648, 0.2474],
  24093. [ 0.6099, -0.3904, 0.9968, -1.2696, -0.4457, -0.9995, 0.4334, 0.2862],
  24094. [ 0.7335, -0.3144, 1.1439, -1.0194, -0.5125, -0.8689, 0.5599, 0.4064],
  24095. [ 0.5387, -0.4068, 1.7121, -0.0485, -0.3344, 0.5094, 0.5716, 0.2331]],
  24096. device='cuda:0', grad_fn=<AddmmBackward>)
  24097. landmarks are: tensor([[[ 0.6421, -0.4008, 1.6055, 0.2160, -0.5076, -0.0534, 1.1020,
  24098. 0.3745],
  24099. [ 0.6132, -0.4122, 1.5478, -1.0619, -0.2709, -1.4314, 0.5500,
  24100. -0.0583],
  24101. [ 0.6273, -0.4393, 1.6402, 0.1313, -0.5076, 0.0467, 1.1532,
  24102. 0.1715],
  24103. [ 0.6128, -0.3828, 1.7499, -0.8386, -0.3344, -1.2620, 0.5792,
  24104. -0.0263],
  24105. [ 0.5178, -0.4480, 1.6748, 0.0620, -0.2767, 0.2083, 0.1067,
  24106. 0.2386],
  24107. [ 0.5713, -0.4538, 1.1028, -1.3659, -0.3831, -1.1273, 0.5340,
  24108. 0.2058],
  24109. [ 0.5915, -0.3682, 1.2187, -1.2313, -0.4326, -0.9541, 0.5778,
  24110. 0.3777],
  24111. [ 0.6014, -0.3883, 1.8192, -0.1150, -0.2882, 0.4008, 0.5201,
  24112. 0.0928]]], device='cuda:0')
  24113. loss_train_step before backward: tensor(0.0127, device='cuda:0', grad_fn=<MseLossBackward>)
  24114. loss_train_step after backward: tensor(0.0127, device='cuda:0', grad_fn=<MseLossBackward>)
  24115. loss_train: 0.8001020578667521
  24116. step: 25
  24117. running loss: 0.032004082314670085
  24118. Train Steps: 25/90 Loss: 0.0320 torch.Size([8, 600, 800])
  24119. torch.Size([8, 8])
  24120. tensor([[0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750],
  24121. [0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019],
  24122. [0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183],
  24123. [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5637, 0.5633],
  24124. [0.6128, 0.4022, 0.8738, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064],
  24125. [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5413, 0.5717],
  24126. [ nan, nan, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650],
  24127. [0.6128, 0.4084, 0.8738, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397]],
  24128. device='cuda:0', dtype=torch.float64)
  24129. predictions are: tensor([[ 0.7484, -0.2701, 0.9943, -1.1130, -0.4974, -0.8206, 0.6677, 0.3461],
  24130. [ 0.5576, -0.4452, 1.8740, -0.1048, -0.5938, -0.3413, 0.9014, 0.0819],
  24131. [ 0.8929, -0.1953, 1.1599, -0.9577, -0.4125, -0.6735, 0.6857, 0.4512],
  24132. [ 0.9197, -0.1350, 1.6207, -0.0053, -0.5782, -0.4007, 0.4676, 0.3432],
  24133. [ 0.7223, -0.2809, 1.6813, 0.0196, -0.0342, 0.1632, 0.1919, 0.1328],
  24134. [ 0.5366, -0.4327, 1.7118, -0.0477, -0.3475, 0.1408, 0.4257, 0.2003],
  24135. [-1.7881, -1.9381, 1.0119, -1.3121, -0.1923, -1.0895, 0.3048, 0.1346],
  24136. [ 0.8429, -0.1873, 1.6756, -0.1547, -0.6246, -0.4101, 0.2499, 0.1570]],
  24137. device='cuda:0', grad_fn=<AddmmBackward>)
  24138. landmarks are: tensor([[[ 0.5805, -0.3818, 1.0282, -1.1774, -0.4903, -0.9310, 0.5894,
  24139. 0.3700],
  24140. [ 0.6372, -0.4129, 1.8942, -0.0765, -0.6173, -0.4768, 0.6999,
  24141. 0.0325],
  24142. [ 0.5759, -0.3948, 1.2072, -0.9849, -0.4672, -0.8156, 0.4912,
  24143. 0.5701],
  24144. [ 0.5902, -0.3493, 1.7961, -0.0072, -0.5942, -0.5615, 0.3180,
  24145. 0.3161],
  24146. [ 0.5446, -0.4280, 1.7499, 0.0543, 0.0156, 0.1301, 0.1918,
  24147. 0.0532],
  24148. [ 0.5432, -0.4462, 1.7557, -0.0380, -0.3113, -0.0765, 0.2141,
  24149. 0.3546],
  24150. [-2.2859, -2.2859, 0.9175, -1.3947, -0.3691, -1.2467, 0.2314,
  24151. 0.3238],
  24152. [ 0.5443, -0.3994, 1.7499, -0.1227, -0.6173, -0.5769, 0.0049,
  24153. 0.2071]]], device='cuda:0')
  24154. loss_train_step before backward: tensor(0.0233, device='cuda:0', grad_fn=<MseLossBackward>)
  24155. loss_train_step after backward: tensor(0.0233, device='cuda:0', grad_fn=<MseLossBackward>)
  24156. loss_train: 0.823372888378799
  24157. step: 26
  24158. running loss: 0.03166818801456919
  24159.  
  24160. Train Steps: 26/90 Loss: 0.0317 torch.Size([8, 600, 800])
  24161. torch.Size([8, 8])
  24162. tensor([[0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317],
  24163. [0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804],
  24164. [0.6259, 0.4133, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947],
  24165. [0.6127, 0.4118, 0.8650, 0.5083, 0.4088, 0.5367, 0.5300, 0.5456],
  24166. [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413],
  24167. [ nan, nan, 0.6793, 0.2110, 0.4012, 0.2167, 0.5112, 0.5583],
  24168. [0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334],
  24169. [0.6118, 0.4052, 0.8463, 0.3917, 0.3538, 0.3450, 0.5053, 0.5593]],
  24170. device='cuda:0', dtype=torch.float64)
  24171. predictions are: tensor([[ 7.5652e-01, -2.7539e-01, 1.5621e+00, 1.8410e-01, -5.1502e-01,
  24172. -3.3857e-01, 3.6783e-01, 2.5592e-01],
  24173. [ 7.1403e-01, -3.2540e-01, 1.6967e+00, -5.6145e-01, -5.9954e-01,
  24174. -3.0913e-01, 7.2090e-01, 7.3376e-02],
  24175. [ 8.6360e-01, -2.0367e-01, 1.3451e+00, -1.1514e+00, -1.1073e-02,
  24176. -1.2515e+00, 5.7336e-01, 1.6534e-01],
  24177. [ 7.4604e-01, -2.7205e-01, 1.5418e+00, 1.2414e-01, -3.3964e-01,
  24178. 4.4680e-01, 3.2297e-01, 3.0321e-01],
  24179. [ 5.8184e-01, -4.4765e-01, 1.8961e+00, -3.1615e-01, -2.8393e-01,
  24180. -6.9362e-01, 1.0925e+00, 1.8052e-01],
  24181. [-1.9204e+00, -2.0319e+00, 8.3474e-01, -1.2677e+00, -4.0795e-01,
  24182. -1.1218e+00, 1.0736e-01, 1.9626e-01],
  24183. [ 5.2281e-01, -4.1657e-01, 1.1166e+00, -1.1776e+00, -5.2418e-01,
  24184. -6.1399e-01, 5.6810e-01, 2.8002e-01],
  24185. [ 7.1040e-01, -2.8060e-01, 1.4571e+00, -4.6101e-01, -5.9290e-01,
  24186. -4.6358e-01, -8.6606e-04, 2.7863e-01]], device='cuda:0',
  24187. grad_fn=<AddmmBackward>)
  24188. landmarks are: tensor([[[ 5.9013e-01, -4.1570e-01, 1.7557e+00, 1.9292e-01, -5.4226e-01,
  24189. -5.9230e-01, 3.5843e-01, 1.6982e-01],
  24190. [ 6.0710e-01, -4.1186e-01, 1.7788e+00, -5.1532e-01, -6.0000e-01,
  24191. -5.6921e-01, 6.5857e-01, -6.7050e-02],
  24192. [ 6.0514e-01, -3.7714e-01, 1.5016e+00, -1.2159e+00, 3.5104e-02,
  24193. -1.5777e+00, 6.0099e-01, -9.2270e-04],
  24194. [ 5.4428e-01, -3.8399e-01, 1.7095e+00, 6.2048e-02, -3.9792e-01,
  24195. 1.9292e-01, 1.6218e-01, 2.3412e-01],
  24196. [ 6.2236e-01, -4.3453e-01, 1.9404e+00, -2.9207e-01, -3.1709e-01,
  24197. -8.7714e-01, 1.0655e+00, 2.1421e-01],
  24198. [-2.2859e+00, -2.2859e+00, 8.5162e-01, -1.3112e+00, -4.3256e-01,
  24199. -1.2851e+00, 7.5520e-02, 2.9299e-01],
  24200. [ 5.7296e-01, -4.5619e-01, 1.2195e+00, -1.2440e+00, -5.4966e-01,
  24201. -7.7109e-01, 5.7045e-01, 1.7788e-01],
  24202. [ 5.3990e-01, -4.1424e-01, 1.6229e+00, -4.7683e-01, -6.5196e-01,
  24203. -6.9238e-01, 4.8058e-02, 2.9724e-01]]], device='cuda:0')
  24204. loss_train_step before backward: tensor(0.0186, device='cuda:0', grad_fn=<MseLossBackward>)
  24205. loss_train_step after backward: tensor(0.0186, device='cuda:0', grad_fn=<MseLossBackward>)
  24206. loss_train: 0.8419636627659202
  24207. step: 27
  24208. running loss: 0.031183839361700747
  24209. Train Steps: 27/90 Loss: 0.0312 torch.Size([8, 600, 800])
  24210. torch.Size([8, 8])
  24211. tensor([[0.6097, 0.3988, 0.8650, 0.5250, 0.4213, 0.5200, 0.5675, 0.5050],
  24212. [0.6223, 0.4028, 0.8988, 0.4200, 0.3763, 0.5733, 0.6375, 0.5167],
  24213. [0.6256, 0.4199, 0.8638, 0.5800, 0.3987, 0.4383, 0.5600, 0.5950],
  24214. [0.6284, 0.4127, 0.8538, 0.5867, 0.4363, 0.5083, 0.6038, 0.5433],
  24215. [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450],
  24216. [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274],
  24217. [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617],
  24218. [0.6091, 0.3997, 0.8314, 0.4334, 0.3788, 0.4550, 0.5213, 0.5656]],
  24219. device='cuda:0', dtype=torch.float64)
  24220. predictions are: tensor([[ 0.6225, -0.3881, 1.5662, -0.0781, -0.2772, 0.1189, 0.2225, 0.0775],
  24221. [ 0.9844, -0.1379, 1.7062, -0.6037, -0.4731, 0.3303, 0.7703, 0.1002],
  24222. [ 0.7208, -0.3073, 1.5178, 0.1953, -0.3616, -0.4154, 0.4075, 0.4878],
  24223. [ 0.6547, -0.3556, 1.4913, 0.1435, -0.2581, -0.0428, 0.4009, 0.1827],
  24224. [ 0.3554, -0.5601, 1.6440, -0.4274, -0.5485, -0.4182, 0.5138, 0.2277],
  24225. [ 0.4400, -0.5040, 1.6787, -0.3374, -0.5218, -0.4229, 0.8214, 0.1786],
  24226. [ 0.3061, -0.5993, 1.7038, -0.4087, -0.4780, -0.3626, 0.3291, 0.2741],
  24227. [ 0.5325, -0.4226, 1.4472, -0.5124, -0.4933, -0.2613, 0.2442, 0.2672]],
  24228. device='cuda:0', grad_fn=<AddmmBackward>)
  24229. landmarks are: tensor([[[ 0.5303, -0.4440, 1.7095, 0.1390, -0.3402, 0.1159, 0.3353,
  24230. 0.0467],
  24231. [ 0.5886, -0.4253, 1.8654, -0.3460, -0.5480, 0.3623, 0.6587,
  24232. 0.1005],
  24233. [ 0.6038, -0.3464, 1.7037, 0.3931, -0.4441, -0.2613, 0.3007,
  24234. 0.4624],
  24235. [ 0.6166, -0.3795, 1.6575, 0.4239, -0.2709, 0.0620, 0.5028,
  24236. 0.2237],
  24237. [ 0.5785, -0.4252, 1.7676, -0.1602, -0.5845, -0.3446, 0.4566,
  24238. 0.2314],
  24239. [ 0.6108, -0.4008, 1.9088, -0.0253, -0.5769, -0.3075, 0.7905,
  24240. 0.1499],
  24241. [ 0.5488, -0.4221, 1.8018, -0.2459, -0.4845, -0.3075, 0.2309,
  24242. 0.3087],
  24243. [ 0.5275, -0.4396, 1.5543, -0.2841, -0.5365, -0.1843, 0.1221,
  24244. 0.3265]]], device='cuda:0')
  24245. loss_train_step before backward: tensor(0.0205, device='cuda:0', grad_fn=<MseLossBackward>)
  24246. loss_train_step after backward: tensor(0.0205, device='cuda:0', grad_fn=<MseLossBackward>)
  24247. loss_train: 0.8624861231073737
  24248. step: 28
  24249. running loss: 0.030803075825263346
  24250. Train Steps: 28/90 Loss: 0.0308 torch.Size([8, 600, 800])
  24251. torch.Size([8, 8])
  24252. tensor([[0.6173, 0.4114, 0.7325, 0.2500, 0.4213, 0.1917, 0.5338, 0.5700],
  24253. [0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133],
  24254. [0.6265, 0.4088, 0.8025, 0.1850, 0.4163, 0.2500, 0.6290, 0.4947],
  24255. [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533],
  24256. [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200],
  24257. [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817],
  24258. [0.6183, 0.4076, 0.8838, 0.4517, 0.3813, 0.4483, 0.5775, 0.5633],
  24259. [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5963, 0.6217]],
  24260. device='cuda:0', dtype=torch.float64)
  24261. predictions are: tensor([[ 0.1548, -0.6666, 1.0475, -1.1813, -0.3556, -1.3011, 0.2236, 0.2432],
  24262. [ 0.7615, -0.3281, 1.7192, 0.2027, -0.5713, -0.0331, 0.7117, 0.0804],
  24263. [ 0.5163, -0.4591, 1.4358, -1.4274, -0.3775, -1.1807, 0.6925, 0.0060],
  24264. [ 0.6276, -0.3860, 1.7572, -0.2639, -0.3414, 0.2167, 0.4277, 0.1755],
  24265. [ 0.3829, -0.5351, 1.7428, -0.1622, -0.5477, -0.1566, 0.4544, 0.1677],
  24266. [ 0.5205, -0.4734, 1.5797, 0.3457, -0.2694, 0.0125, 0.0688, 0.0827],
  24267. [ 0.3644, -0.5406, 1.6827, -0.2990, -0.5216, -0.1021, 0.4127, 0.3489],
  24268. [ 0.4920, -0.4392, 1.0327, -1.2483, -0.3211, -1.2385, 0.4642, 0.4751]],
  24269. device='cuda:0', grad_fn=<AddmmBackward>)
  24270. landmarks are: tensor([[[ 5.6518e-01, -3.8584e-01, 1.0975e+00, -1.1312e+00, -3.4018e-01,
  24271. -1.4006e+00, 1.7945e-01, 3.4688e-01],
  24272. [ 6.1484e-01, -4.1301e-01, 1.6864e+00, 1.6982e-01, -5.3072e-01,
  24273. -1.1501e-01, 6.1247e-01, 8.5142e-02],
  24274. [ 6.0785e-01, -3.9761e-01, 1.4208e+00, -1.4314e+00, -3.6328e-01,
  24275. -1.1312e+00, 6.1950e-01, -9.2270e-04],
  24276. [ 5.4990e-01, -4.2249e-01, 1.8018e+00, -2.9207e-01, -3.0554e-01,
  24277. 5.4350e-02, 4.0462e-01, 2.6990e-01],
  24278. [ 5.8655e-01, -3.9731e-01, 1.8423e+00, -6.8822e-02, -5.1917e-01,
  24279. -2.3048e-01, 4.1617e-01, 1.1594e-01],
  24280. [ 5.1928e-01, -4.6990e-01, 1.5767e+00, 4.0077e-01, -2.4203e-01,
  24281. 7.7444e-02, 1.1776e-01, -6.1038e-02],
  24282. [ 5.7003e-01, -4.0316e-01, 1.7961e+00, -1.9969e-01, -5.2494e-01,
  24283. -2.1509e-01, 3.8152e-01, 3.1609e-01],
  24284. [ 5.6871e-01, -4.0878e-01, 1.0397e+00, -1.1466e+00, -3.1132e-01,
  24285. -1.1928e+00, 4.6813e-01, 5.8553e-01]]], device='cuda:0')
  24286. loss_train_step before backward: tensor(0.0100, device='cuda:0', grad_fn=<MseLossBackward>)
  24287. loss_train_step after backward: tensor(0.0100, device='cuda:0', grad_fn=<MseLossBackward>)
  24288. loss_train: 0.872529580257833
  24289. step: 29
  24290. running loss: 0.030087226905442518
  24291.  
  24292. Train Steps: 29/90 Loss: 0.0301 torch.Size([8, 600, 800])
  24293. torch.Size([8, 8])
  24294. tensor([[0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650],
  24295. [0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133],
  24296. [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649],
  24297. [0.6300, 0.4133, 0.8538, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413],
  24298. [0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750],
  24299. [0.6115, 0.4005, 0.8838, 0.3867, 0.3763, 0.4700, 0.5800, 0.5550],
  24300. [ nan, nan, 0.8488, 0.2300, 0.5563, 0.2100, 0.7390, 0.5679],
  24301. [0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901]],
  24302. device='cuda:0', dtype=torch.float64)
  24303. predictions are: tensor([[ 0.5629, -0.4008, 1.1536, -0.9999, -0.5741, -0.8104, 0.1382, 0.2250],
  24304. [ 0.6017, -0.3874, 1.4440, -1.0393, -0.1726, -1.3285, 0.5830, 0.1753],
  24305. [ 0.9206, -0.2273, 1.5587, 0.5413, -0.6123, 0.1463, 0.4333, 0.0461],
  24306. [ 0.4372, -0.4923, 1.6003, -1.1577, 0.0245, -1.0948, 1.0326, 0.1909],
  24307. [ 0.4308, -0.4589, 1.1894, -0.9453, -0.4832, -0.9390, 0.0674, 0.2499],
  24308. [ 0.9319, -0.1536, 1.6204, -0.4210, -0.6550, 0.0903, 0.3071, 0.1912],
  24309. [-1.7346, -1.9065, 1.4397, -1.1808, 0.0347, -1.2153, 0.9088, 0.3061],
  24310. [ 0.4637, -0.4995, 1.7508, -0.1255, -0.5554, -0.0794, 0.1084, 0.0721]],
  24311. device='cuda:0', grad_fn=<AddmmBackward>)
  24312. landmarks are: tensor([[[ 0.5320, -0.4189, 1.3053, -1.0773, -0.5711, -0.9849, 0.2267,
  24313. 0.3237],
  24314. [ 0.6085, -0.4084, 1.5536, -1.1466, -0.0746, -1.4853, 0.6298,
  24315. 0.0851],
  24316. [ 0.6105, -0.4360, 1.6171, 0.5162, -0.5076, 0.0159, 0.5171,
  24317. -0.1385],
  24318. [ 0.6240, -0.3768, 1.6575, -1.2851, 0.2949, -1.2467, 1.1276,
  24319. 0.2142],
  24320. [ 0.5695, -0.3927, 1.3226, -0.9002, -0.4672, -1.1928, 0.1737,
  24321. 0.3700],
  24322. [ 0.5384, -0.4361, 1.7961, -0.4999, -0.5480, -0.1150, 0.3931,
  24323. 0.2776],
  24324. [-2.2859, -2.2859, 1.6344, -1.2236, 0.2834, -1.3159, 1.1276,
  24325. 0.3371],
  24326. [ 0.5663, -0.3955, 1.7788, -0.2382, -0.4037, -0.2690, 0.0828,
  24327. -0.0220]]], device='cuda:0')
  24328. loss_train_step before backward: tensor(0.0269, device='cuda:0', grad_fn=<MseLossBackward>)
  24329. loss_train_step after backward: tensor(0.0269, device='cuda:0', grad_fn=<MseLossBackward>)
  24330. loss_train: 0.8993881745263934
  24331. step: 30
  24332. running loss: 0.029979605817546447
  24333. Train Steps: 30/90 Loss: 0.0300 torch.Size([8, 600, 800])
  24334. torch.Size([8, 8])
  24335. tensor([[0.6259, 0.4156, 0.8812, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960],
  24336. [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600],
  24337. [0.6135, 0.4115, 0.8838, 0.4667, 0.4288, 0.6050, 0.5778, 0.5097],
  24338. [0.6162, 0.4134, 0.6700, 0.2467, 0.3962, 0.2533, 0.5737, 0.5467],
  24339. [0.6076, 0.3953, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954],
  24340. [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378],
  24341. [0.6075, 0.4007, 0.8275, 0.4917, 0.4050, 0.5100, 0.5167, 0.5280],
  24342. [0.6132, 0.4118, 0.8200, 0.3633, 0.3563, 0.5400, 0.5787, 0.5136]],
  24343. device='cuda:0', dtype=torch.float64)
  24344. predictions are: tensor([[ 0.5029, -0.4627, 1.8340, -1.0026, -0.1339, -1.6149, 0.6156, 0.0087],
  24345. [ 0.5740, -0.4009, 1.7582, -0.1699, -0.1234, -0.0320, 0.4125, 0.2843],
  24346. [ 0.6356, -0.3398, 1.7359, -0.1383, -0.2639, 0.2198, 0.3052, 0.1666],
  24347. [ 0.3308, -0.5629, 0.8794, -1.2483, -0.5392, -1.3448, 0.4178, 0.3191],
  24348. [ 0.4581, -0.5044, 1.5836, -0.5473, -0.7216, -0.5668, 0.2383, 0.0167],
  24349. [ 0.6568, -0.3957, 1.7078, 0.0030, -0.3137, 0.0374, 0.8280, 0.2264],
  24350. [ 0.3297, -0.5640, 1.5501, -0.0372, -0.3852, -0.0481, 0.0236, 0.2139],
  24351. [ 0.3091, -0.5670, 1.4527, -0.6382, -0.6001, -0.0085, 0.4038, 0.2477]],
  24352. device='cuda:0', grad_fn=<AddmmBackward>)
  24353. landmarks are: tensor([[[ 0.6052, -0.3663, 1.7845, -0.8156, -0.0804, -1.4237, 0.5866,
  24354. 0.0051],
  24355. [ 0.5652, -0.4325, 1.7388, -0.1429, -0.0688, 0.3469, 0.4970,
  24356. 0.3007],
  24357. [ 0.5477, -0.3851, 1.7961, -0.1304, -0.3055, 0.5085, 0.3830,
  24358. 0.0682],
  24359. [ 0.5603, -0.3764, 0.8088, -1.1466, -0.4557, -1.1158, 0.3642,
  24360. 0.2391],
  24361. [ 0.5206, -0.4603, 1.4670, -0.4768, -0.6577, -0.3998, 0.1836,
  24362. 0.0021],
  24363. [ 0.6141, -0.4345, 1.6864, -0.0303, -0.2882, 0.1544, 0.9521,
  24364. 0.1982],
  24365. [ 0.5200, -0.4353, 1.5363, -0.0149, -0.4152, 0.0697, 0.1005,
  24366. 0.1530],
  24367. [ 0.5466, -0.3840, 1.5016, -0.6077, -0.6404, 0.2083, 0.3871,
  24368. 0.0862]]], device='cuda:0')
  24369. loss_train_step before backward: tensor(0.0138, device='cuda:0', grad_fn=<MseLossBackward>)
  24370. loss_train_step after backward: tensor(0.0138, device='cuda:0', grad_fn=<MseLossBackward>)
  24371. loss_train: 0.9131455309689045
  24372. step: 31
  24373. running loss: 0.02945630745060982
  24374. Train Steps: 31/90 Loss: 0.0295 torch.Size([8, 600, 800])
  24375. torch.Size([8, 8])
  24376. tensor([[0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892],
  24377. [0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5988, 0.5700],
  24378. [0.6233, 0.4091, 0.8100, 0.2950, 0.3563, 0.3883, 0.6013, 0.5200],
  24379. [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297],
  24380. [ nan, nan, 0.7612, 0.3250, 0.4037, 0.2533, 0.5438, 0.5767],
  24381. [0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869],
  24382. [0.6185, 0.4042, 0.7700, 0.2250, 0.4062, 0.2117, 0.5763, 0.5150],
  24383. [0.6263, 0.4030, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771]],
  24384. device='cuda:0', dtype=torch.float64)
  24385. predictions are: tensor([[ 6.5724e-01, -3.3551e-01, 1.7339e+00, -1.7214e-01, -5.4087e-01,
  24386. -5.7657e-01, -7.2407e-04, 3.0961e-01],
  24387. [ 6.4453e-01, -3.8119e-01, 1.4042e+00, -9.6262e-01, -4.7842e-01,
  24388. -6.2890e-01, 3.6548e-01, 2.1259e-01],
  24389. [ 5.3622e-01, -4.3840e-01, 1.4795e+00, -8.7511e-01, -5.5252e-01,
  24390. -4.6950e-01, 5.1470e-01, 2.1833e-01],
  24391. [ 7.7063e-01, -3.1686e-01, 1.7194e+00, 3.5740e-01, -4.2742e-01,
  24392. -1.5771e-02, 7.1987e-01, 9.3308e-02],
  24393. [-9.5819e-01, -1.3890e+00, 1.2376e+00, -9.8477e-01, -3.6917e-01,
  24394. -1.1233e+00, 1.0399e-01, 3.1658e-01],
  24395. [ 2.9942e-01, -5.8920e-01, 1.5002e+00, -1.1685e+00, -1.3334e-01,
  24396. -1.4075e+00, 5.3227e-01, 5.6465e-02],
  24397. [ 5.8947e-01, -4.2299e-01, 1.2521e+00, -1.2410e+00, -3.8749e-01,
  24398. -1.2859e+00, 3.7489e-01, 5.9306e-02],
  24399. [ 6.0614e-01, -4.2630e-01, 1.8852e+00, -1.1434e-01, -4.4070e-01,
  24400. 1.6359e-01, 5.1914e-01, 5.1127e-02]], device='cuda:0',
  24401. grad_fn=<AddmmBackward>)
  24402. landmarks are: tensor([[[ 5.4544e-01, -4.0531e-01, 1.6633e+00, -1.7660e-01, -6.0577e-01,
  24403. -5.9230e-01, 1.5773e-01, 4.3570e-01],
  24404. [ 5.9601e-01, -3.8879e-01, 1.4840e+00, -1.0095e+00, -6.1155e-01,
  24405. -6.2309e-01, 4.7968e-01, 3.4688e-01],
  24406. [ 5.9319e-01, -3.9615e-01, 1.4554e+00, -9.2333e-01, -6.4042e-01,
  24407. -4.9222e-01, 4.9122e-01, 1.1594e-01],
  24408. [ 6.5036e-01, -3.7434e-01, 1.6171e+00, 4.3156e-01, -5.4226e-01,
  24409. 4.6189e-04, 7.1085e-01, 1.6077e-01],
  24410. [-2.2859e+00, -2.2859e+00, 1.2303e+00, -7.8476e-01, -4.2102e-01,
  24411. -1.1158e+00, 2.2564e-01, 3.7768e-01],
  24412. [ 6.0577e-01, -3.8922e-01, 1.4208e+00, -1.0927e+00, -1.8430e-01,
  24413. -1.4237e+00, 6.1538e-01, -3.6992e-02],
  24414. [ 5.7067e-01, -4.1886e-01, 1.2707e+00, -1.2467e+00, -4.0947e-01,
  24415. -1.3082e+00, 3.7575e-01, 9.2841e-02],
  24416. [ 6.0716e-01, -4.2471e-01, 1.8711e+00, -8.4219e-02, -5.3072e-01,
  24417. 1.0054e-01, 6.7707e-01, -8.2079e-02]]], device='cuda:0')
  24418. loss_train_step before backward: tensor(0.0478, device='cuda:0', grad_fn=<MseLossBackward>)
  24419. loss_train_step after backward: tensor(0.0478, device='cuda:0', grad_fn=<MseLossBackward>)
  24420. loss_train: 0.9609342403709888
  24421. step: 32
  24422. running loss: 0.0300291950115934
  24423. Train Steps: 32/90 Loss: 0.0300 torch.Size([8, 600, 800])
  24424. torch.Size([8, 8])
  24425. tensor([[0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100],
  24426. [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600],
  24427. [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350],
  24428. [0.6263, 0.4030, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771],
  24429. [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683],
  24430. [0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166],
  24431. [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773],
  24432. [0.6286, 0.4055, 0.9000, 0.4717, 0.3763, 0.4683, 0.7018, 0.5494]],
  24433. device='cuda:0', dtype=torch.float64)
  24434. predictions are: tensor([[ 0.5655, -0.4276, 1.7582, -0.6235, -0.6276, -0.6743, 0.5222, 0.1137],
  24435. [ 0.6222, -0.3659, 1.6961, 0.1163, -0.4128, -0.2092, 0.3120, 0.3234],
  24436. [ 0.3312, -0.5525, 1.7061, -0.5412, -0.6122, -0.6128, 0.1507, 0.2349],
  24437. [ 0.4998, -0.4658, 1.7805, -0.3142, -0.5496, -0.0170, 0.4519, 0.0428],
  24438. [ 0.2381, -0.6250, 1.6805, -0.2034, -0.1010, -0.1950, -0.0042, 0.2027],
  24439. [ 0.6927, -0.3407, 1.6134, -0.0583, -0.2813, 0.0023, 0.7085, 0.1154],
  24440. [ 0.3413, -0.5539, 1.7100, -0.2173, -0.0598, -0.2415, -0.0140, 0.1867],
  24441. [ 0.9364, -0.1751, 1.8455, -0.3379, -0.5334, -0.3309, 0.9051, 0.1270]],
  24442. device='cuda:0', grad_fn=<AddmmBackward>)
  24443. landmarks are: tensor([[[ 0.6220, -0.4436, 1.8711, -0.3691, -0.6173, -0.4999, 0.6702,
  24444. 0.0697],
  24445. [ 0.5883, -0.3594, 1.7557, 0.2545, -0.4152, -0.0611, 0.3353,
  24446. 0.3007],
  24447. [ 0.6072, -0.3239, 1.8423, -0.3537, -0.4961, -0.3921, 0.2083,
  24448. 0.1852],
  24449. [ 0.6072, -0.4247, 1.8711, -0.0842, -0.5307, 0.1005, 0.6771,
  24450. -0.0821],
  24451. [ 0.5359, -0.4370, 1.7095, -0.0303, -0.0804, -0.0380, 0.1044,
  24452. 0.3392],
  24453. [ 0.6042, -0.4273, 1.6920, 0.1860, -0.2717, 0.1406, 0.7997,
  24454. 0.1004],
  24455. [ 0.5428, -0.4244, 1.7095, -0.0380, 0.0232, 0.0725, 0.0866,
  24456. 0.3806],
  24457. [ 0.6174, -0.4129, 1.8711, -0.1073, -0.5480, -0.1227, 0.9558,
  24458. 0.2516]]], device='cuda:0')
  24459. loss_train_step before backward: tensor(0.0222, device='cuda:0', grad_fn=<MseLossBackward>)
  24460. loss_train_step after backward: tensor(0.0222, device='cuda:0', grad_fn=<MseLossBackward>)
  24461. loss_train: 0.9831647351384163
  24462. step: 33
  24463. running loss: 0.02979287076177019
  24464.  
  24465. Train Steps: 33/90 Loss: 0.0298 torch.Size([8, 600, 800])
  24466. torch.Size([8, 8])
  24467. tensor([[0.6200, 0.4112, 0.8862, 0.4100, 0.3638, 0.4917, 0.6088, 0.6050],
  24468. [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679],
  24469. [0.6204, 0.4007, 0.7838, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167],
  24470. [0.6048, 0.3928, 0.8538, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719],
  24471. [0.6268, 0.4102, 0.8938, 0.3667, 0.4025, 0.2833, 0.6275, 0.5183],
  24472. [0.6246, 0.4028, 0.8738, 0.4867, 0.4088, 0.5667, 0.6362, 0.5200],
  24473. [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5363, 0.5550],
  24474. [0.6075, 0.4000, 0.8513, 0.5183, 0.4510, 0.5329, 0.5180, 0.5280]],
  24475. device='cuda:0', dtype=torch.float64)
  24476. predictions are: tensor([[ 0.5627, -0.4001, 1.8025, -0.3814, -0.5554, -0.0953, 0.4415, 0.3466],
  24477. [ 0.5360, -0.4458, 1.7127, 0.1136, -0.3603, 0.2607, 0.8678, 0.2304],
  24478. [ 0.6083, -0.3793, 1.3611, -1.2816, -0.2590, -1.5712, 0.4258, 0.0313],
  24479. [ 0.1714, -0.6622, 1.6124, 0.1800, -0.4743, -0.0086, 0.1030, 0.1615],
  24480. [ 0.3095, -0.5768, 1.9215, -0.7524, -0.3930, -1.1578, 0.5276, 0.1364],
  24481. [ 0.6881, -0.3396, 1.8548, -0.0897, -0.3718, 0.2349, 0.6835, 0.0806],
  24482. [ 0.4803, -0.4402, 1.0330, -1.1984, -0.4795, -1.2689, 0.0921, 0.2523],
  24483. [ 0.4115, -0.4979, 1.7126, 0.0244, -0.1545, 0.0139, -0.0112, 0.0896]],
  24484. device='cuda:0', grad_fn=<AddmmBackward>)
  24485. landmarks are: tensor([[[ 0.5777, -0.3868, 1.8076, -0.3921, -0.6058, -0.0149, 0.5259,
  24486. 0.5085],
  24487. [ 0.6257, -0.4249, 1.5998, 0.1236, -0.3806, 0.3084, 0.9887,
  24488. 0.3371],
  24489. [ 0.5796, -0.4351, 1.3342, -1.3159, -0.2189, -1.4853, 0.4046,
  24490. 0.1005],
  24491. [ 0.5079, -0.4714, 1.6575, 0.2237, -0.4961, 0.0774, 0.1465,
  24492. -0.1061],
  24493. [ 0.6092, -0.3913, 1.8423, -0.5923, -0.4268, -0.9772, 0.6125,
  24494. 0.1082],
  24495. [ 0.5990, -0.4256, 1.7499, -0.0380, -0.3979, 0.3315, 0.6529,
  24496. 0.1159],
  24497. [ 0.5448, -0.3859, 0.9242, -1.1466, -0.4152, -1.3005, 0.1910,
  24498. 0.2776],
  24499. [ 0.5202, -0.4382, 1.6460, 0.1082, -0.2029, 0.1754, 0.1067,
  24500. 0.1530]]], device='cuda:0')
  24501. loss_train_step before backward: tensor(0.0109, device='cuda:0', grad_fn=<MseLossBackward>)
  24502. loss_train_step after backward: tensor(0.0109, device='cuda:0', grad_fn=<MseLossBackward>)
  24503. loss_train: 0.994059644639492
  24504. step: 34
  24505. running loss: 0.029237048371749765
  24506. Train Steps: 34/90 Loss: 0.0292 torch.Size([8, 600, 800])
  24507. torch.Size([8, 8])
  24508. tensor([[0.6332, 0.4118, 0.9238, 0.4267, 0.4012, 0.4733, 0.7525, 0.5436],
  24509. [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459],
  24510. [0.6201, 0.4102, 0.7288, 0.2417, 0.4150, 0.2383, 0.6100, 0.5500],
  24511. [0.6361, 0.4076, 0.8862, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297],
  24512. [0.6197, 0.4090, 0.7825, 0.2500, 0.4200, 0.2483, 0.5988, 0.5667],
  24513. [0.6293, 0.4024, 0.8750, 0.5000, 0.4012, 0.5733, 0.7121, 0.5633],
  24514. [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366],
  24515. [0.6325, 0.4165, 0.9000, 0.4617, 0.3813, 0.4900, 0.7485, 0.5447]],
  24516. device='cuda:0', dtype=torch.float64)
  24517. predictions are: tensor([[ 0.5234, -0.4664, 1.8526, -0.3534, -0.4182, -0.0822, 0.6591, 0.2142],
  24518. [ 0.4352, -0.4940, 1.8282, -0.9907, -0.0305, -1.4398, 0.5311, 0.0966],
  24519. [ 0.8282, -0.2346, 1.2444, -1.2653, -0.4695, -1.2162, 0.1952, 0.1267],
  24520. [ 0.4458, -0.5179, 1.8325, 0.1580, -0.5393, -0.1276, 0.4210, 0.1183],
  24521. [ 0.3377, -0.5294, 1.3695, -1.2153, -0.4092, -1.1784, 0.1472, 0.2982],
  24522. [ 0.3720, -0.5473, 1.8003, 0.0639, -0.3505, 0.3625, 0.4583, 0.2098],
  24523. [ 0.4163, -0.5423, 1.5022, 0.1460, -0.4588, -0.0288, 0.5502, 0.1593],
  24524. [ 0.4728, -0.4942, 1.9267, -0.1118, -0.4895, -0.0021, 0.5690, 0.1152]],
  24525. device='cuda:0', grad_fn=<AddmmBackward>)
  24526. landmarks are: tensor([[[ 0.6388, -0.3840, 1.9808, -0.3152, -0.4326, -0.0996, 1.1897,
  24527. 0.2249],
  24528. [ 0.6520, -0.3912, 1.7095, -0.9079, -0.0284, -1.3621, 0.8096,
  24529. 0.2356],
  24530. [ 0.5784, -0.3913, 1.0801, -1.1697, -0.3691, -1.1851, 0.5316,
  24531. 0.2545],
  24532. [ 0.6520, -0.4032, 1.8076, 0.1852, -0.5711, -0.1381, 0.7876,
  24533. 0.1608],
  24534. [ 0.5764, -0.3969, 1.3284, -1.1312, -0.3460, -1.1389, 0.4797,
  24535. 0.3315],
  24536. [ 0.6207, -0.4273, 1.7557, 0.0236, -0.4326, 0.3623, 1.0033,
  24537. 0.3157],
  24538. [ 0.6289, -0.4345, 1.3794, 0.3679, -0.4845, 0.0390, 0.9265,
  24539. 0.1928],
  24540. [ 0.6355, -0.3623, 1.8711, -0.1535, -0.5249, -0.0226, 1.1715,
  24541. 0.2302]]], device='cuda:0')
  24542. loss_train_step before backward: tensor(0.0352, device='cuda:0', grad_fn=<MseLossBackward>)
  24543. loss_train_step after backward: tensor(0.0352, device='cuda:0', grad_fn=<MseLossBackward>)
  24544. loss_train: 1.0292295143008232
  24545. step: 35
  24546. running loss: 0.029406557551452092
  24547. Train Steps: 35/90 Loss: 0.0294 torch.Size([8, 600, 800])
  24548. torch.Size([8, 8])
  24549. tensor([[0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012],
  24550. [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100],
  24551. [0.6286, 0.4055, 0.9000, 0.4717, 0.3763, 0.4683, 0.7018, 0.5494],
  24552. [0.6161, 0.4024, 0.8838, 0.4583, 0.3688, 0.3733, 0.5311, 0.5344],
  24553. [0.6193, 0.4034, 0.7757, 0.2347, 0.3733, 0.2919, 0.5930, 0.4926],
  24554. [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309],
  24555. [0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767],
  24556. [0.6274, 0.4003, 0.8638, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661]],
  24557. device='cuda:0', dtype=torch.float64)
  24558. predictions are: tensor([[ 0.6034, -0.4226, 1.7620, 0.0535, -0.4684, -0.2611, 0.4431, 0.1771],
  24559. [ 0.3709, -0.5737, 1.6665, 0.0294, -0.3556, -0.0046, 0.5314, 0.1333],
  24560. [ 0.6214, -0.4068, 1.9044, -0.2927, -0.4045, -0.1387, 0.9449, 0.1900],
  24561. [ 0.4604, -0.4612, 1.8500, -0.3535, -0.5022, -0.5301, 0.0546, 0.1342],
  24562. [ 0.7181, -0.2917, 1.3510, -1.4534, -0.4692, -0.9523, 0.3789, 0.0521],
  24563. [ 0.6420, -0.3510, 1.8920, -0.1968, -0.3205, -0.5772, 0.5890, 0.1643],
  24564. [ 0.5143, -0.4038, 1.8072, -0.4124, -0.4635, -0.6016, 0.2431, 0.3779],
  24565. [ 0.3334, -0.5918, 1.6142, 0.2640, -0.4179, -0.0050, 0.4461, 0.1309]],
  24566. device='cuda:0', grad_fn=<AddmmBackward>)
  24567. landmarks are: tensor([[[ 0.6249, -0.4352, 1.8018, 0.2545, -0.6173, -0.1997, 0.6401,
  24568. 0.0291],
  24569. [ 0.6125, -0.4273, 1.6402, 0.2006, -0.4788, -0.0303, 0.6182,
  24570. 0.0697],
  24571. [ 0.6174, -0.4129, 1.8711, -0.1073, -0.5480, -0.1227, 0.9558,
  24572. 0.2516],
  24573. [ 0.5598, -0.4273, 1.7961, -0.1689, -0.5827, -0.5615, 0.1671,
  24574. 0.1824],
  24575. [ 0.5746, -0.4225, 1.2968, -1.2019, -0.5615, -0.9374, 0.4533,
  24576. -0.0108],
  24577. [ 0.6520, -0.3623, 1.8885, 0.0313, -0.5538, -0.5384, 0.6926,
  24578. 0.1661],
  24579. [ 0.6058, -0.3216, 1.8423, -0.2536, -0.5885, -0.6000, 0.3353,
  24580. 0.3777],
  24581. [ 0.6120, -0.4371, 1.7037, 0.4701, -0.5827, -0.0226, 0.5354,
  24582. -0.1331]]], device='cuda:0')
  24583. loss_train_step before backward: tensor(0.0150, device='cuda:0', grad_fn=<MseLossBackward>)
  24584. loss_train_step after backward: tensor(0.0150, device='cuda:0', grad_fn=<MseLossBackward>)
  24585. loss_train: 1.044245726428926
  24586. step: 36
  24587. running loss: 0.029006825734136835
  24588. Train Steps: 36/90 Loss: 0.0290 torch.Size([8, 600, 800])
  24589. torch.Size([8, 8])
  24590. tensor([[0.6325, 0.4165, 0.9000, 0.4617, 0.3813, 0.4900, 0.7485, 0.5447],
  24591. [0.6223, 0.4028, 0.8988, 0.4200, 0.3763, 0.5733, 0.6375, 0.5167],
  24592. [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426],
  24593. [0.6151, 0.4125, 0.8738, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483],
  24594. [0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742],
  24595. [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5463, 0.5800],
  24596. [0.6091, 0.3997, 0.8314, 0.4334, 0.3788, 0.4550, 0.5213, 0.5656],
  24597. [0.6145, 0.4007, 0.8775, 0.4533, 0.4562, 0.5533, 0.6088, 0.5533]],
  24598. device='cuda:0', dtype=torch.float64)
  24599. predictions are: tensor([[ 0.6509, -0.3897, 1.9019, -0.1008, -0.3962, -0.0111, 1.1013, 0.1596],
  24600. [ 0.8432, -0.2382, 1.8771, -0.3779, -0.3931, 0.3015, 0.7711, 0.0823],
  24601. [ 0.4317, -0.4885, 1.8653, -0.3814, -0.5239, -0.7898, 0.4372, 0.1703],
  24602. [ 0.5510, -0.3753, 1.7102, -0.2932, -0.5679, -0.6029, 0.1613, 0.2940],
  24603. [ 0.6017, -0.4021, 1.8899, 0.1966, -0.5129, -0.6484, 0.6649, -0.0689],
  24604. [ 0.0397, -0.6923, 1.1594, -0.9537, -0.5354, -1.0902, 0.1489, 0.3543],
  24605. [ 0.4427, -0.4832, 1.5935, -0.3385, -0.4276, -0.2279, 0.1961, 0.2921],
  24606. [ 0.6207, -0.3919, 1.8145, -0.1962, -0.0353, 0.1292, 0.5811, 0.2452]],
  24607. device='cuda:0', grad_fn=<AddmmBackward>)
  24608. landmarks are: tensor([[[ 0.6355, -0.3623, 1.8711, -0.1535, -0.5249, -0.0226, 1.1715,
  24609. 0.2302],
  24610. [ 0.5886, -0.4253, 1.8654, -0.3460, -0.5480, 0.3623, 0.6587,
  24611. 0.1005],
  24612. [ 0.5788, -0.4135, 1.8214, -0.2408, -0.6039, -0.7854, 0.4115,
  24613. 0.2203],
  24614. [ 0.5551, -0.3807, 1.7499, -0.2459, -0.6346, -0.5384, 0.0871,
  24615. 0.2468],
  24616. [ 0.6113, -0.4057, 1.7499, 0.3007, -0.5885, -0.5384, 0.4513,
  24617. -0.0957],
  24618. [ 0.5548, -0.3936, 1.1634, -0.8105, -0.5192, -1.0696, 0.2372,
  24619. 0.3931],
  24620. [ 0.5275, -0.4396, 1.5543, -0.2841, -0.5365, -0.1843, 0.1221,
  24621. 0.3265],
  24622. [ 0.5526, -0.4351, 1.7672, -0.1920, -0.1785, 0.2699, 0.5259,
  24623. 0.2699]]], device='cuda:0')
  24624. loss_train_step before backward: tensor(0.0128, device='cuda:0', grad_fn=<MseLossBackward>)
  24625. loss_train_step after backward: tensor(0.0128, device='cuda:0', grad_fn=<MseLossBackward>)
  24626. loss_train: 1.0570316752418876
  24627. step: 37
  24628. running loss: 0.02856842365518615
  24629.  
  24630. Train Steps: 37/90 Loss: 0.0286 torch.Size([8, 600, 800])
  24631. torch.Size([8, 8])
  24632. tensor([[0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633],
  24633. [0.6109, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567],
  24634. [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413],
  24635. [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378],
  24636. [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483],
  24637. [0.6248, 0.4185, 0.8500, 0.5767, 0.4463, 0.4550, 0.5613, 0.5917],
  24638. [0.6336, 0.4154, 0.8900, 0.2767, 0.4988, 0.2867, 0.7422, 0.5540],
  24639. [0.6208, 0.4082, 0.8538, 0.3067, 0.3588, 0.3717, 0.6112, 0.5517]],
  24640. device='cuda:0', dtype=torch.float64)
  24641. predictions are: tensor([[ 0.4441, -0.4830, 1.6680, -1.0800, -0.2465, -1.0681, 0.7556, 0.2798],
  24642. [ 0.1950, -0.6115, 1.0745, -1.2398, -0.6409, -0.9222, 0.2341, 0.2288],
  24643. [ 0.6605, -0.4018, 2.1028, -0.1091, -0.3462, -0.7841, 0.9333, 0.0228],
  24644. [ 0.7723, -0.3271, 1.7699, 0.2168, -0.3638, 0.3681, 0.7759, 0.0680],
  24645. [ 0.7062, -0.3003, 1.1744, -1.1470, -0.4890, -0.9774, 0.3316, 0.2041],
  24646. [ 0.5624, -0.4178, 1.6652, 0.5979, -0.2604, -0.0412, 0.2334, 0.3466],
  24647. [ 0.0958, -0.7478, 1.8266, -0.7130, -0.0486, -0.9561, 0.8954, 0.2009],
  24648. [ 0.1288, -0.6920, 1.6375, -0.5932, -0.7722, -0.3912, 0.4008, 0.1899]],
  24649. device='cuda:0', grad_fn=<AddmmBackward>)
  24650. landmarks are: tensor([[[ 0.6042, -0.4105, 1.5478, -1.2082, -0.1208, -1.0927, 0.9704,
  24651. 0.3157],
  24652. [ 0.5356, -0.4217, 1.0339, -1.4776, -0.5076, -1.1081, 0.4219,
  24653. 0.2853],
  24654. [ 0.6224, -0.4345, 1.9404, -0.2921, -0.3171, -0.8771, 1.0655,
  24655. 0.2142],
  24656. [ 0.6141, -0.4345, 1.6864, -0.0303, -0.2882, 0.1544, 0.9521,
  24657. 0.1982],
  24658. [ 0.5784, -0.4085, 1.0859, -1.3929, -0.4037, -1.1158, 0.5605,
  24659. 0.2468],
  24660. [ 0.5999, -0.3530, 1.6402, 0.3777, -0.2247, -0.1843, 0.3065,
  24661. 0.4470],
  24662. [ 0.6405, -0.3671, 1.8249, -1.0080, 0.0178, -0.9618, 1.1422,
  24663. 0.2730],
  24664. [ 0.5814, -0.4003, 1.6575, -0.8694, -0.6289, -0.5692, 0.5374,
  24665. 0.2622]]], device='cuda:0')
  24666. loss_train_step before backward: tensor(0.0323, device='cuda:0', grad_fn=<MseLossBackward>)
  24667. loss_train_step after backward: tensor(0.0323, device='cuda:0', grad_fn=<MseLossBackward>)
  24668. loss_train: 1.089333820156753
  24669. step: 38
  24670. running loss: 0.028666679477809293
  24671. Train Steps: 38/90 Loss: 0.0287 torch.Size([8, 600, 800])
  24672. torch.Size([8, 8])
  24673. tensor([[0.6102, 0.3999, 0.8750, 0.5133, 0.3825, 0.4750, 0.5637, 0.5083],
  24674. [0.6218, 0.4185, 0.7338, 0.2650, 0.4625, 0.1950, 0.5687, 0.5800],
  24675. [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118],
  24676. [0.6091, 0.3997, 0.8314, 0.4334, 0.3788, 0.4550, 0.5213, 0.5656],
  24677. [0.6206, 0.4001, 0.8900, 0.3933, 0.3588, 0.3567, 0.5837, 0.5083],
  24678. [0.6087, 0.3951, 0.8387, 0.5833, 0.4188, 0.4933, 0.5146, 0.4830],
  24679. [ nan, nan, 0.8300, 0.3150, 0.3588, 0.3383, 0.5208, 0.5194],
  24680. [0.6079, 0.3964, 0.7420, 0.2958, 0.3563, 0.2917, 0.5351, 0.4980]],
  24681. device='cuda:0', dtype=torch.float64)
  24682. predictions are: tensor([[ 8.8443e-01, -1.9540e-01, 1.8931e+00, 8.7475e-02, -4.4813e-01,
  24683. 1.6257e-03, 6.1814e-01, -9.2434e-03],
  24684. [ 5.0552e-01, -3.8126e-01, 1.2175e+00, -1.0401e+00, -1.1707e-01,
  24685. -1.3590e+00, 5.7183e-01, 4.3613e-01],
  24686. [ 8.3590e-01, -2.1532e-01, 1.6410e+00, 2.7475e-01, -4.1388e-01,
  24687. -1.0416e-01, 7.4442e-01, 4.3882e-01],
  24688. [ 7.6310e-01, -2.6186e-01, 1.6897e+00, -2.8506e-01, -4.8278e-01,
  24689. -1.4364e-01, 4.4101e-01, 2.6590e-01],
  24690. [ 8.4191e-01, -2.4302e-01, 1.9611e+00, -4.4580e-01, -6.0415e-01,
  24691. -6.3684e-01, 6.8879e-01, 9.6068e-03],
  24692. [ 8.2173e-01, -2.3409e-01, 1.7656e+00, 4.0293e-01, -3.2369e-01,
  24693. 5.4397e-02, 5.1337e-01, 2.0862e-02],
  24694. [-2.3548e+00, -2.3337e+00, 1.3308e+00, -7.7752e-01, -6.2998e-01,
  24695. -7.0770e-01, 5.5572e-01, 2.3119e-01],
  24696. [ 8.9218e-01, -1.8210e-01, 1.3559e+00, -9.9058e-01, -5.4404e-01,
  24697. -8.9048e-01, 3.4451e-01, 9.2592e-03]], device='cuda:0',
  24698. grad_fn=<AddmmBackward>)
  24699. landmarks are: tensor([[[ 0.5326, -0.4390, 1.7557, 0.0851, -0.5192, -0.0919, 0.3180,
  24700. 0.0620],
  24701. [ 0.5862, -0.3530, 1.1032, -1.0619, -0.1497, -1.3852, 0.3411,
  24702. 0.3931],
  24703. [ 0.6121, -0.3844, 1.4556, 0.3936, -0.4383, -0.1689, 0.2925,
  24704. 0.5401],
  24705. [ 0.5275, -0.4396, 1.5543, -0.2841, -0.5365, -0.1843, 0.1221,
  24706. 0.3265],
  24707. [ 0.5807, -0.4378, 1.8249, -0.4691, -0.6289, -0.6385, 0.4104,
  24708. 0.0620],
  24709. [ 0.5258, -0.4610, 1.5882, 0.4085, -0.3517, -0.0072, 0.0910,
  24710. -0.0550],
  24711. [-2.2859, -2.2859, 1.5478, -0.8309, -0.6289, -0.7232, 0.1198,
  24712. 0.1133],
  24713. [ 0.5219, -0.4550, 1.1415, -0.9196, -0.6404, -0.9387, 0.1856,
  24714. 0.0141]]], device='cuda:0')
  24715. loss_train_step before backward: tensor(0.0324, device='cuda:0', grad_fn=<MseLossBackward>)
  24716. loss_train_step after backward: tensor(0.0324, device='cuda:0', grad_fn=<MseLossBackward>)
  24717. loss_train: 1.1217780141159892
  24718. step: 39
  24719. running loss: 0.028763538823486902
  24720. Train Steps: 39/90 Loss: 0.0288 torch.Size([8, 600, 800])
  24721. torch.Size([8, 8])
  24722. tensor([[0.6138, 0.4054, 0.8750, 0.4750, 0.4363, 0.5017, 0.5086, 0.5822],
  24723. [0.6346, 0.4086, 0.7938, 0.5500, 0.3962, 0.4867, 0.7343, 0.5702],
  24724. [0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162],
  24725. [0.6126, 0.4039, 0.8237, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138],
  24726. [0.6256, 0.4199, 0.8638, 0.5800, 0.3987, 0.4383, 0.5600, 0.5950],
  24727. [0.6199, 0.4071, 0.7600, 0.2117, 0.4037, 0.2767, 0.6138, 0.5550],
  24728. [0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945],
  24729. [0.6336, 0.4191, 0.8938, 0.5167, 0.3937, 0.3517, 0.7343, 0.5748]],
  24730. device='cuda:0', dtype=torch.float64)
  24731. predictions are: tensor([[ 0.6440, -0.3332, 1.7672, -0.0498, -0.3202, 0.0893, 0.1455, 0.0761],
  24732. [ 0.5725, -0.4177, 1.4867, 0.1828, -0.4769, -0.0204, 1.0719, 0.2339],
  24733. [ 0.5976, -0.4119, 1.8482, -0.3967, -0.5806, -0.0978, 0.7626, 0.1046],
  24734. [ 0.4039, -0.4856, 1.5727, -0.4861, -0.6485, -0.5054, 0.4126, 0.3674],
  24735. [ 0.8189, -0.2302, 1.6996, 0.4369, -0.4249, -0.1943, 0.2815, 0.3349],
  24736. [ 0.6293, -0.3363, 1.3477, -1.3669, -0.4102, -1.0134, 0.5384, 0.2066],
  24737. [ 0.5771, -0.4255, 1.9855, -0.3066, -0.2413, -0.5252, 1.0077, 0.2510],
  24738. [ 0.0591, -0.7700, 1.8177, 0.0866, -0.3968, -0.5449, 0.8709, 0.1763]],
  24739. device='cuda:0', grad_fn=<AddmmBackward>)
  24740. landmarks are: tensor([[[ 0.5491, -0.4132, 1.7557, -0.0919, -0.2709, 0.0313, 0.0635,
  24741. 0.4032],
  24742. [ 0.6454, -0.3984, 1.3804, 0.2545, -0.4557, -0.0380, 1.1057,
  24743. 0.3478],
  24744. [ 0.6059, -0.4442, 1.8921, -0.3422, -0.6230, -0.1461, 0.7489,
  24745. 0.0985],
  24746. [ 0.5438, -0.4206, 1.5189, -0.4537, -0.6115, -0.6231, 0.4365,
  24747. 0.5491],
  24748. [ 0.6038, -0.3464, 1.7037, 0.3931, -0.4441, -0.2613, 0.3007,
  24749. 0.4624],
  24750. [ 0.5775, -0.4054, 1.2245, -1.3082, -0.4210, -1.0080, 0.5490,
  24751. 0.2776],
  24752. [ 0.6471, -0.3719, 1.9866, -0.3921, -0.3806, -0.5538, 1.0070,
  24753. 0.4600],
  24754. [ 0.6405, -0.3503, 1.8423, 0.1005, -0.4672, -0.6616, 1.1057,
  24755. 0.3692]]], device='cuda:0')
  24756. loss_train_step before backward: tensor(0.0169, device='cuda:0', grad_fn=<MseLossBackward>)
  24757. loss_train_step after backward: tensor(0.0169, device='cuda:0', grad_fn=<MseLossBackward>)
  24758. loss_train: 1.1386554995551705
  24759. step: 40
  24760. running loss: 0.028466387488879263
  24761. Train Steps: 40/90 Loss: 0.0285 torch.Size([8, 600, 800])
  24762. torch.Size([8, 8])
  24763. tensor([[0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461],
  24764. [0.6361, 0.4076, 0.8862, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297],
  24765. [0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835],
  24766. [0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058],
  24767. [0.6068, 0.3963, 0.8650, 0.4317, 0.4037, 0.5083, 0.5253, 0.4999],
  24768. [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683],
  24769. [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533],
  24770. [0.6246, 0.4008, 0.8757, 0.5088, 0.4101, 0.5392, 0.6644, 0.5133]],
  24771. device='cuda:0', dtype=torch.float64)
  24772. predictions are: tensor([[ 0.6078, -0.3606, 1.5451, -0.5208, -0.7137, -0.4924, 0.4499, 0.2869],
  24773. [ 0.7464, -0.3133, 1.7544, 0.2531, -0.5698, -0.1807, 1.0396, 0.2625],
  24774. [ 0.3502, -0.5323, 1.8783, -0.4356, -0.3516, -1.3380, 0.7168, 0.1584],
  24775. [ 0.4409, -0.4870, 1.5813, -0.2868, -0.5962, -0.1480, 0.2479, 0.2000],
  24776. [ 0.5728, -0.4096, 1.6914, -0.2556, -0.5003, 0.0082, 0.4210, 0.1230],
  24777. [ 0.3514, -0.5419, 1.7136, 0.0634, -0.1271, -0.0333, 0.4373, 0.2821],
  24778. [ 0.8080, -0.2471, 1.7488, -0.1996, -0.3355, 0.0799, 0.5897, 0.2356],
  24779. [ 0.6162, -0.4141, 1.7027, 0.1096, -0.3821, 0.1523, 1.0530, 0.2057]],
  24780. device='cuda:0', grad_fn=<AddmmBackward>)
  24781. landmarks are: tensor([[[ 0.5124, -0.4483, 1.5420, -0.6231, -0.7155, -0.4152, 0.2679,
  24782. 0.2365],
  24783. [ 0.6520, -0.4032, 1.8076, 0.1852, -0.5711, -0.1381, 0.7876,
  24784. 0.1608],
  24785. [ 0.6117, -0.3877, 1.8249, -0.5230, -0.3517, -1.2851, 0.4200,
  24786. -0.0529],
  24787. [ 0.5245, -0.4361, 1.5940, -0.2921, -0.5480, -0.0919, 0.2432,
  24788. 0.0502],
  24789. [ 0.5168, -0.4556, 1.7095, -0.2921, -0.4210, 0.0620, 0.1404,
  24790. 0.0231],
  24791. [ 0.5359, -0.4370, 1.7095, -0.0303, -0.0804, -0.0380, 0.1044,
  24792. 0.3392],
  24793. [ 0.5499, -0.4225, 1.8018, -0.2921, -0.3055, 0.0543, 0.4046,
  24794. 0.2699],
  24795. [ 0.5993, -0.4345, 1.7587, 0.0641, -0.3917, 0.2048, 0.7827,
  24796. 0.0852]]], device='cuda:0')
  24797. loss_train_step before backward: tensor(0.0150, device='cuda:0', grad_fn=<MseLossBackward>)
  24798. loss_train_step after backward: tensor(0.0150, device='cuda:0', grad_fn=<MseLossBackward>)
  24799. loss_train: 1.1536407247185707
  24800. step: 41
  24801. running loss: 0.028137578651672455
  24802.  
  24803. Train Steps: 41/90 Loss: 0.0281 torch.Size([8, 600, 800])
  24804. torch.Size([8, 8])
  24805. tensor([[0.6110, 0.3984, 0.8750, 0.4933, 0.4625, 0.4950, 0.5578, 0.5676],
  24806. [0.6185, 0.4098, 0.8838, 0.4900, 0.4537, 0.5800, 0.6288, 0.5400],
  24807. [0.6034, 0.4011, 0.7350, 0.2533, 0.3438, 0.3367, 0.5516, 0.5084],
  24808. [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600],
  24809. [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395],
  24810. [0.6293, 0.4024, 0.8750, 0.5000, 0.4012, 0.5733, 0.7121, 0.5633],
  24811. [0.6282, 0.4034, 0.7830, 0.2080, 0.4532, 0.2080, 0.6404, 0.5323],
  24812. [0.6198, 0.4101, 0.8838, 0.5283, 0.3763, 0.5267, 0.5913, 0.5567]],
  24813. device='cuda:0', dtype=torch.float64)
  24814. predictions are: tensor([[ 0.8460, -0.2558, 1.8253, 0.0117, -0.2213, -0.0403, 0.4005, 0.2357],
  24815. [ 0.7772, -0.2915, 1.8138, 0.0177, -0.2384, 0.2574, 0.7008, 0.2388],
  24816. [ 0.4894, -0.4668, 1.1931, -1.0988, -0.6886, -0.7637, 0.4019, 0.1059],
  24817. [ 0.4321, -0.5275, 1.7577, 0.3191, -0.4827, -0.1046, 0.5689, 0.3247],
  24818. [ 0.0895, -0.7132, 1.6919, 0.1463, -0.7607, -0.6208, 0.4622, 0.2670],
  24819. [ 0.5809, -0.4362, 1.8019, 0.0882, -0.4585, 0.2636, 1.0704, 0.2819],
  24820. [ 0.4095, -0.5098, 1.3689, -1.2173, -0.2903, -1.4032, 0.7033, 0.1959],
  24821. [ 0.7698, -0.3002, 1.7510, 0.2490, -0.5625, 0.0499, 0.6418, 0.3161]],
  24822. device='cuda:0', grad_fn=<AddmmBackward>)
  24823. landmarks are: tensor([[[ 5.3637e-01, -4.4573e-01, 1.7557e+00, -7.2363e-03, -1.4965e-01,
  24824. 4.6189e-04, 2.9049e-01, 3.3573e-01],
  24825. [ 5.7090e-01, -3.9330e-01, 1.7961e+00, -2.2633e-02, -1.9007e-01,
  24826. 3.9307e-01, 6.1824e-01, 2.0831e-01],
  24827. [ 5.0092e-01, -4.3333e-01, 1.1090e+00, -1.1158e+00, -6.9815e-01,
  24828. -7.3087e-01, 2.6170e-01, 6.2199e-02],
  24829. [ 5.8834e-01, -3.5935e-01, 1.7557e+00, 2.5450e-01, -4.1524e-01,
  24830. -6.1124e-02, 3.3533e-01, 3.0069e-01],
  24831. [ 5.6966e-01, -4.5138e-01, 1.7420e+00, 2.6720e-01, -6.0553e-01,
  24832. -6.3118e-01, 3.4489e-01, 2.0578e-01],
  24833. [ 6.2072e-01, -4.2731e-01, 1.7557e+00, 2.3557e-02, -4.3256e-01,
  24834. 3.6228e-01, 1.0033e+00, 3.1574e-01],
  24835. [ 6.1577e-01, -4.2249e-01, 1.3307e+00, -1.3253e+00, -1.9244e-01,
  24836. -1.3252e+00, 6.7213e-01, 1.7271e-01],
  24837. [ 5.7696e-01, -3.9176e-01, 1.7961e+00, 1.5443e-01, -5.4804e-01,
  24838. 1.4673e-01, 4.4503e-01, 2.8530e-01]]], device='cuda:0')
  24839. loss_train_step before backward: tensor(0.0151, device='cuda:0', grad_fn=<MseLossBackward>)
  24840. loss_train_step after backward: tensor(0.0151, device='cuda:0', grad_fn=<MseLossBackward>)
  24841. loss_train: 1.168755921535194
  24842. step: 42
  24843. running loss: 0.027827521941314142
  24844. Train Steps: 42/90 Loss: 0.0278 torch.Size([8, 600, 800])
  24845. torch.Size([8, 8])
  24846. tensor([[ nan, nan, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621],
  24847. [0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496],
  24848. [0.6289, 0.4019, 0.8113, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332],
  24849. [0.6259, 0.4133, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947],
  24850. [0.6113, 0.4006, 0.8700, 0.5350, 0.3638, 0.3767, 0.5097, 0.4882],
  24851. [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290],
  24852. [0.6273, 0.4143, 0.8750, 0.5700, 0.3987, 0.4717, 0.6013, 0.5467],
  24853. [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578]],
  24854. device='cuda:0', dtype=torch.float64)
  24855. predictions are: tensor([[-2.3789, -2.3153, 1.4070, -1.0507, -0.0678, -1.0513, 0.8482, 0.3953],
  24856. [ 0.8099, -0.2315, 1.6563, -0.7828, -0.4231, -0.8245, 0.5477, 0.2148],
  24857. [ 0.8783, -0.2303, 1.5719, 0.3158, -0.5748, 0.1391, 0.9184, 0.1658],
  24858. [ 0.7176, -0.2993, 1.4880, -0.9552, -0.0990, -1.3317, 0.5871, 0.2044],
  24859. [ 0.6668, -0.3495, 1.7761, 0.3125, -0.7050, -0.3645, 0.2513, 0.0879],
  24860. [ 0.7174, -0.3138, 1.5611, -0.6533, -0.6790, -0.3239, 0.7629, 0.2343],
  24861. [ 0.6578, -0.3283, 1.6729, 0.4010, -0.5496, 0.0428, 0.5500, 0.2054],
  24862. [ 0.7935, -0.2433, 0.9686, -1.1348, -0.4325, -1.1478, 0.3664, 0.3375]],
  24863. device='cuda:0', grad_fn=<AddmmBackward>)
  24864. landmarks are: tensor([[[-2.2859e+00, -2.2859e+00, 1.7210e+00, -9.7721e-01, 1.8522e-01,
  24865. -1.3698e+00, 7.9859e-01, 3.1039e-01],
  24866. [ 5.7904e-01, -4.0308e-01, 1.6915e+00, -9.5640e-01, -4.1518e-01,
  24867. -1.1063e+00, 4.4251e-01, 2.5281e-01],
  24868. [ 6.1907e-01, -4.2971e-01, 1.4612e+00, 2.3911e-01, -4.9607e-01,
  24869. 3.1255e-02, 1.1166e+00, 1.7680e-01],
  24870. [ 6.0514e-01, -3.7714e-01, 1.5016e+00, -1.2159e+00, 3.5104e-02,
  24871. -1.5777e+00, 6.0099e-01, -9.2270e-04],
  24872. [ 5.3788e-01, -4.3580e-01, 1.7326e+00, 1.8522e-01, -6.0577e-01,
  24873. -5.4611e-01, 6.8408e-02, -3.0981e-02],
  24874. [ 5.6966e-01, -4.5379e-01, 1.5308e+00, -8.7027e-01, -6.5720e-01,
  24875. -3.6388e-01, 5.7392e-01, 1.5759e-01],
  24876. [ 6.1149e-01, -3.7244e-01, 1.7557e+00, 3.4688e-01, -4.4411e-01,
  24877. -1.0731e-01, 4.9122e-01, 2.3911e-01],
  24878. [ 5.2315e-01, -4.1886e-01, 9.3580e-01, -1.4006e+00, -3.9215e-01,
  24879. -1.3698e+00, 2.5553e-01, 2.9064e-01]]], device='cuda:0')
  24880. loss_train_step before backward: tensor(0.0226, device='cuda:0', grad_fn=<MseLossBackward>)
  24881. loss_train_step after backward: tensor(0.0226, device='cuda:0', grad_fn=<MseLossBackward>)
  24882. loss_train: 1.1913132583722472
  24883. step: 43
  24884. running loss: 0.027704959497029006
  24885. Train Steps: 43/90 Loss: 0.0277 torch.Size([8, 600, 800])
  24886. torch.Size([8, 8])
  24887. tensor([[0.6293, 0.4097, 0.8800, 0.2517, 0.5262, 0.2600, 0.7430, 0.5378],
  24888. [0.6263, 0.4233, 0.7924, 0.4626, 0.3788, 0.2883, 0.5573, 0.6047],
  24889. [0.6193, 0.4050, 0.7313, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656],
  24890. [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383],
  24891. [0.6277, 0.4103, 0.8087, 0.5717, 0.4188, 0.4750, 0.5663, 0.6083],
  24892. [0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947],
  24893. [0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019],
  24894. [ nan, nan, 0.7268, 0.2333, 0.4125, 0.1933, 0.5112, 0.5383]],
  24895. device='cuda:0', dtype=torch.float64)
  24896. predictions are: tensor([[ 0.7687, -0.3415, 1.8001, -0.9061, 0.0913, -1.0675, 1.0970, 0.1657],
  24897. [ 0.6264, -0.3643, 1.4945, 0.0356, -0.5496, -0.5861, 0.2921, 0.4837],
  24898. [ 0.4643, -0.4759, 1.1842, -1.0252, -0.4246, -1.0295, 0.3400, 0.2287],
  24899. [ 0.8361, -0.2656, 1.6420, -0.5933, -0.7136, -0.2909, 0.6242, 0.2097],
  24900. [ 0.8740, -0.2414, 1.6095, 0.5303, -0.4926, 0.1541, 0.4730, 0.3609],
  24901. [ 0.6245, -0.4205, 1.4561, -1.2001, -0.3846, -1.1205, 0.6975, 0.0619],
  24902. [ 1.0365, -0.1871, 1.9594, 0.1347, -0.6793, -0.2144, 0.7839, 0.0429],
  24903. [-2.2615, -2.2909, 1.0155, -0.9692, -0.4837, -1.0926, 0.3438, 0.3451]],
  24904. device='cuda:0', grad_fn=<AddmmBackward>)
  24905. landmarks are: tensor([[[ 6.2072e-01, -3.9360e-01, 1.7788e+00, -1.1235e+00, 1.4480e-01,
  24906. -1.0850e+00, 1.1459e+00, 1.9818e-01],
  24907. [ 6.0687e-01, -3.3095e-01, 1.3742e+00, -1.4927e-01, -5.3649e-01,
  24908. -9.5412e-01, 2.8843e-01, 5.0705e-01],
  24909. [ 5.7460e-01, -4.1527e-01, 1.0917e+00, -1.1620e+00, -4.0370e-01,
  24910. -1.3082e+00, 3.2339e-01, 3.2671e-01],
  24911. [ 5.7794e-01, -4.2748e-01, 1.5894e+00, -8.3617e-01, -6.5774e-01,
  24912. -5.1532e-01, 5.6051e-01, 2.0062e-01],
  24913. [ 6.1339e-01, -3.9099e-01, 1.4497e+00, 3.5458e-01, -3.5173e-01,
  24914. -9.1917e-02, 3.2956e-01, 5.2394e-01],
  24915. [ 6.0739e-01, -4.0631e-01, 1.3630e+00, -1.5238e+00, -3.1132e-01,
  24916. -1.2390e+00, 6.1950e-01, -9.2270e-04],
  24917. [ 6.3718e-01, -4.1286e-01, 1.8942e+00, -7.6520e-02, -6.1732e-01,
  24918. -4.7683e-01, 6.9989e-01, 3.2524e-02],
  24919. [-2.2859e+00, -2.2859e+00, 1.0712e+00, -1.2085e+00, -3.8060e-01,
  24920. -1.3929e+00, 7.5520e-02, 2.0062e-01]]], device='cuda:0')
  24921. loss_train_step before backward: tensor(0.0253, device='cuda:0', grad_fn=<MseLossBackward>)
  24922. loss_train_step after backward: tensor(0.0253, device='cuda:0', grad_fn=<MseLossBackward>)
  24923. loss_train: 1.2166265537962317
  24924. step: 44
  24925. running loss: 0.027650603495368905
  24926.  
  24927. Train Steps: 44/90 Loss: 0.0277 torch.Size([8, 600, 800])
  24928. torch.Size([8, 8])
  24929. tensor([[0.6329, 0.4196, 0.9238, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748],
  24930. [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6063, 0.5617],
  24931. [0.6151, 0.4125, 0.8738, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483],
  24932. [0.6262, 0.4163, 0.8850, 0.5183, 0.3763, 0.4150, 0.6025, 0.5500],
  24933. [0.6286, 0.4078, 0.8063, 0.2267, 0.4788, 0.1533, 0.5953, 0.4913],
  24934. [0.6260, 0.4214, 0.8538, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983],
  24935. [0.6199, 0.4102, 0.8950, 0.4417, 0.4012, 0.5367, 0.6112, 0.5967],
  24936. [0.6149, 0.4054, 0.6713, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695]],
  24937. device='cuda:0', dtype=torch.float64)
  24938. predictions are: tensor([[ 0.3303, -0.6281, 1.9273, -0.1490, -0.3484, -0.8306, 1.1438, 0.2584],
  24939. [ 0.4734, -0.4978, 1.7926, -0.0348, -0.6413, -0.0802, 0.6553, 0.2107],
  24940. [ 0.5969, -0.3857, 1.6875, -0.2029, -0.6595, -0.4201, 0.2257, 0.2553],
  24941. [ 0.5582, -0.4341, 1.7493, 0.0950, -0.6047, -0.2364, 0.6304, 0.2153],
  24942. [ 0.9523, -0.1903, 1.3590, -1.0891, -0.0964, -1.3481, 0.5336, 0.0817],
  24943. [ 0.5434, -0.4403, 1.5534, 0.2962, -0.5464, -0.3564, 0.3453, 0.4208],
  24944. [ 0.6677, -0.3801, 1.7239, -0.2702, -0.4015, 0.3340, 0.7229, 0.3309],
  24945. [ 0.0454, -0.7430, 0.8856, -1.2511, -0.3786, -1.2249, 0.1690, 0.3114]],
  24946. device='cuda:0', grad_fn=<AddmmBackward>)
  24947. landmarks are: tensor([[[ 0.6372, -0.3479, 1.9808, -0.1689, -0.3171, -0.9310, 1.1057,
  24948. 0.3692],
  24949. [ 0.5797, -0.3965, 1.8480, -0.0765, -0.6173, -0.1535, 0.5143,
  24950. 0.3084],
  24951. [ 0.5551, -0.3807, 1.7499, -0.2459, -0.6346, -0.5384, 0.0871,
  24952. 0.2468],
  24953. [ 0.6066, -0.3632, 1.8018, 0.1082, -0.5480, -0.3691, 0.4970,
  24954. 0.2545],
  24955. [ 0.6177, -0.4022, 1.4381, -1.2390, -0.0746, -1.5777, 0.4639,
  24956. -0.0168],
  24957. [ 0.6055, -0.3393, 1.6575, 0.2545, -0.5942, -0.5461, 0.2949,
  24958. 0.4778],
  24959. [ 0.5772, -0.3913, 1.8480, -0.2459, -0.4326, 0.1929, 0.5374,
  24960. 0.4701],
  24961. [ 0.5544, -0.4133, 0.8145, -1.2082, -0.4268, -1.3544, 0.1221,
  24962. 0.3446]]], device='cuda:0')
  24963. loss_train_step before backward: tensor(0.0183, device='cuda:0', grad_fn=<MseLossBackward>)
  24964. loss_train_step after backward: tensor(0.0183, device='cuda:0', grad_fn=<MseLossBackward>)
  24965. loss_train: 1.2349761677905917
  24966. step: 45
  24967. running loss: 0.027443914839790928
  24968. Train Steps: 45/90 Loss: 0.0274 torch.Size([8, 600, 800])
  24969. torch.Size([8, 8])
  24970. tensor([[0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938],
  24971. [0.6264, 0.3972, 0.8853, 0.4771, 0.3853, 0.4511, 0.6293, 0.5334],
  24972. [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297],
  24973. [0.6204, 0.4110, 0.7913, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367],
  24974. [0.6223, 0.4028, 0.8988, 0.4200, 0.3763, 0.5733, 0.6375, 0.5167],
  24975. [0.6229, 0.4066, 0.7612, 0.2967, 0.4250, 0.2167, 0.5925, 0.5350],
  24976. [0.6128, 0.4084, 0.8738, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397],
  24977. [ nan, nan, 0.7268, 0.2333, 0.4125, 0.1933, 0.5112, 0.5383]],
  24978. device='cuda:0', dtype=torch.float64)
  24979. predictions are: tensor([[ 0.9098, -0.2486, 1.7183, 0.3586, -0.4905, -0.0874, 0.6515, 0.0615],
  24980. [ 0.8520, -0.2772, 1.7976, -0.0532, -0.5135, -0.2319, 0.7554, 0.2187],
  24981. [ 0.7001, -0.3704, 1.6785, 0.3245, -0.5234, -0.0619, 0.8263, 0.1866],
  24982. [ 0.9059, -0.2046, 1.4341, -1.0454, -0.3405, -1.1364, 0.4722, 0.2305],
  24983. [ 0.4765, -0.4984, 1.7903, -0.3978, -0.5214, 0.2342, 0.6698, 0.2583],
  24984. [ 0.7447, -0.3002, 1.2680, -1.0002, -0.2925, -1.2363, 0.4049, 0.2539],
  24985. [ 0.6188, -0.3739, 1.7510, -0.1212, -0.6343, -0.6131, 0.1172, 0.2812],
  24986. [-1.8595, -2.0222, 1.0496, -1.0688, -0.3478, -1.2907, 0.3034, 0.3780]],
  24987. device='cuda:0', grad_fn=<AddmmBackward>)
  24988. landmarks are: tensor([[[ 6.3949e-01, -4.2132e-01, 1.7037e+00, 3.6228e-01, -4.3256e-01,
  24989. -1.0731e-01, 6.5602e-01, -4.8817e-03],
  24990. [ 6.0754e-01, -4.5138e-01, 1.8032e+00, -8.2167e-02, -5.0606e-01,
  24991. -2.0228e-01, 6.2076e-01, 1.7788e-01],
  24992. [ 6.5036e-01, -3.7434e-01, 1.6171e+00, 4.3156e-01, -5.4226e-01,
  24993. 4.6189e-04, 7.1085e-01, 1.6077e-01],
  24994. [ 5.7962e-01, -3.8776e-01, 1.3688e+00, -1.0542e+00, -4.0947e-01,
  24995. -1.1312e+00, 5.8938e-01, 1.9292e-01],
  24996. [ 5.8857e-01, -4.2525e-01, 1.8654e+00, -3.4596e-01, -5.4804e-01,
  24997. 3.6228e-01, 6.5866e-01, 1.0054e-01],
  24998. [ 5.9107e-01, -4.0805e-01, 1.2303e+00, -9.1563e-01, -3.2286e-01,
  24999. -1.2851e+00, 4.5081e-01, 1.8522e-01],
  25000. [ 5.4434e-01, -3.9938e-01, 1.7499e+00, -1.2271e-01, -6.1732e-01,
  25001. -5.7691e-01, 4.8756e-03, 2.0706e-01],
  25002. [-2.2859e+00, -2.2859e+00, 1.0712e+00, -1.2085e+00, -3.8060e-01,
  25003. -1.3929e+00, 7.5520e-02, 2.0062e-01]]], device='cuda:0')
  25004. loss_train_step before backward: tensor(0.0147, device='cuda:0', grad_fn=<MseLossBackward>)
  25005. loss_train_step after backward: tensor(0.0147, device='cuda:0', grad_fn=<MseLossBackward>)
  25006. loss_train: 1.2496287617832422
  25007. step: 46
  25008. running loss: 0.027165842647461788
  25009. Train Steps: 46/90 Loss: 0.0272 torch.Size([8, 600, 800])
  25010. torch.Size([8, 8])
  25011. tensor([[0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5737, 0.5350],
  25012. [0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250],
  25013. [0.6357, 0.4097, 0.9038, 0.3883, 0.4213, 0.2950, 0.6686, 0.5390],
  25014. [0.6276, 0.4235, 0.8888, 0.5333, 0.3800, 0.3117, 0.5427, 0.6164],
  25015. [0.6293, 0.4097, 0.8800, 0.2517, 0.5262, 0.2600, 0.7430, 0.5378],
  25016. [0.6179, 0.4118, 0.7278, 0.4237, 0.3588, 0.3400, 0.5675, 0.5917],
  25017. [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167],
  25018. [0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585]],
  25019. device='cuda:0', dtype=torch.float64)
  25020. predictions are: tensor([[ 0.6078, -0.4097, 1.7259, -0.0724, -0.5862, -0.7533, 0.3447, 0.1366],
  25021. [ 0.5774, -0.4651, 1.8405, -0.4214, -0.6620, -0.7016, 0.6259, 0.1238],
  25022. [ 0.5436, -0.4845, 1.7181, -0.5656, -0.4144, -0.8562, 0.7794, 0.1202],
  25023. [ 0.5757, -0.4130, 1.6117, 0.2724, -0.5682, -0.5975, 0.3891, 0.3914],
  25024. [ 0.5848, -0.4938, 1.6619, -1.2172, 0.1144, -1.1763, 1.1502, 0.1497],
  25025. [ 0.5044, -0.4762, 1.2517, -0.6019, -0.6974, -0.6827, 0.2724, 0.4284],
  25026. [ 0.3666, -0.5354, 1.0874, -0.8050, -0.0549, -1.2401, 0.2733, 0.4696],
  25027. [-0.0309, -0.8409, 1.6236, -0.3864, -0.3763, 0.3611, 0.4533, 0.2188]],
  25028. device='cuda:0', grad_fn=<AddmmBackward>)
  25029. landmarks are: tensor([[[ 0.5901, -0.4000, 1.8423, -0.0688, -0.5307, -0.9233, 0.3642,
  25030. 0.1852],
  25031. [ 0.6111, -0.3828, 1.8885, -0.3844, -0.5654, -0.8079, 0.5663,
  25032. 0.1390],
  25033. [ 0.6504, -0.3936, 1.8885, -0.4922, -0.3402, -0.9233, 0.8022,
  25034. 0.2035],
  25035. [ 0.6131, -0.3299, 1.8192, 0.1775, -0.5307, -0.8463, 0.2206,
  25036. 0.5612],
  25037. [ 0.6207, -0.3936, 1.7788, -1.1235, 0.1448, -1.0850, 1.1459,
  25038. 0.1982],
  25039. [ 0.5680, -0.3840, 1.0756, -0.3290, -0.6289, -0.7155, 0.3353,
  25040. 0.4470],
  25041. [ 0.6161, -0.3075, 1.1678, -0.6500, 0.0813, -1.4006, 0.2545,
  25042. 0.5624],
  25043. [ 0.5115, -0.4332, 1.7557, -0.3152, -0.2536, 0.3931, 0.4139,
  25044. 0.2936]]], device='cuda:0')
  25045. loss_train_step before backward: tensor(0.0190, device='cuda:0', grad_fn=<MseLossBackward>)
  25046. loss_train_step after backward: tensor(0.0190, device='cuda:0', grad_fn=<MseLossBackward>)
  25047. loss_train: 1.2686103787273169
  25048. step: 47
  25049. running loss: 0.026991710185687592
  25050. Train Steps: 47/90 Loss: 0.0270 torch.Size([8, 600, 800])
  25051. torch.Size([8, 8])
  25052. tensor([[0.6075, 0.4007, 0.8275, 0.4917, 0.4050, 0.5100, 0.5167, 0.5280],
  25053. [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367],
  25054. [0.6241, 0.4143, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550],
  25055. [ nan, nan, 0.8463, 0.2550, 0.5850, 0.2133, 0.7129, 0.6072],
  25056. [0.6164, 0.4119, 0.7913, 0.2650, 0.3538, 0.3500, 0.5614, 0.5038],
  25057. [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170],
  25058. [0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153],
  25059. [ nan, nan, 0.7425, 0.2117, 0.3937, 0.2433, 0.5438, 0.5567]],
  25060. device='cuda:0', dtype=torch.float64)
  25061. predictions are: tensor([[ 0.7934, -0.2449, 1.5154, 0.1089, -0.4596, -0.0577, 0.1022, 0.1616],
  25062. [ 0.7675, -0.2803, 1.3185, -0.8539, -0.6104, -0.9508, 0.4776, 0.2612],
  25063. [ 0.7659, -0.2658, 1.7689, -0.0246, -0.5172, -0.0341, 0.4878, 0.2810],
  25064. [-1.6360, -1.8705, 1.5813, -0.9753, 0.1505, -1.2824, 1.0845, 0.4963],
  25065. [ 0.7661, -0.2863, 1.3170, -0.9580, -0.6673, -0.7572, 0.3504, 0.1004],
  25066. [ 1.1099, -0.0842, 1.4959, -0.8644, -0.0398, -1.5079, 0.7408, 0.0879],
  25067. [ 0.8245, -0.2819, 1.6245, 0.2629, -0.2925, 0.0168, 0.4868, 0.1323],
  25068. [-1.3622, -1.6601, 1.1851, -0.9392, -0.4514, -1.1801, 0.2310, 0.3086]],
  25069. device='cuda:0', grad_fn=<AddmmBackward>)
  25070. landmarks are: tensor([[[ 0.5200, -0.4353, 1.5363, -0.0149, -0.4152, 0.0697, 0.1005,
  25071. 0.1530],
  25072. [ 0.5697, -0.4442, 1.3529, -0.9515, -0.5774, -0.7801, 0.5253,
  25073. 0.1931],
  25074. [ 0.5968, -0.3725, 1.8423, -0.1381, -0.4037, 0.1852, 0.6009,
  25075. 0.2776],
  25076. [-2.2859, -2.2859, 1.6229, -1.1081, 0.4162, -1.3005, 1.0070,
  25077. 0.5188],
  25078. [ 0.5614, -0.3834, 1.3688, -1.0619, -0.6520, -0.6693, 0.3069,
  25079. 0.0412],
  25080. [ 0.6471, -0.3840, 1.5767, -1.0311, -0.0457, -1.5007, 0.6889,
  25081. 0.1020],
  25082. [ 0.5548, -0.4682, 1.7309, 0.0966, -0.1394, 0.1757, 0.5045,
  25083. 0.0942],
  25084. [-2.2859, -2.2859, 1.1436, -1.3082, -0.4672, -1.1620, 0.2256,
  25085. 0.2853]]], device='cuda:0')
  25086. loss_train_step before backward: tensor(0.0482, device='cuda:0', grad_fn=<MseLossBackward>)
  25087. loss_train_step after backward: tensor(0.0482, device='cuda:0', grad_fn=<MseLossBackward>)
  25088. loss_train: 1.3167862426489592
  25089. step: 48
  25090. running loss: 0.027433046721853316
  25091.  
  25092. Train Steps: 48/90 Loss: 0.0274 torch.Size([8, 600, 800])
  25093. torch.Size([8, 8])
  25094. tensor([[0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329],
  25095. [0.6143, 0.4034, 0.8800, 0.4833, 0.4512, 0.5367, 0.5289, 0.5097],
  25096. [0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719],
  25097. [0.6201, 0.3970, 0.8413, 0.4950, 0.4413, 0.5183, 0.6088, 0.5400],
  25098. [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297],
  25099. [0.6097, 0.4000, 0.7325, 0.2667, 0.3450, 0.3517, 0.5284, 0.5045],
  25100. [0.6293, 0.3982, 0.8700, 0.5300, 0.3763, 0.4717, 0.7050, 0.5297],
  25101. [0.6361, 0.4076, 0.8862, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297]],
  25102. device='cuda:0', dtype=torch.float64)
  25103. predictions are: tensor([[ 0.6029, -0.4274, 1.6532, -0.8771, -0.5784, -0.8858, 0.4872, 0.2820],
  25104. [ 0.3579, -0.5499, 1.8201, -0.1189, -0.1082, -0.1930, 0.1005, 0.1570],
  25105. [ 0.2582, -0.6223, 1.5448, -0.6216, -0.5208, -0.6415, 0.1641, 0.4178],
  25106. [ 0.4707, -0.5316, 1.7071, -0.0668, -0.2033, -0.1684, 0.5052, 0.3034],
  25107. [ 0.5893, -0.4368, 1.6961, 0.2884, -0.4750, -0.2678, 0.7490, 0.2008],
  25108. [ 0.4832, -0.4854, 1.1430, -1.1649, -0.5456, -0.9242, 0.1531, 0.1965],
  25109. [ 0.5110, -0.4910, 1.6933, 0.1101, -0.4684, -0.2975, 0.7455, 0.2333],
  25110. [ 0.6602, -0.4004, 1.7785, 0.1148, -0.5072, -0.3808, 0.7389, 0.2826]],
  25111. device='cuda:0', grad_fn=<AddmmBackward>)
  25112. landmarks are: tensor([[[ 5.6801e-01, -4.3934e-01, 1.5920e+00, -6.6715e-01, -6.4527e-01,
  25113. -5.4566e-01, 5.1492e-01, 1.7534e-01],
  25114. [ 5.5155e-01, -4.2249e-01, 1.7788e+00, -5.3426e-02, -2.0162e-01,
  25115. 1.9292e-01, 1.5683e-01, 6.8210e-02],
  25116. [ 5.3672e-01, -4.2941e-01, 1.5709e+00, -4.9992e-01, -6.6928e-01,
  25117. -3.0747e-01, 2.4546e-01, 3.5585e-01],
  25118. [ 5.7812e-01, -4.5219e-01, 1.5998e+00, 4.6189e-04, -2.4781e-01,
  25119. 1.0824e-01, 5.2587e-01, 2.0831e-01],
  25120. [ 6.5036e-01, -3.7434e-01, 1.6171e+00, 4.3156e-01, -5.4226e-01,
  25121. 4.6189e-04, 7.1085e-01, 1.6077e-01],
  25122. [ 5.3031e-01, -4.3841e-01, 1.0975e+00, -1.0542e+00, -6.9238e-01,
  25123. -6.6159e-01, 1.5477e-01, 4.4164e-02],
  25124. [ 6.2072e-01, -4.4656e-01, 1.7326e+00, 1.6212e-01, -5.4804e-01,
  25125. -1.0731e-01, 9.7040e-01, 1.6077e-01],
  25126. [ 6.5201e-01, -4.0323e-01, 1.8076e+00, 1.8522e-01, -5.7113e-01,
  25127. -1.3811e-01, 7.8762e-01, 1.6077e-01]]], device='cuda:0')
  25128. loss_train_step before backward: tensor(0.0196, device='cuda:0', grad_fn=<MseLossBackward>)
  25129. loss_train_step after backward: tensor(0.0196, device='cuda:0', grad_fn=<MseLossBackward>)
  25130. loss_train: 1.336402228102088
  25131. step: 49
  25132. running loss: 0.027273514859226286
  25133. Train Steps: 49/90 Loss: 0.0273 torch.Size([8, 600, 800])
  25134. torch.Size([8, 8])
  25135. tensor([[0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5787, 0.5117],
  25136. [0.6213, 0.4001, 0.7712, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183],
  25137. [0.6179, 0.4008, 0.7505, 0.2678, 0.4368, 0.1891, 0.5831, 0.5263],
  25138. [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750],
  25139. [0.6053, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168],
  25140. [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679],
  25141. [0.6219, 0.4097, 0.8738, 0.3400, 0.3563, 0.4117, 0.5975, 0.5683],
  25142. [0.6030, 0.3969, 0.7988, 0.3917, 0.3450, 0.3667, 0.5266, 0.4700]],
  25143. device='cuda:0', dtype=torch.float64)
  25144. predictions are: tensor([[ 0.4697, -0.4949, 1.2264, -1.1615, -0.3115, -1.3537, 0.3968, 0.1538],
  25145. [ 0.4581, -0.5103, 1.2444, -1.2177, -0.2020, -1.4147, 0.4720, 0.1634],
  25146. [ 0.7304, -0.3405, 1.2307, -0.9687, -0.2543, -1.3712, 0.4215, 0.2137],
  25147. [ 0.5282, -0.5000, 1.8005, 0.1556, -0.2945, -0.0716, 0.3245, 0.3234],
  25148. [-0.5072, -1.1353, 0.9631, -1.1802, -0.3185, -1.3331, 0.3396, 0.2513],
  25149. [ 0.3824, -0.5971, 1.8023, 0.2948, -0.4119, 0.2892, 1.0153, 0.3397],
  25150. [ 0.5125, -0.4644, 1.8052, -0.4985, -0.6533, -0.4799, 0.4674, 0.3349],
  25151. [ 0.4898, -0.4892, 1.6309, -0.3511, -0.6462, -0.5951, 0.1274, 0.1258]],
  25152. device='cuda:0', grad_fn=<AddmmBackward>)
  25153. landmarks are: tensor([[[ 0.5598, -0.4201, 1.1898, -1.3005, -0.3806, -1.3313, 0.3873,
  25154. 0.0774],
  25155. [ 0.5838, -0.4379, 1.2764, -1.3082, -0.2824, -1.4545, 0.4162,
  25156. 0.1082],
  25157. [ 0.5680, -0.4345, 1.1806, -1.0490, -0.2683, -1.4127, 0.4074,
  25158. 0.1449],
  25159. [ 0.5525, -0.4463, 1.7326, 0.0313, -0.2536, -0.0688, 0.1968,
  25160. 0.3700],
  25161. [ 0.5102, -0.4223, 0.8999, -1.4011, -0.4383, -1.3082, 0.2267,
  25162. 0.1013],
  25163. [ 0.6257, -0.4249, 1.5998, 0.1236, -0.3806, 0.3084, 0.9887,
  25164. 0.3371],
  25165. [ 0.5867, -0.3937, 1.7499, -0.7155, -0.6404, -0.3844, 0.4739,
  25166. 0.3392],
  25167. [ 0.4992, -0.4525, 1.4035, -0.4768, -0.6924, -0.5923, 0.1465,
  25168. -0.1151]]], device='cuda:0')
  25169. loss_train_step before backward: tensor(0.0342, device='cuda:0', grad_fn=<MseLossBackward>)
  25170. loss_train_step after backward: tensor(0.0342, device='cuda:0', grad_fn=<MseLossBackward>)
  25171. loss_train: 1.370650378987193
  25172. step: 50
  25173. running loss: 0.027413007579743864
  25174. Train Steps: 50/90 Loss: 0.0274 torch.Size([8, 600, 800])
  25175. torch.Size([8, 8])
  25176. tensor([[0.6064, 0.4019, 0.8650, 0.4517, 0.4037, 0.5367, 0.5703, 0.5609],
  25177. [0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528],
  25178. [0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6063, 0.6083],
  25179. [0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148],
  25180. [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483],
  25181. [0.6122, 0.3993, 0.8738, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927],
  25182. [0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162],
  25183. [ nan, nan, 0.8463, 0.2550, 0.5850, 0.2133, 0.7129, 0.6072]],
  25184. device='cuda:0', dtype=torch.float64)
  25185. predictions are: tensor([[ 0.5260, -0.4512, 1.5853, -0.1327, -0.3905, 0.0258, 0.3438, 0.2133],
  25186. [ 0.7047, -0.3586, 1.7669, -0.6288, -0.3302, -0.8932, 0.9293, 0.2138],
  25187. [ 0.6988, -0.2873, 1.4754, -0.5759, -0.5988, -0.5795, 0.3050, 0.4355],
  25188. [ 0.7808, -0.3050, 1.6218, 0.2099, -0.4843, -0.3894, 0.3224, 0.0286],
  25189. [ 0.6624, -0.3391, 1.5512, -0.4366, -0.5093, -0.2878, 0.2257, 0.2136],
  25190. [ 0.5930, -0.4043, 1.6772, -0.0725, -0.2312, -0.2132, 0.0442, -0.0188],
  25191. [ 0.5222, -0.4644, 1.6615, -0.3672, -0.5627, -0.3673, 0.6173, 0.2334],
  25192. [-1.8953, -2.0281, 1.5608, -1.1411, 0.3558, -1.4730, 1.0002, 0.4625]],
  25193. device='cuda:0', grad_fn=<AddmmBackward>)
  25194. landmarks are: tensor([[[ 0.5151, -0.4296, 1.7095, -0.1997, -0.4210, 0.1929, 0.3484,
  25195. 0.3047],
  25196. [ 0.6388, -0.3623, 1.9173, -0.7386, -0.3517, -0.5846, 1.1495,
  25197. 0.2676],
  25198. [ 0.5774, -0.3868, 1.6286, -0.5692, -0.6462, -0.2767, 0.5143,
  25199. 0.5239],
  25200. [ 0.5773, -0.4316, 1.7399, 0.1287, -0.5153, -0.0817, 0.4313,
  25201. 0.0919],
  25202. [ 0.5546, -0.4620, 1.7788, -0.4229, -0.5192, -0.0226, 0.4277,
  25203. 0.2468],
  25204. [ 0.5417, -0.4417, 1.7499, -0.1304, -0.1994, -0.0324, 0.0951,
  25205. -0.0099],
  25206. [ 0.6059, -0.4442, 1.8921, -0.3422, -0.6230, -0.1461, 0.7489,
  25207. 0.0985],
  25208. [-2.2859, -2.2859, 1.6229, -1.1081, 0.4162, -1.3005, 1.0070,
  25209. 0.5188]]], device='cuda:0')
  25210. loss_train_step before backward: tensor(0.0200, device='cuda:0', grad_fn=<MseLossBackward>)
  25211. loss_train_step after backward: tensor(0.0200, device='cuda:0', grad_fn=<MseLossBackward>)
  25212. loss_train: 1.3906042091548443
  25213. step: 51
  25214. running loss: 0.027266749199114593
  25215. Train Steps: 51/90 Loss: 0.0273 torch.Size([8, 600, 800])
  25216. torch.Size([8, 8])
  25217. tensor([[0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901],
  25218. [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283],
  25219. [0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967],
  25220. [0.6216, 0.4099, 0.7225, 0.2033, 0.4188, 0.2217, 0.5975, 0.5283],
  25221. [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483],
  25222. [0.6224, 0.4097, 0.7438, 0.2267, 0.3850, 0.2850, 0.5988, 0.5250],
  25223. [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726],
  25224. [0.6110, 0.3984, 0.8750, 0.4933, 0.4625, 0.4950, 0.5578, 0.5676]],
  25225. device='cuda:0', dtype=torch.float64)
  25226. predictions are: tensor([[ 0.4308, -0.5393, 1.9275, -0.0703, -0.3942, -0.2634, 0.1682, 0.1026],
  25227. [ 0.3524, -0.5766, 1.8549, -0.2104, -0.5534, -0.6176, 0.5192, 0.3368],
  25228. [ 0.3679, -0.5808, 1.7765, -0.1534, -0.5801, 0.0536, 0.4781, 0.1789],
  25229. [ 0.4266, -0.5153, 1.0804, -1.1756, -0.2470, -1.4130, 0.3684, 0.2618],
  25230. [ 0.5300, -0.4728, 1.1145, -1.2576, -0.2983, -1.2850, 0.6322, 0.2698],
  25231. [ 0.2888, -0.6171, 1.1986, -1.1513, -0.3956, -1.0440, 0.6077, 0.2181],
  25232. [ 0.6086, -0.4195, 1.2452, -0.9810, -0.4951, -1.0781, 0.2751, 0.0469],
  25233. [ 0.2659, -0.6734, 1.8575, 0.0737, -0.0717, -0.0615, 0.3565, 0.3574]],
  25234. device='cuda:0', grad_fn=<AddmmBackward>)
  25235. landmarks are: tensor([[[ 5.6634e-01, -3.9546e-01, 1.7788e+00, -2.3818e-01, -4.0370e-01,
  25236. -2.6898e-01, 8.2802e-02, -2.1963e-02],
  25237. [ 5.8799e-01, -3.8868e-01, 1.8423e+00, -3.3056e-01, -6.2309e-01,
  25238. -5.2302e-01, 4.0462e-01, 1.5443e-01],
  25239. [ 4.9740e-01, -4.4819e-01, 1.6633e+00, -3.3056e-01, -6.1732e-01,
  25240. 1.3133e-01, 2.9255e-01, 8.0947e-03],
  25241. [ 5.8505e-01, -3.9246e-01, 1.0513e+00, -1.3467e+00, -3.5173e-01,
  25242. -1.2620e+00, 4.7390e-01, 1.5443e-01],
  25243. [ 5.7841e-01, -4.0847e-01, 1.0859e+00, -1.3929e+00, -4.0370e-01,
  25244. -1.1158e+00, 5.6051e-01, 2.4681e-01],
  25245. [ 5.8909e-01, -3.9369e-01, 1.1494e+00, -1.2390e+00, -5.0762e-01,
  25246. -9.6952e-01, 4.7968e-01, 1.3903e-01],
  25247. [ 4.9971e-01, -4.4465e-01, 1.1610e+00, -9.7721e-01, -6.0577e-01,
  25248. -1.0311e+00, 1.4038e-01, -1.0312e-01],
  25249. [ 5.3637e-01, -4.4573e-01, 1.7557e+00, -7.2363e-03, -1.4965e-01,
  25250. 4.6189e-04, 2.9049e-01, 3.3573e-01]]], device='cuda:0')
  25251. loss_train_step before backward: tensor(0.0164, device='cuda:0', grad_fn=<MseLossBackward>)
  25252. loss_train_step after backward: tensor(0.0164, device='cuda:0', grad_fn=<MseLossBackward>)
  25253. loss_train: 1.4069728069007397
  25254. step: 52
  25255. running loss: 0.027057169363475762
  25256.  
  25257. Train Steps: 52/90 Loss: 0.0271 torch.Size([8, 600, 800])
  25258. torch.Size([8, 8])
  25259. tensor([[0.6190, 0.4135, 0.8000, 0.4883, 0.3566, 0.3647, 0.5613, 0.5900],
  25260. [0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019],
  25261. [0.6257, 0.4167, 0.8775, 0.3433, 0.3563, 0.4133, 0.6200, 0.5667],
  25262. [0.6168, 0.4029, 0.8523, 0.3417, 0.3588, 0.5000, 0.6125, 0.5400],
  25263. [ nan, nan, 0.6793, 0.2110, 0.4012, 0.2167, 0.5112, 0.5583],
  25264. [0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343],
  25265. [0.6113, 0.4088, 0.6859, 0.2208, 0.4363, 0.1700, 0.5188, 0.5533],
  25266. [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466]],
  25267. device='cuda:0', dtype=torch.float64)
  25268. predictions are: tensor([[ 0.6663, -0.3088, 1.6004, 0.0154, -0.4911, -0.6876, 0.0623, 0.3197],
  25269. [ 0.8698, -0.2878, 1.8808, -0.0299, -0.4834, -0.4837, 0.6643, 0.0741],
  25270. [ 0.7847, -0.2747, 1.7890, -0.5887, -0.5645, -0.3924, 0.5472, 0.3043],
  25271. [ 0.5769, -0.4142, 1.6525, -0.7106, -0.5385, 0.0847, 0.6281, 0.2195],
  25272. [-2.3713, -2.3257, 1.1043, -1.3198, -0.1935, -1.3207, 0.1519, 0.2115],
  25273. [ 0.7094, -0.3830, 1.8119, -0.5825, -0.2013, -0.9141, 0.9488, 0.0407],
  25274. [ 0.3497, -0.5628, 0.9207, -1.2847, -0.1244, -1.5964, 0.2215, 0.2328],
  25275. [ 0.5139, -0.4596, 1.3448, -0.5620, -0.5908, -0.3688, 0.0872, 0.2178]],
  25276. device='cuda:0', grad_fn=<AddmmBackward>)
  25277. landmarks are: tensor([[[ 0.5732, -0.3761, 1.4092, -0.0303, -0.6388, -0.6012, 0.3065,
  25278. 0.4393],
  25279. [ 0.6372, -0.4129, 1.8942, -0.0765, -0.6173, -0.4768, 0.6999,
  25280. 0.0325],
  25281. [ 0.6040, -0.3614, 1.7672, -0.7001, -0.6404, -0.3768, 0.5778,
  25282. 0.3315],
  25283. [ 0.5631, -0.4249, 1.6509, -0.7078, -0.6289, 0.0236, 0.5432,
  25284. 0.2083],
  25285. [-2.2859, -2.2859, 0.8516, -1.3112, -0.4326, -1.2851, 0.0755,
  25286. 0.2930],
  25287. [ 0.6125, -0.4369, 1.9173, -0.5384, -0.2594, -0.8386, 0.9741,
  25288. 0.1821],
  25289. [ 0.5377, -0.3978, 0.8824, -1.2663, -0.2709, -1.5007, 0.1102,
  25290. 0.2699],
  25291. [ 0.5076, -0.4443, 1.2337, -0.5023, -0.6808, -0.3614, 0.0866,
  25292. 0.2386]]], device='cuda:0')
  25293. loss_train_step before backward: tensor(0.0103, device='cuda:0', grad_fn=<MseLossBackward>)
  25294. loss_train_step after backward: tensor(0.0103, device='cuda:0', grad_fn=<MseLossBackward>)
  25295. loss_train: 1.4172964449971914
  25296. step: 53
  25297. running loss: 0.026741442358437575
  25298. Train Steps: 53/90 Loss: 0.0267 torch.Size([8, 600, 800])
  25299. torch.Size([8, 8])
  25300. tensor([[0.6164, 0.4076, 0.8838, 0.4117, 0.3713, 0.5550, 0.6238, 0.5350],
  25301. [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667],
  25302. [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208],
  25303. [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517],
  25304. [0.6250, 0.4054, 0.8770, 0.4723, 0.4662, 0.5367, 0.6162, 0.5433],
  25305. [0.6265, 0.4071, 0.8875, 0.3367, 0.3975, 0.3350, 0.6312, 0.5250],
  25306. [ nan, nan, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563],
  25307. [0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317]],
  25308. device='cuda:0', dtype=torch.float64)
  25309. predictions are: tensor([[ 0.5588, -0.4190, 1.8389, -0.3034, -0.4850, 0.2965, 0.6388, 0.2382],
  25310. [ 0.5971, -0.3907, 1.5570, -0.6910, -0.6359, -0.7238, 0.2675, 0.2644],
  25311. [ 0.5777, -0.4140, 1.4146, -1.0344, -0.2636, -1.2908, 0.4869, 0.0262],
  25312. [ 0.3478, -0.5392, 0.8901, -1.1096, -0.2750, -1.3271, 0.4063, 0.2873],
  25313. [ 0.6769, -0.3749, 1.8106, -0.0091, -0.1602, 0.2302, 0.5604, 0.2009],
  25314. [ 0.6794, -0.3695, 1.8073, -0.6632, -0.4131, -0.8047, 0.6484, 0.1249],
  25315. [-1.6592, -1.8379, 1.3722, -0.7628, -0.4710, -1.0075, -0.0038, 0.1739],
  25316. [ 0.3202, -0.5807, 1.1559, -1.1361, -0.2836, -1.1998, 0.3835, 0.2031]],
  25317. device='cuda:0', grad_fn=<AddmmBackward>)
  25318. landmarks are: tensor([[[ 0.5614, -0.4032, 1.7961, -0.3844, -0.5711, 0.2776, 0.5952,
  25319. 0.1852],
  25320. [ 0.5768, -0.3857, 1.5305, -0.7694, -0.6462, -0.6308, 0.3988,
  25321. 0.3315],
  25322. [ 0.5885, -0.4429, 1.4266, -0.9926, -0.4383, -1.2313, 0.4228,
  25323. 0.1195],
  25324. [ 0.5713, -0.3671, 0.8665, -1.0696, -0.3691, -1.2236, 0.3527,
  25325. 0.2622],
  25326. [ 0.6009, -0.4135, 1.7651, -0.1043, -0.1323, 0.1929, 0.5605,
  25327. 0.2237],
  25328. [ 0.6078, -0.4057, 1.8134, -0.7309, -0.4499, -0.7386, 0.6298,
  25329. 0.1390],
  25330. [-2.2859, -2.2859, 1.4006, -0.8105, -0.6115, -0.8232, 0.0419,
  25331. 0.2837],
  25332. [ 0.5849, -0.3836, 1.2649, -1.2236, -0.3460, -1.2313, 0.4508,
  25333. 0.1698]]], device='cuda:0')
  25334. loss_train_step before backward: tensor(0.0165, device='cuda:0', grad_fn=<MseLossBackward>)
  25335. loss_train_step after backward: tensor(0.0165, device='cuda:0', grad_fn=<MseLossBackward>)
  25336. loss_train: 1.4337830413132906
  25337. step: 54
  25338. running loss: 0.026551537802097975
  25339. Train Steps: 54/90 Loss: 0.0266 torch.Size([8, 600, 800])
  25340. torch.Size([8, 8])
  25341. tensor([[0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6088, 0.5583],
  25342. [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791],
  25343. [ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600],
  25344. [0.6179, 0.3993, 0.8925, 0.4789, 0.3879, 0.4900, 0.6041, 0.5279],
  25345. [0.6154, 0.4112, 0.7037, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600],
  25346. [0.6197, 0.4050, 0.7527, 0.2000, 0.4042, 0.2249, 0.5895, 0.4995],
  25347. [0.6286, 0.4055, 0.9000, 0.4717, 0.3763, 0.4683, 0.7018, 0.5494],
  25348. [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167]],
  25349. device='cuda:0', dtype=torch.float64)
  25350. predictions are: tensor([[ 0.5739, -0.4075, 1.0099, -1.2497, -0.4887, -1.0567, 0.4248, 0.2145],
  25351. [ 0.6474, -0.3825, 1.8956, 0.0107, -0.5090, 0.2286, 0.5319, 0.0127],
  25352. [-2.0089, -2.0760, 1.3973, -1.0495, -0.2714, -1.0593, 0.1613, 0.2091],
  25353. [ 0.5967, -0.3914, 1.8916, -0.0790, -0.5793, 0.0853, 0.3931, 0.1261],
  25354. [ 0.6433, -0.3461, 0.9987, -1.1937, -0.3592, -1.3822, 0.1790, 0.2450],
  25355. [ 0.6662, -0.3432, 1.2133, -1.2991, -0.4466, -1.1694, 0.4190, 0.0401],
  25356. [ 0.4311, -0.5357, 1.9630, -0.0693, -0.5812, -0.0296, 0.7530, 0.2171],
  25357. [ 0.4983, -0.3966, 1.1779, -0.7674, -0.0606, -1.3045, 0.2264, 0.4292]],
  25358. device='cuda:0', grad_fn=<AddmmBackward>)
  25359. landmarks are: tensor([[[ 0.5992, -0.3968, 0.9307, -1.3497, -0.4730, -1.0465, 0.5259,
  25360. 0.2930],
  25361. [ 0.6074, -0.4223, 1.8192, -0.0303, -0.4152, 0.1236, 0.6524,
  25362. -0.0731],
  25363. [-2.2859, -2.2859, 1.4266, -1.1389, -0.4499, -1.1235, 0.2891,
  25364. 0.3007],
  25365. [ 0.5680, -0.4417, 1.8365, -0.0740, -0.4941, -0.0227, 0.5045,
  25366. 0.1525],
  25367. [ 0.5565, -0.3864, 0.9647, -1.2159, -0.3286, -1.4391, 0.1852,
  25368. 0.3007],
  25369. [ 0.5762, -0.4153, 1.1908, -1.3622, -0.4190, -1.2471, 0.4368,
  25370. 0.0213],
  25371. [ 0.6174, -0.4129, 1.8711, -0.1073, -0.5480, -0.1227, 0.9558,
  25372. 0.2516],
  25373. [ 0.6161, -0.3075, 1.1678, -0.6500, 0.0813, -1.4006, 0.2545,
  25374. 0.5624]]], device='cuda:0')
  25375. loss_train_step before backward: tensor(0.0087, device='cuda:0', grad_fn=<MseLossBackward>)
  25376. loss_train_step after backward: tensor(0.0087, device='cuda:0', grad_fn=<MseLossBackward>)
  25377. loss_train: 1.4424332054331899
  25378. step: 55
  25379. running loss: 0.026226058280603454
  25380. Train Steps: 55/90 Loss: 0.0262 torch.Size([8, 600, 800])
  25381. torch.Size([8, 8])
  25382. tensor([[0.6097, 0.4024, 0.8488, 0.3717, 0.3875, 0.5517, 0.5836, 0.5591],
  25383. [ nan, nan, 0.6469, 0.1943, 0.4025, 0.2000, 0.5125, 0.5533],
  25384. [0.6346, 0.4086, 0.7938, 0.5500, 0.3962, 0.4867, 0.7343, 0.5702],
  25385. [0.6132, 0.4037, 0.6963, 0.2217, 0.4100, 0.1950, 0.5395, 0.5175],
  25386. [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202],
  25387. [0.6200, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913],
  25388. [0.6127, 0.4115, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495],
  25389. [0.6197, 0.4050, 0.7527, 0.2000, 0.4042, 0.2249, 0.5895, 0.4995]],
  25390. device='cuda:0', dtype=torch.float64)
  25391. predictions are: tensor([[ 0.4872, -0.4469, 1.7565, -0.5619, -0.4863, 0.3109, 0.5125, 0.2862],
  25392. [-1.7611, -1.9040, 0.9830, -1.3490, -0.3156, -1.5371, 0.0834, 0.3199],
  25393. [ 0.5838, -0.4097, 1.6127, 0.1440, -0.5238, -0.0731, 0.8983, 0.3705],
  25394. [ 0.5559, -0.3909, 1.0559, -1.3005, -0.4127, -1.4387, 0.2621, 0.0786],
  25395. [ 0.6276, -0.3755, 1.8506, 0.0234, -0.3498, 0.1800, 0.3969, 0.1361],
  25396. [ 0.4281, -0.5543, 1.8947, 0.0836, -0.3881, 0.0256, 0.3248, 0.0540],
  25397. [ 0.5170, -0.4141, 1.1629, -1.0955, -0.6339, -0.8938, 0.1307, 0.2361],
  25398. [ 0.7367, -0.3005, 1.2420, -1.3779, -0.4522, -1.2652, 0.4379, 0.0683]],
  25399. device='cuda:0', grad_fn=<AddmmBackward>)
  25400. landmarks are: tensor([[[ 0.5301, -0.4273, 1.6344, -0.5692, -0.4961, 0.2622, 0.4098,
  25401. 0.2966],
  25402. [-2.2859, -2.2859, 0.7023, -1.3883, -0.4268, -1.3621, 0.0813,
  25403. 0.2699],
  25404. [ 0.6454, -0.3984, 1.3804, 0.2545, -0.4557, -0.0380, 1.1057,
  25405. 0.3478],
  25406. [ 0.5465, -0.4214, 0.9300, -1.2620, -0.3921, -1.3852, 0.2062,
  25407. 0.1043],
  25408. [ 0.5614, -0.4586, 1.7587, 0.0641, -0.2998, 0.1712, 0.4958,
  25409. 0.1170],
  25410. [ 0.5776, -0.4784, 1.7730, 0.1236, -0.4037, 0.0697, 0.5132,
  25411. -0.0168],
  25412. [ 0.5442, -0.3855, 1.0224, -0.9541, -0.6115, -0.9233, 0.1745,
  25413. 0.2522],
  25414. [ 0.5762, -0.4153, 1.1908, -1.3622, -0.4190, -1.2471, 0.4368,
  25415. 0.0213]]], device='cuda:0')
  25416. loss_train_step before backward: tensor(0.0146, device='cuda:0', grad_fn=<MseLossBackward>)
  25417. loss_train_step after backward: tensor(0.0146, device='cuda:0', grad_fn=<MseLossBackward>)
  25418. loss_train: 1.4570275181904435
  25419. step: 56
  25420. running loss: 0.026018348539115062
  25421.  
  25422. Train Steps: 56/90 Loss: 0.0260 torch.Size([8, 600, 800])
  25423. torch.Size([8, 8])
  25424. tensor([[0.6233, 0.4091, 0.8100, 0.2950, 0.3563, 0.3883, 0.6013, 0.5200],
  25425. [0.6268, 0.4029, 0.8500, 0.2683, 0.3937, 0.3500, 0.6860, 0.5297],
  25426. [0.6248, 0.4032, 0.7738, 0.1900, 0.4813, 0.1400, 0.5941, 0.4904],
  25427. [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773],
  25428. [0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343],
  25429. [0.6219, 0.4097, 0.8738, 0.3400, 0.3563, 0.4117, 0.5975, 0.5683],
  25430. [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650],
  25431. [0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904]],
  25432. device='cuda:0', dtype=torch.float64)
  25433. predictions are: tensor([[ 0.4687, -0.4761, 1.3885, -0.9022, -0.7235, -0.3510, 0.5231, 0.1867],
  25434. [ 0.5900, -0.4263, 1.4820, -1.0055, -0.6282, -0.5480, 0.8535, 0.1611],
  25435. [ 0.5019, -0.4647, 1.1006, -1.1980, -0.0975, -1.5659, 0.3672, 0.0922],
  25436. [ 0.2821, -0.6068, 1.6277, -0.0317, -0.0875, 0.1353, -0.0203, 0.2949],
  25437. [ 0.4712, -0.5273, 1.7103, -0.5717, -0.4082, -0.7752, 0.8000, 0.1014],
  25438. [ 0.3002, -0.5729, 1.6020, -0.7157, -0.7064, -0.2938, 0.3533, 0.2870],
  25439. [ 0.1574, -0.6228, 1.4466, -0.3927, -0.6264, -0.6890, 0.0769, 0.4375],
  25440. [ 0.1880, -0.6467, 1.4838, -1.2031, -0.1930, -1.1545, 0.4361, 0.0573]],
  25441. device='cuda:0', grad_fn=<AddmmBackward>)
  25442. landmarks are: tensor([[[ 0.5932, -0.3962, 1.4554, -0.9233, -0.6404, -0.4922, 0.4912,
  25443. 0.1159],
  25444. [ 0.6092, -0.4249, 1.6402, -1.0465, -0.4672, -0.6693, 0.8827,
  25445. 0.1608],
  25446. [ 0.5999, -0.4236, 1.2880, -1.4083, -0.0630, -1.6393, 0.4584,
  25447. -0.0208],
  25448. [ 0.5428, -0.4244, 1.7095, -0.0380, 0.0232, 0.0725, 0.0866,
  25449. 0.3806],
  25450. [ 0.6125, -0.4369, 1.9173, -0.5384, -0.2594, -0.8386, 0.9741,
  25451. 0.1821],
  25452. [ 0.5867, -0.3937, 1.7499, -0.7155, -0.6404, -0.3844, 0.4739,
  25453. 0.3392],
  25454. [ 0.5894, -0.3503, 1.6402, -0.3614, -0.5827, -0.7925, 0.3238,
  25455. 0.3238],
  25456. [ 0.6126, -0.3871, 1.6229, -1.0773, -0.2132, -1.3698, 0.5829,
  25457. -0.0209]]], device='cuda:0')
  25458. loss_train_step before backward: tensor(0.0224, device='cuda:0', grad_fn=<MseLossBackward>)
  25459. loss_train_step after backward: tensor(0.0224, device='cuda:0', grad_fn=<MseLossBackward>)
  25460. loss_train: 1.4794276067987084
  25461. step: 57
  25462. running loss: 0.02595487029471418
  25463. Train Steps: 57/90 Loss: 0.0260 torch.Size([8, 600, 800])
  25464. torch.Size([8, 8])
  25465. tensor([[0.6275, 0.4050, 0.9038, 0.3767, 0.3838, 0.3533, 0.7074, 0.5575],
  25466. [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153],
  25467. [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767],
  25468. [0.6175, 0.3957, 0.8700, 0.4817, 0.4662, 0.5133, 0.5800, 0.5517],
  25469. [0.6168, 0.4029, 0.8523, 0.3417, 0.3588, 0.5000, 0.6125, 0.5400],
  25470. [0.6204, 0.4110, 0.7913, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367],
  25471. [0.6126, 0.4039, 0.8237, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138],
  25472. [0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400]],
  25473. device='cuda:0', dtype=torch.float64)
  25474. predictions are: tensor([[ 0.5395, -0.4694, 1.7540, -0.6162, -0.3841, -0.6810, 0.9446, 0.2789],
  25475. [ 0.5050, -0.4520, 1.3034, -1.0162, -0.6673, -0.7612, 0.3418, 0.0111],
  25476. [ 0.4511, -0.4448, 1.5851, -0.2066, -0.5379, -0.8791, 0.2361, 0.2449],
  25477. [ 0.3331, -0.6058, 1.6994, -0.1982, -0.0322, 0.0858, 0.1891, 0.1661],
  25478. [ 0.4013, -0.5315, 1.5514, -0.8512, -0.5781, 0.1099, 0.5527, 0.2199],
  25479. [ 0.5276, -0.4425, 1.3316, -1.2870, -0.3980, -1.1945, 0.4547, 0.0442],
  25480. [ 0.3599, -0.5380, 1.4635, -0.6173, -0.6199, -0.5927, 0.3554, 0.4287],
  25481. [ 0.3724, -0.5610, 1.7453, -0.5381, -0.5880, -0.0687, 0.4760, 0.2247]],
  25482. device='cuda:0', grad_fn=<AddmmBackward>)
  25483. landmarks are: tensor([[[ 0.6125, -0.4153, 1.8885, -0.5461, -0.5134, -0.6539, 0.9814,
  25484. 0.2890],
  25485. [ 0.5614, -0.4514, 1.3252, -0.7240, -0.6237, -0.7215, 0.4299,
  25486. 0.0942],
  25487. [ 0.5781, -0.3848, 1.7441, -0.0996, -0.5769, -0.8002, 0.2021,
  25488. 0.3778],
  25489. [ 0.5662, -0.4581, 1.7326, -0.0611, -0.1323, 0.0851, 0.3931,
  25490. 0.2622],
  25491. [ 0.5631, -0.4249, 1.6509, -0.7078, -0.6289, 0.0236, 0.5432,
  25492. 0.2083],
  25493. [ 0.5796, -0.3878, 1.3688, -1.0542, -0.4095, -1.1312, 0.5894,
  25494. 0.1929],
  25495. [ 0.5438, -0.4206, 1.5189, -0.4537, -0.6115, -0.6231, 0.4365,
  25496. 0.5491],
  25497. [ 0.5778, -0.4393, 1.8018, -0.4614, -0.6693, -0.1381, 0.5490,
  25498. 0.2083]]], device='cuda:0')
  25499. loss_train_step before backward: tensor(0.0119, device='cuda:0', grad_fn=<MseLossBackward>)
  25500. loss_train_step after backward: tensor(0.0119, device='cuda:0', grad_fn=<MseLossBackward>)
  25501. loss_train: 1.491361235268414
  25502. step: 58
  25503. running loss: 0.02571312474600714
  25504. Train Steps: 58/90 Loss: 0.0257 torch.Size([8, 600, 800])
  25505. torch.Size([8, 8])
  25506. tensor([[ nan, nan, 0.7512, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617],
  25507. [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5513, 0.5750],
  25508. [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261],
  25509. [0.6201, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044],
  25510. [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012],
  25511. [0.6225, 0.4116, 0.8662, 0.3517, 0.3663, 0.3233, 0.5837, 0.5317],
  25512. [0.6230, 0.4152, 0.7588, 0.2283, 0.4012, 0.2883, 0.6200, 0.5767],
  25513. [0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117]],
  25514. device='cuda:0', dtype=torch.float64)
  25515. predictions are: tensor([[-1.0428, -1.4417, 1.1209, -1.3021, -0.3286, -1.2498, 0.1953, 0.2366],
  25516. [-0.5951, -1.1337, 0.8469, -1.3820, -0.3139, -1.3906, 0.1110, 0.3584],
  25517. [ 0.7358, -0.2749, 1.8279, -0.0705, -0.3832, 0.1568, 0.2909, 0.1203],
  25518. [ 0.6467, -0.3631, 1.2392, -1.4253, -0.4089, -1.1979, 0.4387, 0.0469],
  25519. [ 0.7252, -0.3748, 1.8057, 0.2314, -0.5462, -0.1135, 0.6252, 0.0910],
  25520. [ 0.7185, -0.2933, 1.6671, -0.7556, -0.6140, -0.6193, 0.4088, 0.1966],
  25521. [ 0.6939, -0.3297, 1.2181, -1.2268, -0.5244, -0.8829, 0.6186, 0.3606],
  25522. [ 0.6410, -0.4152, 1.8992, -0.1185, -0.5496, 0.0964, 0.7300, 0.1463]],
  25523. device='cuda:0', grad_fn=<AddmmBackward>)
  25524. landmarks are: tensor([[[-2.2859, -2.2859, 1.1841, -1.3082, -0.3055, -1.3621, 0.3007,
  25525. 0.3084],
  25526. [ 0.5663, -0.4396, 0.8261, -1.1312, -0.2940, -1.3929, 0.2603,
  25527. 0.3700],
  25528. [ 0.5746, -0.3623, 1.7961, -0.1150, -0.3691, -0.0380, 0.2208,
  25529. 0.1439],
  25530. [ 0.5785, -0.4129, 1.2785, -1.3996, -0.3227, -1.3259, 0.4258,
  25531. 0.0438],
  25532. [ 0.6249, -0.4352, 1.8018, 0.2545, -0.6173, -0.1997, 0.6401,
  25533. 0.0291],
  25534. [ 0.5893, -0.3847, 1.7152, -0.6616, -0.5942, -0.7925, 0.4104,
  25535. 0.1698],
  25536. [ 0.5915, -0.3682, 1.2187, -1.2313, -0.4326, -0.9541, 0.5778,
  25537. 0.3777],
  25538. [ 0.6262, -0.4461, 1.8480, -0.0534, -0.5827, -0.1227, 0.6587,
  25539. 0.0774]]], device='cuda:0')
  25540. loss_train_step before backward: tensor(0.0705, device='cuda:0', grad_fn=<MseLossBackward>)
  25541. loss_train_step after backward: tensor(0.0705, device='cuda:0', grad_fn=<MseLossBackward>)
  25542. loss_train: 1.5618963139131665
  25543. step: 59
  25544. running loss: 0.026472818879884178
  25545. Train Steps: 59/90 Loss: 0.0265 torch.Size([8, 600, 800])
  25546. torch.Size([8, 8])
  25547. tensor([[0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058],
  25548. [0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183],
  25549. [0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6362, 0.5200],
  25550. [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852],
  25551. [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317],
  25552. [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050],
  25553. [ nan, nan, 0.7425, 0.2117, 0.3937, 0.2433, 0.5438, 0.5567],
  25554. [0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631]],
  25555. device='cuda:0', dtype=torch.float64)
  25556. predictions are: tensor([[ 0.6830, -0.3160, 1.5683, -0.4438, -0.5378, -0.1399, 0.1804, 0.0487],
  25557. [ 0.6534, -0.3342, 1.1585, -1.1978, -0.4531, -0.9316, 0.6637, 0.5017],
  25558. [ 0.7875, -0.2731, 1.6926, -0.7453, -0.5681, 0.1508, 0.7731, 0.1326],
  25559. [ 0.4850, -0.4299, 1.6709, -0.0909, -0.1282, -0.1104, 0.3001, 0.3511],
  25560. [ 0.7053, -0.2990, 1.6272, 0.0098, -0.5108, -0.6850, 0.4937, 0.1394],
  25561. [ 0.7281, -0.3108, 1.7384, -0.3913, -0.5842, -0.4616, 0.5438, -0.0194],
  25562. [-2.2420, -2.2402, 1.1993, -1.3444, -0.4029, -1.2837, 0.1392, 0.1897],
  25563. [ 0.4973, -0.4148, 1.5378, -0.5538, -0.5770, -0.0541, 0.4557, 0.2367]],
  25564. device='cuda:0', grad_fn=<AddmmBackward>)
  25565. landmarks are: tensor([[[ 0.5245, -0.4361, 1.5940, -0.2921, -0.5480, -0.0919, 0.2432,
  25566. 0.0502],
  25567. [ 0.5759, -0.3948, 1.2072, -0.9849, -0.4672, -0.8156, 0.4912,
  25568. 0.5701],
  25569. [ 0.5841, -0.4199, 1.7961, -0.6693, -0.6231, 0.1082, 0.6529,
  25570. 0.1159],
  25571. [ 0.5500, -0.4060, 1.7326, 0.0236, -0.1554, -0.0226, 0.1438,
  25572. 0.4171],
  25573. [ 0.5901, -0.4157, 1.7557, 0.1929, -0.5423, -0.5923, 0.3584,
  25574. 0.1698],
  25575. [ 0.5844, -0.4466, 1.8423, -0.1997, -0.5942, -0.3998, 0.4219,
  25576. 0.0467],
  25577. [-2.2859, -2.2859, 1.1436, -1.3082, -0.4672, -1.1620, 0.2256,
  25578. 0.2853],
  25579. [ 0.5027, -0.4289, 1.5478, -0.4229, -0.6346, -0.0303, 0.3223,
  25580. 0.3148]]], device='cuda:0')
  25581. loss_train_step before backward: tensor(0.0100, device='cuda:0', grad_fn=<MseLossBackward>)
  25582. loss_train_step after backward: tensor(0.0100, device='cuda:0', grad_fn=<MseLossBackward>)
  25583. loss_train: 1.5718588447198272
  25584. step: 60
  25585. running loss: 0.026197647411997118
  25586.  
  25587. Train Steps: 60/90 Loss: 0.0262 torch.Size([8, 600, 800])
  25588. torch.Size([8, 8])
  25589. tensor([[ nan, nan, 0.7512, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617],
  25590. [ nan, nan, 0.8363, 0.3317, 0.3563, 0.3367, 0.5329, 0.5142],
  25591. [0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862],
  25592. [ nan, nan, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733],
  25593. [0.6161, 0.4024, 0.8662, 0.4683, 0.4935, 0.5364, 0.6063, 0.5567],
  25594. [0.6164, 0.4076, 0.8838, 0.4117, 0.3713, 0.5550, 0.6238, 0.5350],
  25595. [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413],
  25596. [0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090]],
  25597. device='cuda:0', dtype=torch.float64)
  25598. predictions are: tensor([[-0.9843, -1.3892, 1.0133, -1.2959, -0.4340, -1.2826, 0.1560, 0.2911],
  25599. [-0.4680, -1.0566, 1.4186, -0.9007, -0.7730, -0.6571, 0.1447, 0.1514],
  25600. [ 0.9725, -0.1446, 1.7648, -0.3390, -0.6778, -0.0703, 0.5688, 0.0236],
  25601. [-1.6986, -1.8749, 0.8447, -1.3785, -0.2807, -1.4518, 0.1705, 0.4223],
  25602. [ 0.9834, -0.1479, 1.6662, -0.1350, -0.1114, 0.1979, 0.4884, 0.2888],
  25603. [ 1.0038, -0.0991, 1.7135, -0.3682, -0.5607, 0.4052, 0.5918, 0.2180],
  25604. [ 0.6469, -0.3788, 1.8183, -0.9520, -0.1124, -1.1401, 0.9596, 0.2261],
  25605. [ 1.1452, -0.0534, 1.6800, -0.5457, -0.5165, -0.8915, 0.6578, 0.0302]],
  25606. device='cuda:0', grad_fn=<AddmmBackward>)
  25607. landmarks are: tensor([[[-2.2859, -2.2859, 1.1841, -1.3082, -0.3055, -1.3621, 0.3007,
  25608. 0.3084],
  25609. [-2.2859, -2.2859, 1.5767, -0.7540, -0.6404, -0.7309, 0.1753,
  25610. 0.0893],
  25611. [ 0.6072, -0.4075, 1.8942, -0.3537, -0.5423, -0.1612, 0.6277,
  25612. -0.0400],
  25613. [-2.2859, -2.2859, 0.7106, -1.4468, -0.2882, -1.4237, 0.2430,
  25614. 0.3623],
  25615. [ 0.5598, -0.4273, 1.7152, -0.1227, -0.0065, 0.1917, 0.5143,
  25616. 0.2853],
  25617. [ 0.5614, -0.4032, 1.7961, -0.3844, -0.5711, 0.2776, 0.5952,
  25618. 0.1852],
  25619. [ 0.6108, -0.4201, 1.8711, -0.7848, -0.0053, -1.2236, 1.0362,
  25620. 0.2142],
  25621. [ 0.6165, -0.4249, 1.7961, -0.5384, -0.4499, -0.9695, 0.6401,
  25622. 0.0652]]], device='cuda:0')
  25623. loss_train_step before backward: tensor(0.1461, device='cuda:0', grad_fn=<MseLossBackward>)
  25624. loss_train_step after backward: tensor(0.1461, device='cuda:0', grad_fn=<MseLossBackward>)
  25625. loss_train: 1.7179200714454055
  25626. step: 61
  25627. running loss: 0.028162624122055828
  25628. Train Steps: 61/90 Loss: 0.0282 torch.Size([8, 600, 800])
  25629. torch.Size([8, 8])
  25630. tensor([[0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272],
  25631. [0.6307, 0.4045, 0.8025, 0.5833, 0.3775, 0.4867, 0.6892, 0.5459],
  25632. [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290],
  25633. [0.6226, 0.4103, 0.8575, 0.3450, 0.4388, 0.2067, 0.5787, 0.5383],
  25634. [ nan, nan, 0.7268, 0.2333, 0.4125, 0.1933, 0.5112, 0.5383],
  25635. [0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183],
  25636. [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100],
  25637. [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389]],
  25638. device='cuda:0', dtype=torch.float64)
  25639. predictions are: tensor([[ 0.4069, -0.4887, 1.5738, -0.8161, -0.4371, -0.9213, 0.2029, 0.1648],
  25640. [ 0.6786, -0.3535, 1.4581, 0.0995, -0.4370, 0.0541, 0.8146, 0.2342],
  25641. [ 0.4607, -0.4850, 1.5293, -0.9322, -0.6656, -0.2563, 0.6008, 0.1711],
  25642. [ 0.8734, -0.1844, 1.6444, -0.8083, -0.2172, -1.0457, 0.3793, 0.2408],
  25643. [-2.2036, -2.2297, 1.0053, -1.3418, -0.4457, -1.3178, 0.0063, 0.2084],
  25644. [ 0.6053, -0.4024, 1.7200, -0.1428, -0.5565, -0.3159, 0.4685, 0.1977],
  25645. [ 0.6157, -0.4194, 1.7860, -0.4878, -0.5991, -0.3522, 0.7102, 0.1331],
  25646. [ 0.4235, -0.4903, 1.1394, -1.3845, -0.5097, -0.8972, 0.5451, 0.3152]],
  25647. device='cuda:0', grad_fn=<AddmmBackward>)
  25648. landmarks are: tensor([[[ 0.5711, -0.4015, 1.6979, -0.6770, -0.5365, -1.0619, 0.1712,
  25649. 0.1494],
  25650. [ 0.6273, -0.4177, 1.4208, 0.4085, -0.5423, -0.0380, 0.8973,
  25651. 0.2356],
  25652. [ 0.5697, -0.4538, 1.5308, -0.8703, -0.6572, -0.3639, 0.5739,
  25653. 0.1576],
  25654. [ 0.5898, -0.3908, 1.6748, -0.6924, -0.2594, -1.3313, 0.3873,
  25655. 0.2006],
  25656. [-2.2859, -2.2859, 1.0712, -1.2085, -0.3806, -1.3929, 0.0755,
  25657. 0.2006],
  25658. [ 0.5944, -0.4543, 1.8018, 0.0082, -0.6058, -0.4306, 0.4162,
  25659. 0.1082],
  25660. [ 0.6220, -0.4436, 1.8711, -0.3691, -0.6173, -0.4999, 0.6702,
  25661. 0.0697],
  25662. [ 0.5697, -0.4466, 1.1973, -1.1871, -0.4571, -0.9965, 0.5219,
  25663. 0.2032]]], device='cuda:0')
  25664. loss_train_step before backward: tensor(0.0114, device='cuda:0', grad_fn=<MseLossBackward>)
  25665. loss_train_step after backward: tensor(0.0114, device='cuda:0', grad_fn=<MseLossBackward>)
  25666. loss_train: 1.7292997566983104
  25667. step: 62
  25668. running loss: 0.027891931559650168
  25669. Train Steps: 62/90 Loss: 0.0279 torch.Size([8, 600, 800])
  25670. torch.Size([8, 8])
  25671. tensor([[0.6128, 0.4118, 0.8638, 0.5333, 0.4625, 0.5267, 0.5193, 0.5475],
  25672. [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650],
  25673. [0.6277, 0.4057, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823],
  25674. [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208],
  25675. [0.6228, 0.4119, 0.7938, 0.2233, 0.4674, 0.1773, 0.6188, 0.5433],
  25676. [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383],
  25677. [0.6321, 0.4048, 0.8738, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927],
  25678. [0.6307, 0.4029, 0.8650, 0.5200, 0.3763, 0.4017, 0.7311, 0.5366]],
  25679. device='cuda:0', dtype=torch.float64)
  25680. predictions are: tensor([[ 0.4144, -0.4975, 1.6276, -0.0421, -0.2300, 0.1761, 0.1585, 0.3201],
  25681. [ 0.0341, -0.7305, 1.5266, -0.5993, -0.6648, -0.6568, 0.3565, 0.5090],
  25682. [ 0.5211, -0.4751, 1.5493, -1.3137, -0.4079, -1.2031, 0.6441, 0.0187],
  25683. [ 0.1711, -0.6814, 1.3945, -1.2608, -0.5439, -1.0559, 0.4239, 0.1091],
  25684. [ 0.6617, -0.3810, 1.2957, -1.4371, -0.2348, -1.3179, 0.5682, 0.2448],
  25685. [ 0.4271, -0.5189, 1.6395, -0.2360, -0.3430, 0.0163, 0.2549, 0.3236],
  25686. [ 0.4806, -0.5065, 1.6840, 0.0585, -0.6031, -0.1559, 0.5047, 0.0424],
  25687. [ 0.3668, -0.6003, 1.7427, -0.1002, -0.6017, -0.3306, 0.8768, 0.1167]],
  25688. device='cuda:0', grad_fn=<AddmmBackward>)
  25689. landmarks are: tensor([[[ 5.4475e-01, -3.8383e-01, 1.7037e+00, 1.7752e-01, -1.4965e-01,
  25690. 1.4673e-01, 1.1283e-01, 2.4313e-01],
  25691. [ 5.8942e-01, -3.5027e-01, 1.6402e+00, -3.6135e-01, -5.8268e-01,
  25692. -7.9246e-01, 3.2379e-01, 3.2379e-01],
  25693. [ 6.1316e-01, -4.1224e-01, 1.5478e+00, -1.0619e+00, -2.7090e-01,
  25694. -1.4314e+00, 5.5000e-01, -5.8318e-02],
  25695. [ 5.8851e-01, -4.4288e-01, 1.4266e+00, -9.9261e-01, -4.3834e-01,
  25696. -1.2313e+00, 4.2276e-01, 1.1948e-01],
  25697. [ 5.9076e-01, -3.8322e-01, 1.3804e+00, -1.2543e+00, -1.2695e-01,
  25698. -1.4671e+00, 5.7206e-01, 2.2371e-01],
  25699. [ 5.3603e-01, -4.6490e-01, 1.6517e+00, 4.6189e-04, -2.8245e-01,
  25700. -6.8822e-02, 2.3086e-01, 2.0046e-01],
  25701. [ 6.3355e-01, -4.1617e-01, 1.7499e+00, 3.0839e-01, -4.9607e-01,
  25702. -2.4588e-01, 6.5236e-01, -1.0225e-02],
  25703. [ 6.2730e-01, -4.2490e-01, 1.7095e+00, 1.1594e-01, -5.4804e-01,
  25704. -4.3064e-01, 1.0910e+00, 1.9283e-01]]], device='cuda:0')
  25705. loss_train_step before backward: tensor(0.0269, device='cuda:0', grad_fn=<MseLossBackward>)
  25706. loss_train_step after backward: tensor(0.0269, device='cuda:0', grad_fn=<MseLossBackward>)
  25707. loss_train: 1.7561887232586741
  25708. step: 63
  25709. running loss: 0.027876011480296416
  25710. Train Steps: 63/90 Loss: 0.0279 torch.Size([8, 600, 800])
  25711. torch.Size([8, 8])
  25712. tensor([[0.6144, 0.4032, 0.8563, 0.3283, 0.3525, 0.4200, 0.5775, 0.5583],
  25713. [0.6076, 0.3953, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954],
  25714. [0.6126, 0.4067, 0.8638, 0.5383, 0.4188, 0.4850, 0.5016, 0.5392],
  25715. [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795],
  25716. [0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6262, 0.5167],
  25717. [0.6200, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183],
  25718. [0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283],
  25719. [0.6128, 0.4118, 0.8638, 0.5333, 0.4625, 0.5267, 0.5193, 0.5475]],
  25720. device='cuda:0', dtype=torch.float64)
  25721. predictions are: tensor([[ 0.6315, -0.3615, 1.5528, -0.9199, -0.7058, -0.4619, 0.3850, 0.2241],
  25722. [ 0.3478, -0.5686, 1.4856, -0.6887, -0.6811, -0.5351, 0.2528, 0.0194],
  25723. [ 0.2640, -0.5953, 1.5867, 0.0825, -0.2929, -0.1107, 0.1025, 0.2531],
  25724. [ 0.5650, -0.4369, 1.6092, 0.0533, -0.3942, 0.1878, 1.0112, 0.3344],
  25725. [-0.0204, -0.8140, 1.8506, -0.5604, -0.4599, -0.8181, 0.6399, 0.2778],
  25726. [ 0.4338, -0.5079, 1.4570, -1.0512, -0.5840, -1.0056, 0.5227, 0.0549],
  25727. [ 0.4230, -0.5198, 1.8174, -0.6804, -0.5331, -0.5642, 0.7121, 0.2244],
  25728. [ 0.4962, -0.4371, 1.6423, 0.0260, -0.1258, -0.0331, 0.1809, 0.2964]],
  25729. device='cuda:0', grad_fn=<AddmmBackward>)
  25730. landmarks are: tensor([[[ 0.5520, -0.4237, 1.6691, -0.7694, -0.6577, -0.3460, 0.3815,
  25731. 0.2930],
  25732. [ 0.5206, -0.4603, 1.4670, -0.4768, -0.6577, -0.3998, 0.1836,
  25733. 0.0021],
  25734. [ 0.5436, -0.4076, 1.7037, 0.2006, -0.3517, -0.0457, 0.0311,
  25735. 0.2048],
  25736. [ 0.6355, -0.4080, 1.6113, 0.1852, -0.4730, 0.1467, 0.9996,
  25737. 0.3905],
  25738. [ 0.6094, -0.3947, 1.8885, -0.2998, -0.5769, -0.6770, 0.6067,
  25739. 0.1005],
  25740. [ 0.5781, -0.4129, 1.4142, -0.9057, -0.5115, -0.9937, 0.4621,
  25741. 0.1080],
  25742. [ 0.6073, -0.4058, 1.8885, -0.4999, -0.5942, -0.4768, 0.6413,
  25743. 0.1544],
  25744. [ 0.5447, -0.3838, 1.7037, 0.1775, -0.1497, 0.1467, 0.1128,
  25745. 0.2431]]], device='cuda:0')
  25746. loss_train_step before backward: tensor(0.0206, device='cuda:0', grad_fn=<MseLossBackward>)
  25747. loss_train_step after backward: tensor(0.0206, device='cuda:0', grad_fn=<MseLossBackward>)
  25748. loss_train: 1.7767991842702031
  25749. step: 64
  25750. running loss: 0.027762487254221924
  25751.  
  25752. Train Steps: 64/90 Loss: 0.0278 torch.Size([8, 600, 800])
  25753. torch.Size([8, 8])
  25754. tensor([[0.6264, 0.4049, 0.8988, 0.4633, 0.3813, 0.4983, 0.6326, 0.4843],
  25755. [0.6203, 0.4056, 0.8942, 0.4086, 0.3643, 0.3617, 0.5917, 0.5482],
  25756. [0.6201, 0.3970, 0.8413, 0.4950, 0.4413, 0.5183, 0.6088, 0.5400],
  25757. [0.6182, 0.4058, 0.8738, 0.4350, 0.3563, 0.3400, 0.5290, 0.5822],
  25758. [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092],
  25759. [0.6185, 0.4067, 0.8838, 0.4450, 0.4037, 0.4733, 0.5213, 0.5142],
  25760. [0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496],
  25761. [0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631]],
  25762. device='cuda:0', dtype=torch.float64)
  25763. predictions are: tensor([[ 0.4024, -0.5408, 1.8413, -0.2530, -0.4907, -0.0938, 0.6912, 0.0739],
  25764. [ 0.2597, -0.6268, 1.7614, -0.5272, -0.5981, -0.6706, 0.5482, 0.1919],
  25765. [ 0.5895, -0.4250, 1.6090, -0.0598, -0.2045, -0.0404, 0.6184, 0.3056],
  25766. [ 0.3920, -0.5026, 1.5573, -0.3926, -0.6188, -0.8363, 0.1831, 0.3305],
  25767. [ 0.3983, -0.5696, 1.7242, 0.0177, -0.4467, -0.0459, 0.4527, 0.1228],
  25768. [ 0.2151, -0.6292, 1.6645, -0.2897, -0.3916, -0.1674, 0.1652, 0.1324],
  25769. [ 0.7846, -0.2886, 1.6542, -1.0929, -0.3467, -1.0930, 0.6925, 0.1557],
  25770. [ 0.0871, -0.7089, 1.5407, -0.4913, -0.5601, -0.1321, 0.4519, 0.3129]],
  25771. device='cuda:0', grad_fn=<AddmmBackward>)
  25772. landmarks are: tensor([[[ 6.0751e-01, -4.1586e-01, 1.8654e+00, -1.4580e-01, -5.2494e-01,
  25773. 1.5858e-02, 6.3595e-01, -4.9015e-02],
  25774. [ 5.7910e-01, -4.1270e-01, 1.8442e+00, -3.9854e-01, -6.0306e-01,
  25775. -6.1538e-01, 4.4726e-01, 2.4636e-01],
  25776. [ 5.7812e-01, -4.5219e-01, 1.5998e+00, 4.6189e-04, -2.4781e-01,
  25777. 1.0824e-01, 5.2587e-01, 2.0831e-01],
  25778. [ 5.6969e-01, -4.1132e-01, 1.7499e+00, -2.7667e-01, -6.4042e-01,
  25779. -7.1547e-01, 1.5756e-01, 4.0319e-01],
  25780. [ 5.7829e-01, -4.3634e-01, 1.7724e+00, 4.3211e-02, -5.1531e-01,
  25781. 8.7136e-02, 4.8401e-01, 6.6263e-02],
  25782. [ 5.7079e-01, -4.0747e-01, 1.7961e+00, -2.3048e-01, -4.2102e-01,
  25783. -9.9615e-02, 1.2187e-01, 8.9251e-02],
  25784. [ 5.7904e-01, -4.0308e-01, 1.6915e+00, -9.5640e-01, -4.1518e-01,
  25785. -1.1063e+00, 4.4251e-01, 2.5281e-01],
  25786. [ 5.0266e-01, -4.2895e-01, 1.5478e+00, -4.2294e-01, -6.3464e-01,
  25787. -3.0331e-02, 3.2234e-01, 3.1483e-01]]], device='cuda:0')
  25788. loss_train_step before backward: tensor(0.0183, device='cuda:0', grad_fn=<MseLossBackward>)
  25789. loss_train_step after backward: tensor(0.0183, device='cuda:0', grad_fn=<MseLossBackward>)
  25790. loss_train: 1.7951224902644753
  25791. step: 65
  25792. running loss: 0.02761726908099193
  25793. Train Steps: 65/90 Loss: 0.0276 torch.Size([8, 600, 800])
  25794. torch.Size([8, 8])
  25795. tensor([[0.6200, 0.3999, 0.8653, 0.5207, 0.4100, 0.5125, 0.5975, 0.5103],
  25796. [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550],
  25797. [0.6120, 0.4014, 0.6863, 0.2817, 0.3700, 0.2783, 0.5513, 0.5667],
  25798. [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879],
  25799. [0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901],
  25800. [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050],
  25801. [0.6202, 0.4064, 0.7879, 0.2179, 0.4567, 0.1725, 0.5955, 0.5478],
  25802. [0.6193, 0.3930, 0.8949, 0.4437, 0.3852, 0.5435, 0.6263, 0.5263]],
  25803. device='cuda:0', dtype=torch.float64)
  25804. predictions are: tensor([[ 4.6338e-01, -5.2227e-01, 1.7455e+00, 1.4470e-01, -4.0361e-01,
  25805. -9.9533e-03, 5.7737e-01, 2.0882e-01],
  25806. [ 5.8856e-01, -4.1358e-01, 1.8907e+00, -1.4242e-01, -1.5540e-03,
  25807. -1.9423e-01, 5.4205e-01, 3.0218e-01],
  25808. [ 1.7223e-02, -7.4201e-01, 9.9789e-01, -9.9806e-01, -6.9489e-01,
  25809. -9.9325e-01, 1.8740e-01, 3.3484e-01],
  25810. [ 5.7928e-01, -4.4168e-01, 1.8534e+00, -2.0523e-01, -4.6537e-01,
  25811. 1.8911e-01, 5.8318e-01, 4.3507e-02],
  25812. [ 2.2809e-01, -6.4014e-01, 1.8076e+00, -2.9703e-01, -4.9680e-01,
  25813. -3.3525e-01, 1.6166e-01, 8.1175e-02],
  25814. [ 5.3848e-01, -4.6859e-01, 1.8441e+00, -2.2687e-01, -6.5880e-01,
  25815. -4.8382e-01, 6.1689e-01, 7.4508e-02],
  25816. [ 5.3224e-01, -4.6670e-01, 1.2817e+00, -1.2365e+00, -2.9876e-01,
  25817. -1.5286e+00, 5.0515e-01, 2.2926e-01],
  25818. [ 1.9818e-01, -6.9324e-01, 1.7965e+00, -2.1564e-01, -5.4197e-01,
  25819. 4.7281e-02, 5.9085e-01, 1.7062e-01]], device='cuda:0',
  25820. grad_fn=<AddmmBackward>)
  25821. landmarks are: tensor([[[ 0.5778, -0.4389, 1.7107, 0.1192, -0.3921, 0.0815, 0.4741,
  25822. 0.0711],
  25823. [ 0.5351, -0.4321, 1.7326, -0.1381, 0.0463, 0.1049, 0.4162,
  25824. 0.2776],
  25825. [ 0.5410, -0.4321, 0.8838, -0.9849, -0.5769, -1.0003, 0.2603,
  25826. 0.3315],
  25827. [ 0.5676, -0.4327, 1.8252, -0.2095, -0.4886, 0.3084, 0.4727,
  25828. -0.0322],
  25829. [ 0.5663, -0.3955, 1.7788, -0.2382, -0.4037, -0.2690, 0.0828,
  25830. -0.0220],
  25831. [ 0.5844, -0.4466, 1.8423, -0.1997, -0.5942, -0.3998, 0.4219,
  25832. 0.0467],
  25833. [ 0.5787, -0.4086, 1.3535, -1.2794, -0.1764, -1.4891, 0.4645,
  25834. 0.2442],
  25835. [ 0.5746, -0.4706, 1.8476, -0.2365, -0.5068, 0.2245, 0.6069,
  25836. 0.1449]]], device='cuda:0')
  25837. loss_train_step before backward: tensor(0.0180, device='cuda:0', grad_fn=<MseLossBackward>)
  25838. loss_train_step after backward: tensor(0.0180, device='cuda:0', grad_fn=<MseLossBackward>)
  25839. loss_train: 1.813149274326861
  25840. step: 66
  25841. running loss: 0.027471958701922136
  25842. Train Steps: 66/90 Loss: 0.0275 torch.Size([8, 600, 800])
  25843. torch.Size([8, 8])
  25844. tensor([[0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400],
  25845. [0.6148, 0.3996, 0.8488, 0.3867, 0.3488, 0.4067, 0.5863, 0.5000],
  25846. [ nan, nan, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999],
  25847. [0.6128, 0.4116, 0.8450, 0.4583, 0.3675, 0.4867, 0.5337, 0.5446],
  25848. [0.6268, 0.4061, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300],
  25849. [0.6317, 0.4038, 0.8287, 0.5900, 0.3800, 0.4717, 0.6295, 0.4986],
  25850. [0.6069, 0.3975, 0.8625, 0.5083, 0.4388, 0.5483, 0.5650, 0.4967],
  25851. [0.6357, 0.4118, 0.8400, 0.2500, 0.5413, 0.1633, 0.6725, 0.5586]],
  25852. device='cuda:0', dtype=torch.float64)
  25853. predictions are: tensor([[ 0.6237, -0.3526, 1.4897, -0.8053, -0.2185, -1.1974, 0.4056, 0.2007],
  25854. [ 0.5143, -0.4223, 1.6992, -0.3952, -0.7517, -0.3259, 0.4469, 0.1026],
  25855. [-2.5826, -2.4747, 1.1780, -1.1872, -0.4906, -1.2475, 0.2174, 0.1164],
  25856. [ 0.4828, -0.4180, 1.6138, -0.1066, -0.5749, -0.0056, 0.3014, 0.1830],
  25857. [ 0.1862, -0.6344, 1.5743, -1.1071, -0.3166, -1.1941, 0.6868, 0.2255],
  25858. [ 0.8689, -0.2395, 1.6244, 0.4121, -0.6324, 0.0112, 0.6183, 0.0872],
  25859. [ 0.6781, -0.3274, 1.6930, 0.1251, -0.3152, 0.2715, 0.3076, 0.0937],
  25860. [ 0.8013, -0.2742, 1.6472, -1.0357, 0.1264, -1.3641, 0.7931, 0.2533]],
  25861. device='cuda:0', grad_fn=<AddmmBackward>)
  25862. landmarks are: tensor([[[ 0.5911, -0.3888, 1.4727, -0.9541, -0.0919, -1.4930, 0.3988,
  25863. 0.2083],
  25864. [ 0.5538, -0.4400, 1.6344, -0.4999, -0.6751, -0.4075, 0.4219,
  25865. 0.0236],
  25866. [-2.2859, -2.2859, 1.3400, -1.0388, -0.3055, -1.4930, 0.1157,
  25867. 0.0231],
  25868. [ 0.5444, -0.3846, 1.6171, -0.1689, -0.5885, -0.0380, 0.1791,
  25869. 0.2296],
  25870. [ 0.6093, -0.4104, 1.5709, -1.1620, -0.1727, -1.2313, 0.6471,
  25871. 0.1621],
  25872. [ 0.6320, -0.4206, 1.5420, 0.4393, -0.5307, -0.1073, 0.6216,
  25873. 0.0171],
  25874. [ 0.5174, -0.4497, 1.6979, 0.0620, -0.2594, 0.2468, 0.3238,
  25875. 0.0082],
  25876. [ 0.6504, -0.3840, 1.5940, -1.1312, 0.2141, -1.5315, 0.8205,
  25877. 0.2944]]], device='cuda:0')
  25878. loss_train_step before backward: tensor(0.0157, device='cuda:0', grad_fn=<MseLossBackward>)
  25879. loss_train_step after backward: tensor(0.0157, device='cuda:0', grad_fn=<MseLossBackward>)
  25880. loss_train: 1.8288374664261937
  25881. step: 67
  25882. running loss: 0.027296081588450653
  25883. Train Steps: 67/90 Loss: 0.0273 torch.Size([8, 600, 800])
  25884. torch.Size([8, 8])
  25885. tensor([[0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617],
  25886. [0.6197, 0.4118, 0.8688, 0.5517, 0.4037, 0.5233, 0.5875, 0.5600],
  25887. [0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433],
  25888. [0.6275, 0.4048, 0.8488, 0.2883, 0.4463, 0.2033, 0.6321, 0.5155],
  25889. [0.6231, 0.3973, 0.8650, 0.3950, 0.3625, 0.3183, 0.5837, 0.5167],
  25890. [0.6304, 0.4024, 0.8925, 0.4800, 0.3937, 0.4817, 0.7485, 0.5297],
  25891. [0.6122, 0.3993, 0.8738, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927],
  25892. [0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742]],
  25893. device='cuda:0', dtype=torch.float64)
  25894. predictions are: tensor([[ 0.3651, -0.5206, 1.6997, 0.1684, -0.4600, -0.2442, 0.4703, 0.4270],
  25895. [ 0.7231, -0.3273, 1.6843, 0.2476, -0.3617, 0.2596, 0.3580, 0.2787],
  25896. [ 0.2887, -0.6014, 1.3200, -1.1043, -0.6701, -0.5464, 0.3305, 0.1842],
  25897. [ 0.5762, -0.4481, 1.7215, -0.9292, -0.2008, -1.3391, 0.6507, 0.1040],
  25898. [ 0.5290, -0.4668, 1.7715, -0.5002, -0.5998, -0.8600, 0.4262, 0.1119],
  25899. [ 0.3244, -0.6056, 1.7703, -0.1491, -0.4256, 0.0223, 0.7856, 0.1435],
  25900. [ 0.4005, -0.5445, 1.8250, -0.1427, -0.1946, 0.0466, 0.0948, 0.0051],
  25901. [ 0.4814, -0.5089, 1.8568, 0.1708, -0.5595, -0.5359, 0.4954, -0.0020]],
  25902. device='cuda:0', grad_fn=<AddmmBackward>)
  25903. landmarks are: tensor([[[ 0.5897, -0.3527, 1.8018, 0.2545, -0.5307, -0.3229, 0.3122,
  25904. 0.3084],
  25905. [ 0.5762, -0.3840, 1.7268, 0.2622, -0.4210, 0.1313, 0.4277,
  25906. 0.3007],
  25907. [ 0.5665, -0.3963, 1.3457, -0.9926, -0.6058, -0.6308, 0.5721,
  25908. 0.2237],
  25909. [ 0.6126, -0.4161, 1.6344, -0.9541, -0.2247, -1.3467, 0.6339,
  25910. 0.0953],
  25911. [ 0.5924, -0.4507, 1.7095, -0.4614, -0.6115, -0.8156, 0.4104,
  25912. 0.1005],
  25913. [ 0.6257, -0.4273, 1.8365, -0.0688, -0.4672, -0.0611, 1.1715,
  25914. 0.1608],
  25915. [ 0.5417, -0.4417, 1.7499, -0.1304, -0.1994, -0.0324, 0.0951,
  25916. -0.0099],
  25917. [ 0.6113, -0.4057, 1.7499, 0.3007, -0.5885, -0.5384, 0.4513,
  25918. -0.0957]]], device='cuda:0')
  25919. loss_train_step before backward: tensor(0.0129, device='cuda:0', grad_fn=<MseLossBackward>)
  25920. loss_train_step after backward: tensor(0.0129, device='cuda:0', grad_fn=<MseLossBackward>)
  25921. loss_train: 1.8417663387954235
  25922. step: 68
  25923. running loss: 0.027084799099932697
  25924.  
  25925. Train Steps: 68/90 Loss: 0.0271 torch.Size([8, 600, 800])
  25926. torch.Size([8, 8])
  25927. tensor([[0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6188, 0.5433],
  25928. [0.6250, 0.4013, 0.8525, 0.5417, 0.4037, 0.5117, 0.6325, 0.5017],
  25929. [0.6261, 0.4029, 0.8720, 0.3364, 0.3665, 0.3753, 0.6531, 0.5183],
  25930. [0.6265, 0.4071, 0.8875, 0.3367, 0.3975, 0.3350, 0.6312, 0.5250],
  25931. [0.6129, 0.4069, 0.8750, 0.5067, 0.3875, 0.4233, 0.5235, 0.5881],
  25932. [0.6346, 0.4144, 0.9088, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899],
  25933. [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646],
  25934. [0.6260, 0.4214, 0.8538, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983]],
  25935. device='cuda:0', dtype=torch.float64)
  25936. predictions are: tensor([[ 0.7885, -0.2755, 1.8507, -0.0734, -0.4111, 0.0235, 0.3560, 0.0462],
  25937. [ 0.7404, -0.3294, 1.7126, 0.2041, -0.3258, 0.0337, 0.6419, -0.0222],
  25938. [ 0.9415, -0.2172, 1.7814, -0.6563, -0.5288, -0.5494, 0.6929, 0.0388],
  25939. [ 0.4125, -0.5422, 1.9219, -0.7111, -0.3666, -0.8003, 0.6918, 0.0999],
  25940. [ 0.6063, -0.3980, 1.7486, 0.0841, -0.4291, -0.2599, 0.1080, 0.2190],
  25941. [ 0.3537, -0.5790, 1.9364, -0.1108, -0.2992, -0.2235, 0.9382, 0.2877],
  25942. [-1.0781, -1.4772, 1.3764, -0.8284, -0.6005, -0.8796, -0.0899, 0.0682],
  25943. [ 0.6193, -0.3623, 1.5289, 0.2866, -0.4413, -0.5597, 0.1839, 0.4190]],
  25944. device='cuda:0', grad_fn=<AddmmBackward>)
  25945. landmarks are: tensor([[[ 0.5773, -0.4105, 1.8192, -0.1304, -0.5307, 0.0467, 0.5721,
  25946. 0.2237],
  25947. [ 0.6010, -0.4321, 1.6517, 0.2160, -0.4210, 0.0774, 0.6356,
  25948. 0.0313],
  25949. [ 0.6059, -0.4249, 1.7420, -0.7321, -0.5931, -0.5523, 0.7307,
  25950. 0.1080],
  25951. [ 0.6078, -0.4057, 1.8134, -0.7309, -0.4499, -0.7386, 0.6298,
  25952. 0.1390],
  25953. [ 0.5452, -0.4067, 1.7557, 0.0543, -0.4961, -0.3306, 0.1323,
  25954. 0.4306],
  25955. [ 0.6454, -0.3719, 1.9115, -0.1304, -0.5076, -0.2844, 1.0033,
  25956. 0.4386],
  25957. [ 0.5483, -0.4105, 1.4208, -0.8002, -0.6000, -0.9002, 0.0511,
  25958. 0.3220],
  25959. [ 0.6055, -0.3393, 1.6575, 0.2545, -0.5942, -0.5461, 0.2949,
  25960. 0.4778]]], device='cuda:0')
  25961. loss_train_step before backward: tensor(0.0728, device='cuda:0', grad_fn=<MseLossBackward>)
  25962. loss_train_step after backward: tensor(0.0728, device='cuda:0', grad_fn=<MseLossBackward>)
  25963. loss_train: 1.9145381338894367
  25964. step: 69
  25965. running loss: 0.027746929476658504
  25966. Train Steps: 69/90 Loss: 0.0277 torch.Size([8, 600, 800])
  25967. torch.Size([8, 8])
  25968. tensor([[0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317],
  25969. [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301],
  25970. [0.6200, 0.4118, 0.8287, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799],
  25971. [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749],
  25972. [0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835],
  25973. [0.6126, 0.3954, 0.8538, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350],
  25974. [0.6263, 0.4065, 0.9038, 0.4317, 0.3588, 0.4550, 0.6325, 0.5250],
  25975. [0.6246, 0.4028, 0.8738, 0.4867, 0.4088, 0.5667, 0.6362, 0.5200]],
  25976. device='cuda:0', dtype=torch.float64)
  25977. predictions are: tensor([[-0.1347, -0.8905, 1.3869, -1.0465, -0.3008, -1.1903, 0.5167, 0.1891],
  25978. [ 0.5789, -0.4325, 1.7945, 0.1725, -0.3334, 0.1157, 0.6113, 0.0700],
  25979. [ 0.1578, -0.6654, 1.5865, -0.2879, -0.4701, -0.9997, 0.2556, 0.2780],
  25980. [ 0.3835, -0.5182, 1.2272, -0.7243, -0.6110, -0.6875, 0.1796, 0.2398],
  25981. [ 0.6624, -0.3772, 2.0066, -0.3205, -0.2315, -1.2892, 0.5583, 0.0233],
  25982. [ 1.0005, -0.1530, 1.7703, 0.2153, -0.3215, -0.0169, 0.2178, 0.0890],
  25983. [ 0.5081, -0.4877, 1.9618, -0.1362, -0.5452, -0.0896, 0.5822, 0.0863],
  25984. [ 0.6206, -0.3973, 1.8513, 0.1067, -0.3296, 0.4381, 0.6588, 0.0824]],
  25985. device='cuda:0', grad_fn=<AddmmBackward>)
  25986. landmarks are: tensor([[[ 0.5849, -0.3836, 1.2649, -1.2236, -0.3460, -1.2313, 0.4508,
  25987. 0.1698],
  25988. [ 0.6010, -0.4562, 1.7198, -0.0090, -0.3464, 0.0108, 0.6294,
  25989. 0.1627],
  25990. [ 0.5779, -0.3840, 1.5420, -0.4306, -0.5423, -0.9772, 0.2041,
  25991. 0.3928],
  25992. [ 0.5419, -0.4160, 1.1810, -0.8939, -0.6808, -0.7463, 0.2485,
  25993. 0.3694],
  25994. [ 0.6117, -0.3877, 1.8249, -0.5230, -0.3517, -1.2851, 0.4200,
  25995. -0.0529],
  25996. [ 0.5435, -0.4597, 1.6575, 0.0159, -0.3229, -0.1150, 0.1877,
  25997. 0.1854],
  25998. [ 0.6070, -0.4085, 1.8885, -0.2921, -0.6289, -0.1843, 0.6356,
  25999. 0.1390],
  26000. [ 0.5990, -0.4256, 1.7499, -0.0380, -0.3979, 0.3315, 0.6529,
  26001. 0.1159]]], device='cuda:0')
  26002. loss_train_step before backward: tensor(0.0296, device='cuda:0', grad_fn=<MseLossBackward>)
  26003. loss_train_step after backward: tensor(0.0296, device='cuda:0', grad_fn=<MseLossBackward>)
  26004. loss_train: 1.9441209603101015
  26005. step: 70
  26006. running loss: 0.027773156575858594
  26007. Train Steps: 70/90 Loss: 0.0278 torch.Size([8, 600, 800])
  26008. torch.Size([8, 8])
  26009. tensor([[0.6110, 0.3984, 0.8750, 0.4933, 0.4625, 0.4950, 0.5578, 0.5676],
  26010. [0.6135, 0.3994, 0.7913, 0.3050, 0.3625, 0.3050, 0.5837, 0.5050],
  26011. [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117],
  26012. [0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204],
  26013. [0.6197, 0.4091, 0.8800, 0.4783, 0.3538, 0.4767, 0.5950, 0.5550],
  26014. [0.6161, 0.4024, 0.8662, 0.4683, 0.4935, 0.5364, 0.6063, 0.5567],
  26015. [0.6102, 0.3999, 0.8750, 0.5133, 0.3825, 0.4750, 0.5637, 0.5083],
  26016. [0.6125, 0.3999, 0.8750, 0.4883, 0.4750, 0.4700, 0.5533, 0.5617]],
  26017. device='cuda:0', dtype=torch.float64)
  26018. predictions are: tensor([[ 0.6819, -0.3631, 1.7929, 0.0338, -0.1769, -0.1166, 0.2615, 0.2761],
  26019. [ 0.2571, -0.6269, 1.4967, -0.7880, -0.6286, -0.9278, 0.4804, 0.0170],
  26020. [ 0.1597, -0.7106, 1.8510, -0.6569, -0.5832, -0.9929, 0.5346, 0.0192],
  26021. [ 0.6173, -0.4197, 1.6777, 0.3250, -0.4016, 0.1249, 0.7606, 0.0746],
  26022. [ 0.7167, -0.3175, 1.7641, -0.0079, -0.6089, -0.2013, 0.5217, 0.2568],
  26023. [ 0.8350, -0.2456, 1.8566, -0.0243, -0.0949, -0.0249, 0.4935, 0.2472],
  26024. [ 0.5563, -0.4414, 1.8026, 0.2169, -0.5440, -0.1765, 0.2853, 0.0218],
  26025. [ 0.5862, -0.4149, 1.7682, -0.0044, -0.1104, -0.1928, 0.2206, 0.2828]],
  26026. device='cuda:0', grad_fn=<AddmmBackward>)
  26027. landmarks are: tensor([[[ 5.3637e-01, -4.4573e-01, 1.7557e+00, -7.2363e-03, -1.4965e-01,
  26028. 4.6189e-04, 2.9049e-01, 3.3573e-01],
  26029. [ 5.4769e-01, -4.4126e-01, 1.3688e+00, -8.7714e-01, -6.1155e-01,
  26030. -8.7714e-01, 4.1039e-01, 4.6651e-02],
  26031. [ 5.7991e-01, -4.3295e-01, 1.7210e+00, -7.6936e-01, -5.7113e-01,
  26032. -8.7714e-01, 3.9885e-01, 7.7444e-02],
  26033. [ 6.0425e-01, -4.2731e-01, 1.7198e+00, 2.1845e-01, -3.4783e-01,
  26034. 1.1492e-01, 8.0616e-01, 1.1755e-01],
  26035. [ 5.7633e-01, -3.9630e-01, 1.7788e+00, -7.6520e-02, -6.5196e-01,
  26036. -8.4219e-02, 4.6236e-01, 2.7760e-01],
  26037. [ 5.5978e-01, -4.2731e-01, 1.7152e+00, -1.2271e-01, -6.4698e-03,
  26038. 1.9169e-01, 5.1432e-01, 2.8530e-01],
  26039. [ 5.3262e-01, -4.3895e-01, 1.7557e+00, 8.5142e-02, -5.1917e-01,
  26040. -9.1917e-02, 3.1801e-01, 6.2048e-02],
  26041. [ 5.4319e-01, -4.3880e-01, 1.7557e+00, -3.0331e-02, -9.1917e-02,
  26042. -1.1501e-01, 2.6993e-01, 3.0867e-01]]], device='cuda:0')
  26043. loss_train_step before backward: tensor(0.0130, device='cuda:0', grad_fn=<MseLossBackward>)
  26044. loss_train_step after backward: tensor(0.0130, device='cuda:0', grad_fn=<MseLossBackward>)
  26045. loss_train: 1.9570787139236927
  26046. step: 71
  26047. running loss: 0.027564488928502714
  26048. Train Steps: 71/90 Loss: 0.0276 torch.Size([8, 600, 800])
  26049. torch.Size([8, 8])
  26050. tensor([[0.6201, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044],
  26051. [0.6102, 0.4020, 0.8638, 0.3717, 0.3625, 0.5017, 0.6038, 0.5500],
  26052. [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600],
  26053. [0.6219, 0.3934, 0.8688, 0.5267, 0.4313, 0.4967, 0.5988, 0.4983],
  26054. [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901],
  26055. [0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272],
  26056. [0.6197, 0.3986, 0.8800, 0.4617, 0.4188, 0.4783, 0.5687, 0.5550],
  26057. [0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183]],
  26058. device='cuda:0', dtype=torch.float64)
  26059. predictions are: tensor([[ 0.2929, -0.6306, 1.3494, -1.0974, -0.2760, -1.2471, 0.5076, 0.1783],
  26060. [ 0.5876, -0.4227, 1.8131, -0.3082, -0.5716, 0.0681, 0.6366, 0.1759],
  26061. [ 0.4308, -0.4823, 1.2201, -0.8724, -0.5480, -0.7434, 0.4204, 0.2808],
  26062. [ 0.5981, -0.4646, 1.8456, 0.3831, -0.3125, 0.1098, 0.6071, -0.0576],
  26063. [ 0.3022, -0.6097, 1.5507, -0.6316, -0.4949, -1.1007, 0.2263, 0.1064],
  26064. [ 0.7622, -0.3001, 1.7877, -0.3726, -0.3335, -1.0181, 0.2477, 0.1684],
  26065. [ 0.6255, -0.4040, 1.8410, 0.0971, -0.3022, 0.0756, 0.4581, 0.1748],
  26066. [ 0.5553, -0.4920, 1.8473, 0.2026, -0.5247, -0.4048, 0.5851, 0.1129]],
  26067. device='cuda:0', grad_fn=<AddmmBackward>)
  26068. landmarks are: tensor([[[ 0.5785, -0.4129, 1.2785, -1.3996, -0.3227, -1.3259, 0.4258,
  26069. 0.0438],
  26070. [ 0.5324, -0.4293, 1.7037, -0.5692, -0.6115, 0.0313, 0.5028,
  26071. 0.2545],
  26072. [ 0.5466, -0.4080, 1.0668, -1.1764, -0.6289, -0.7617, 0.4855,
  26073. 0.3007],
  26074. [ 0.5864, -0.4690, 1.7268, 0.1467, -0.2940, 0.0082, 0.4797,
  26075. 0.0159],
  26076. [ 0.5515, -0.4129, 1.4785, -1.0080, -0.5192, -1.1004, 0.1034,
  26077. -0.0220],
  26078. [ 0.5711, -0.4015, 1.6979, -0.6770, -0.5365, -1.0619, 0.1712,
  26079. 0.1494],
  26080. [ 0.5763, -0.4450, 1.7788, -0.1535, -0.3517, -0.0765, 0.3411,
  26081. 0.2776],
  26082. [ 0.5944, -0.4543, 1.8018, 0.0082, -0.6058, -0.4306, 0.4162,
  26083. 0.1082]]], device='cuda:0')
  26084. loss_train_step before backward: tensor(0.0202, device='cuda:0', grad_fn=<MseLossBackward>)
  26085. loss_train_step after backward: tensor(0.0202, device='cuda:0', grad_fn=<MseLossBackward>)
  26086. loss_train: 1.9772933050990105
  26087. step: 72
  26088. running loss: 0.027462407015264034
  26089.  
  26090. Train Steps: 72/90 Loss: 0.0275 torch.Size([8, 600, 800])
  26091. torch.Size([8, 8])
  26092. tensor([[0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750],
  26093. [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283],
  26094. [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5787, 0.5550],
  26095. [0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240],
  26096. [ nan, nan, 0.6688, 0.2513, 0.4113, 0.2117, 0.5193, 0.5933],
  26097. [0.6204, 0.4049, 0.7975, 0.2700, 0.3937, 0.2567, 0.5700, 0.5183],
  26098. [0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431],
  26099. [0.6189, 0.4029, 0.8375, 0.5767, 0.4745, 0.4829, 0.5551, 0.5598]],
  26100. device='cuda:0', dtype=torch.float64)
  26101. predictions are: tensor([[ 6.5583e-01, -3.7811e-01, 1.2696e+00, -7.9125e-01, -4.5153e-01,
  26102. -8.8227e-01, 6.5653e-01, 3.0809e-01],
  26103. [ 9.5700e-01, -1.8838e-01, 2.0054e+00, -3.3127e-02, -6.0763e-01,
  26104. -4.2334e-01, 6.2569e-01, 8.3709e-02],
  26105. [ 9.3682e-01, -2.0342e-01, 1.6857e+00, -6.1299e-01, -6.7221e-01,
  26106. -5.1084e-01, 4.4851e-01, 6.2466e-02],
  26107. [ 2.5737e-01, -6.2742e-01, 1.3201e+00, -9.9313e-01, -2.6889e-01,
  26108. -1.3625e+00, 3.5545e-01, 1.6475e-01],
  26109. [-1.3298e+00, -1.6522e+00, 1.0108e+00, -9.3262e-01, -3.8892e-01,
  26110. -1.2307e+00, 3.4897e-01, 3.3487e-01],
  26111. [ 3.9808e-01, -5.5571e-01, 1.6143e+00, -7.6801e-01, -4.6551e-01,
  26112. -1.0459e+00, 3.7800e-01, 6.5839e-02],
  26113. [ 6.9094e-01, -3.6286e-01, 1.8962e+00, 1.8539e-01, -2.5873e-01,
  26114. 7.2056e-02, 2.3904e-01, 1.9110e-02],
  26115. [ 6.2867e-01, -4.1586e-01, 1.7483e+00, 5.4881e-01, -1.1541e-01,
  26116. -1.4957e-03, 4.3051e-01, 1.5835e-01]], device='cuda:0',
  26117. grad_fn=<AddmmBackward>)
  26118. landmarks are: tensor([[[ 0.5805, -0.3818, 1.0282, -1.1774, -0.4903, -0.9310, 0.5894,
  26119. 0.3700],
  26120. [ 0.5880, -0.3887, 1.8423, -0.3306, -0.6231, -0.5230, 0.4046,
  26121. 0.1544],
  26122. [ 0.5647, -0.4129, 1.4901, -1.0619, -0.6462, -0.5846, 0.3873,
  26123. 0.2776],
  26124. [ 0.5676, -0.4112, 1.1898, -1.2467, -0.2940, -1.4622, 0.2103,
  26125. 0.1343],
  26126. [-2.2859, -2.2859, 0.8033, -1.1250, -0.3864, -1.3082, 0.1126,
  26127. 0.4543],
  26128. [ 0.5798, -0.4156, 1.3977, -1.0388, -0.4672, -1.1004, 0.3469,
  26129. 0.1082],
  26130. [ 0.5351, -0.4153, 1.7326, -0.0457, -0.2214, -0.0466, 0.0434,
  26131. 0.2228],
  26132. [ 0.5726, -0.4249, 1.5824, 0.3777, -0.0942, -0.0556, 0.2782,
  26133. 0.2997]]], device='cuda:0')
  26134. loss_train_step before backward: tensor(0.0512, device='cuda:0', grad_fn=<MseLossBackward>)
  26135. loss_train_step after backward: tensor(0.0512, device='cuda:0', grad_fn=<MseLossBackward>)
  26136. loss_train: 2.028471328318119
  26137. step: 73
  26138. running loss: 0.02778727847011122
  26139. Train Steps: 73/90 Loss: 0.0278 torch.Size([8, 600, 800])
  26140. torch.Size([8, 8])
  26141. tensor([[0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459],
  26142. [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617],
  26143. [ nan, nan, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733],
  26144. [0.6087, 0.3976, 0.8337, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300],
  26145. [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578],
  26146. [0.6126, 0.3954, 0.8538, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350],
  26147. [0.6293, 0.4097, 0.8800, 0.2517, 0.5262, 0.2600, 0.7430, 0.5378],
  26148. [0.6185, 0.4067, 0.8838, 0.4450, 0.4037, 0.4733, 0.5213, 0.5142]],
  26149. device='cuda:0', dtype=torch.float64)
  26150. predictions are: tensor([[ 0.6416, -0.3906, 1.7688, -0.7565, -0.0875, -1.1663, 0.7043, 0.1075],
  26151. [ 0.7966, -0.2625, 1.7855, 0.4262, -0.6310, -0.1006, 0.5265, 0.2256],
  26152. [-1.6472, -1.8579, 1.0097, -1.0127, -0.2529, -1.2835, 0.3738, 0.4171],
  26153. [ 0.6535, -0.3692, 1.6415, -0.3841, -0.6620, -0.6364, 0.3850, 0.0625],
  26154. [ 0.6294, -0.3894, 1.0214, -1.1424, -0.4673, -1.1836, 0.2044, 0.2154],
  26155. [ 0.6398, -0.3967, 1.7436, 0.1655, -0.4967, 0.0448, 0.2775, 0.0594],
  26156. [ 0.5993, -0.4317, 1.8031, -0.9034, 0.0458, -1.0696, 0.9169, 0.1342],
  26157. [ 0.7419, -0.3240, 1.7888, 0.0059, -0.5461, 0.0415, 0.1476, 0.0034]],
  26158. device='cuda:0', grad_fn=<AddmmBackward>)
  26159. landmarks are: tensor([[[ 0.6520, -0.3912, 1.7095, -0.9079, -0.0284, -1.3621, 0.8096,
  26160. 0.2356],
  26161. [ 0.5897, -0.3527, 1.8018, 0.2545, -0.5307, -0.3229, 0.3122,
  26162. 0.3084],
  26163. [-2.2859, -2.2859, 0.7106, -1.4468, -0.2882, -1.4237, 0.2430,
  26164. 0.3623],
  26165. [ 0.5255, -0.4495, 1.5651, -0.4999, -0.5711, -0.8463, 0.4566,
  26166. 0.1621],
  26167. [ 0.5232, -0.4189, 0.9358, -1.4006, -0.3921, -1.3698, 0.2555,
  26168. 0.2906],
  26169. [ 0.5435, -0.4597, 1.6575, 0.0159, -0.3229, -0.1150, 0.1877,
  26170. 0.1854],
  26171. [ 0.6207, -0.3936, 1.7788, -1.1235, 0.1448, -1.0850, 1.1459,
  26172. 0.1982],
  26173. [ 0.5708, -0.4075, 1.7961, -0.2305, -0.4210, -0.0996, 0.1219,
  26174. 0.0893]]], device='cuda:0')
  26175. loss_train_step before backward: tensor(0.0284, device='cuda:0', grad_fn=<MseLossBackward>)
  26176. loss_train_step after backward: tensor(0.0284, device='cuda:0', grad_fn=<MseLossBackward>)
  26177. loss_train: 2.056894328445196
  26178. step: 74
  26179. running loss: 0.027795869303313462
  26180. Train Steps: 74/90 Loss: 0.0278 torch.Size([8, 600, 800])
  26181. torch.Size([8, 8])
  26182. tensor([[0.6286, 0.4086, 0.8408, 0.2801, 0.4163, 0.2800, 0.6725, 0.5393],
  26183. [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320],
  26184. [0.6159, 0.4085, 0.6900, 0.2283, 0.4088, 0.1950, 0.5123, 0.5397],
  26185. [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524],
  26186. [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646],
  26187. [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285],
  26188. [0.6251, 0.4163, 0.8662, 0.4467, 0.3625, 0.3567, 0.6038, 0.5533],
  26189. [0.6127, 0.4084, 0.8700, 0.4467, 0.3987, 0.4317, 0.5013, 0.5471]],
  26190. device='cuda:0', dtype=torch.float64)
  26191. predictions are: tensor([[ 6.8721e-01, -3.6603e-01, 1.6273e+00, -9.3910e-01, -2.3020e-01,
  26192. -1.0608e+00, 8.1830e-01, 2.8156e-01],
  26193. [ 2.6261e-01, -6.6006e-01, 1.7028e+00, 6.9278e-02, -5.0285e-01,
  26194. 6.1316e-02, 9.1670e-01, 9.0216e-02],
  26195. [ 7.8071e-02, -7.5531e-01, 9.7520e-01, -1.1406e+00, -3.2744e-01,
  26196. -1.3896e+00, 1.0609e-01, 2.5919e-01],
  26197. [ 6.7547e-01, -3.5546e-01, 1.7788e+00, -4.9031e-02, -5.2201e-01,
  26198. -2.5553e-01, 2.1430e-01, 1.6061e-01],
  26199. [ 9.4422e-01, -1.8775e-01, 1.4252e+00, -7.5169e-01, -5.3062e-01,
  26200. -8.7556e-01, -3.1819e-02, 1.9832e-01],
  26201. [ 4.0847e-01, -5.7434e-01, 1.6866e+00, 2.0950e-01, -4.9814e-01,
  26202. 1.7580e-03, 7.6417e-01, 9.1559e-02],
  26203. [ 6.3641e-01, -3.8298e-01, 1.7680e+00, -2.8971e-01, -5.1222e-01,
  26204. -7.0556e-01, 4.5485e-01, 2.9442e-01],
  26205. [ 4.7471e-01, -4.8755e-01, 1.7431e+00, -2.4426e-01, -4.8224e-01,
  26206. -3.1786e-01, 1.0535e-01, 1.8711e-01]], device='cuda:0',
  26207. grad_fn=<AddmmBackward>)
  26208. landmarks are: tensor([[[ 6.1742e-01, -3.9842e-01, 1.5975e+00, -9.9214e-01, -3.6328e-01,
  26209. -9.9261e-01, 8.2047e-01, 2.0505e-01],
  26210. [ 6.2730e-01, -4.3934e-01, 1.6402e+00, 1.3133e-01, -5.0762e-01,
  26211. 4.6651e-02, 1.1532e+00, 1.7146e-01],
  26212. [ 5.5912e-01, -3.9900e-01, 9.0115e-01, -1.2313e+00, -3.9792e-01,
  26213. -1.3852e+00, 8.0445e-02, 2.0706e-01],
  26214. [ 5.5289e-01, -3.8106e-01, 1.7788e+00, -3.8029e-02, -5.3072e-01,
  26215. -2.0739e-01, 7.2734e-02, 2.6568e-01],
  26216. [ 5.4825e-01, -4.1045e-01, 1.4208e+00, -8.0015e-01, -6.0000e-01,
  26217. -9.0023e-01, 5.1142e-02, 3.2204e-01],
  26218. [ 6.5365e-01, -3.9601e-01, 1.6517e+00, 3.1609e-01, -4.9607e-01,
  26219. 4.6189e-04, 7.6203e-01, 1.5543e-01],
  26220. [ 6.0162e-01, -3.6328e-01, 1.7152e+00, -2.2279e-01, -6.1155e-01,
  26221. -6.3849e-01, 5.0277e-01, 2.6990e-01],
  26222. [ 5.4405e-01, -3.9969e-01, 1.7326e+00, -2.2279e-01, -4.4411e-01,
  26223. -2.9207e-01, 2.9551e-02, 2.4088e-01]]], device='cuda:0')
  26224. loss_train_step before backward: tensor(0.0174, device='cuda:0', grad_fn=<MseLossBackward>)
  26225. loss_train_step after backward: tensor(0.0174, device='cuda:0', grad_fn=<MseLossBackward>)
  26226. loss_train: 2.0743119940161705
  26227. step: 75
  26228. running loss: 0.02765749325354894
  26229.  
  26230. Train Steps: 75/90 Loss: 0.0277 torch.Size([8, 600, 800])
  26231. torch.Size([8, 8])
  26232. tensor([[0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835],
  26233. [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611],
  26234. [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377],
  26235. [0.6212, 0.4171, 0.7875, 0.3633, 0.3813, 0.2933, 0.5675, 0.5700],
  26236. [0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385],
  26237. [0.6262, 0.4163, 0.8850, 0.5183, 0.3763, 0.4150, 0.6025, 0.5500],
  26238. [ nan, nan, 0.9088, 0.3783, 0.4562, 0.2617, 0.6741, 0.5575],
  26239. [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892]],
  26240. device='cuda:0', dtype=torch.float64)
  26241. predictions are: tensor([[ 0.8649, -0.2485, 1.7574, -0.6191, -0.2934, -1.1843, 0.4251, 0.1032],
  26242. [ 0.4361, -0.5190, 1.6486, -0.2521, -0.3616, -0.0664, 0.2143, 0.2285],
  26243. [ 0.3731, -0.5932, 1.5713, 0.0356, -0.5071, -0.1149, 0.5145, 0.1103],
  26244. [ 0.4553, -0.4569, 1.2745, -0.7047, -0.4934, -0.8180, 0.3275, 0.3119],
  26245. [ 0.7047, -0.3546, 1.3248, -1.2568, -0.2380, -1.1344, 0.7567, 0.2289],
  26246. [ 0.6938, -0.3485, 1.6413, -0.0666, -0.5683, -0.3663, 0.4046, 0.1990],
  26247. [-0.3272, -1.0349, 1.8241, -0.5274, -0.2300, -0.9745, 0.7401, 0.2870],
  26248. [ 0.7128, -0.3070, 1.5699, -0.2741, -0.6330, -0.5391, 0.0100, 0.2550]],
  26249. device='cuda:0', grad_fn=<AddmmBackward>)
  26250. landmarks are: tensor([[[ 0.6117, -0.3877, 1.8249, -0.5230, -0.3517, -1.2851, 0.4200,
  26251. -0.0529],
  26252. [ 0.5491, -0.4290, 1.7788, -0.1073, -0.2651, -0.1073, 0.2555,
  26253. 0.3057],
  26254. [ 0.5775, -0.4607, 1.6741, 0.1962, -0.4036, -0.1212, 0.4588,
  26255. 0.1979],
  26256. [ 0.5832, -0.3593, 1.3515, -0.6077, -0.5249, -0.9310, 0.3353,
  26257. 0.3469],
  26258. [ 0.6174, -0.3936, 1.4586, -1.1709, -0.2420, -1.1389, 0.8296,
  26259. 0.2012],
  26260. [ 0.6066, -0.3632, 1.8018, 0.1082, -0.5480, -0.3691, 0.4970,
  26261. 0.2545],
  26262. [-2.2859, -2.2859, 1.9115, -0.5384, -0.1785, -1.0773, 0.8278,
  26263. 0.2890],
  26264. [ 0.5454, -0.4053, 1.6633, -0.1766, -0.6058, -0.5923, 0.1577,
  26265. 0.4357]]], device='cuda:0')
  26266. loss_train_step before backward: tensor(0.0945, device='cuda:0', grad_fn=<MseLossBackward>)
  26267. loss_train_step after backward: tensor(0.0945, device='cuda:0', grad_fn=<MseLossBackward>)
  26268. loss_train: 2.168797843158245
  26269. step: 76
  26270. running loss: 0.028536813725766382
  26271. Train Steps: 76/90 Loss: 0.0285 torch.Size([8, 600, 800])
  26272. torch.Size([8, 8])
  26273. tensor([[0.6245, 0.4115, 0.8700, 0.4883, 0.4625, 0.5517, 0.6100, 0.5217],
  26274. [0.6154, 0.4112, 0.7037, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600],
  26275. [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167],
  26276. [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391],
  26277. [0.6250, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6088, 0.5183],
  26278. [0.6197, 0.4090, 0.7825, 0.2500, 0.4200, 0.2483, 0.5988, 0.5667],
  26279. [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150],
  26280. [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100]],
  26281. device='cuda:0', dtype=torch.float64)
  26282. predictions are: tensor([[ 0.5790, -0.4246, 1.7650, -0.0211, -0.2898, 0.0674, 0.4351, 0.2170],
  26283. [ 0.7075, -0.3480, 1.0163, -1.1494, -0.3797, -1.5016, 0.1140, 0.3749],
  26284. [ 0.5688, -0.4119, 1.2305, -0.8478, -0.4889, -1.1087, 0.3657, 0.4492],
  26285. [ 0.5267, -0.4787, 1.4408, -0.9947, -0.4450, -1.2665, 0.3767, 0.1211],
  26286. [ 0.5472, -0.4517, 1.7947, -0.1033, -0.2681, 0.1028, 0.4809, 0.2489],
  26287. [ 0.4713, -0.5011, 1.2762, -1.1014, -0.4127, -1.2061, 0.4065, 0.3346],
  26288. [ 0.3543, -0.5869, 1.7987, -0.0647, -0.3814, 0.2580, 0.5179, 0.1634],
  26289. [ 0.3446, -0.6577, 1.8439, -0.3129, -0.6954, -0.6245, 0.6689, 0.0432]],
  26290. device='cuda:0', grad_fn=<AddmmBackward>)
  26291. landmarks are: tensor([[[ 0.5987, -0.3852, 1.7326, -0.0303, -0.1497, 0.2622, 0.5316,
  26292. 0.1236],
  26293. [ 0.5565, -0.3864, 0.9647, -1.2159, -0.3286, -1.4391, 0.1852,
  26294. 0.3007],
  26295. [ 0.5631, -0.4129, 1.2129, -0.9233, -0.4152, -1.0311, 0.4566,
  26296. 0.5624],
  26297. [ 0.5785, -0.4259, 1.4228, -1.0261, -0.4190, -1.2189, 0.4763,
  26298. 0.2043],
  26299. [ 0.6012, -0.3846, 1.7326, -0.0457, -0.1381, 0.2853, 0.5259,
  26300. 0.1082],
  26301. [ 0.5764, -0.3969, 1.3284, -1.1312, -0.3460, -1.1389, 0.4797,
  26302. 0.3315],
  26303. [ 0.6014, -0.3883, 1.8192, -0.1150, -0.2882, 0.4008, 0.5201,
  26304. 0.0928],
  26305. [ 0.6220, -0.4436, 1.8711, -0.3691, -0.6173, -0.4999, 0.6702,
  26306. 0.0697]]], device='cuda:0')
  26307. loss_train_step before backward: tensor(0.0095, device='cuda:0', grad_fn=<MseLossBackward>)
  26308. loss_train_step after backward: tensor(0.0095, device='cuda:0', grad_fn=<MseLossBackward>)
  26309. loss_train: 2.1782545149326324
  26310. step: 77
  26311. running loss: 0.028289019674449772
  26312. Train Steps: 77/90 Loss: 0.0283 torch.Size([8, 600, 800])
  26313. torch.Size([8, 8])
  26314. tensor([[0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124],
  26315. [0.6117, 0.4019, 0.8538, 0.4067, 0.3513, 0.3583, 0.5663, 0.5133],
  26316. [0.6221, 0.4107, 0.7788, 0.3033, 0.3950, 0.2817, 0.6075, 0.5517],
  26317. [0.6243, 0.4128, 0.7762, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417],
  26318. [ nan, nan, 0.8300, 0.3150, 0.3588, 0.3383, 0.5208, 0.5194],
  26319. [0.6250, 0.4054, 0.8770, 0.4723, 0.4662, 0.5367, 0.6162, 0.5433],
  26320. [0.6137, 0.4038, 0.8563, 0.4050, 0.3813, 0.2550, 0.5106, 0.4954],
  26321. [0.6206, 0.4001, 0.8900, 0.3933, 0.3588, 0.3567, 0.5837, 0.5083]],
  26322. device='cuda:0', dtype=torch.float64)
  26323. predictions are: tensor([[ 0.7530, -0.2939, 1.4046, -1.0338, -0.3590, -1.0755, 0.5026, 0.1979],
  26324. [ 0.4401, -0.4800, 1.5502, -0.4485, -0.5286, -0.7753, 0.2494, 0.2087],
  26325. [ 0.6913, -0.3143, 1.2940, -0.9489, -0.3650, -1.0523, 0.4405, 0.3707],
  26326. [ 0.7265, -0.2805, 1.2429, -1.0236, -0.4281, -0.9581, 0.4826, 0.3433],
  26327. [-1.5196, -1.7774, 1.3053, -0.7442, -0.5791, -0.8100, 0.3340, 0.2008],
  26328. [ 0.4117, -0.5339, 1.6772, -0.1784, -0.2167, 0.0569, 0.6415, 0.2582],
  26329. [ 0.5965, -0.3613, 1.5313, -0.4395, -0.3732, -1.0792, 0.1596, 0.1984],
  26330. [ 0.6020, -0.4144, 1.7071, -0.4450, -0.5114, -0.7503, 0.4338, 0.1238]],
  26331. device='cuda:0', grad_fn=<AddmmBackward>)
  26332. landmarks are: tensor([[[ 5.7898e-01, -4.0793e-01, 1.5929e+00, -1.0630e+00, -4.7294e-01,
  26333. -1.0725e+00, 4.1374e-01, 8.0707e-02],
  26334. [ 5.3926e-01, -4.2941e-01, 1.6575e+00, -4.0754e-01, -6.6351e-01,
  26335. -6.3079e-01, 3.2956e-01, 8.5142e-02],
  26336. [ 5.8747e-01, -3.8876e-01, 1.3111e+00, -8.8483e-01, -4.6143e-01,
  26337. -9.8491e-01, 5.2009e-01, 2.6220e-01],
  26338. [ 5.9766e-01, -3.7916e-01, 1.2995e+00, -1.0311e+00, -5.1917e-01,
  26339. -8.3865e-01, 5.8360e-01, 2.1601e-01],
  26340. [-2.2859e+00, -2.2859e+00, 1.5478e+00, -8.3095e-01, -6.2887e-01,
  26341. -7.2317e-01, 1.1982e-01, 1.1330e-01],
  26342. [ 6.0087e-01, -4.1347e-01, 1.7651e+00, -1.0433e-01, -1.3233e-01,
  26343. 1.9292e-01, 5.6051e-01, 2.2371e-01],
  26344. [ 5.4850e-01, -4.2094e-01, 1.6691e+00, -4.1524e-01, -5.2494e-01,
  26345. -1.1081e+00, 7.2521e-02, 2.0831e-03],
  26346. [ 5.8072e-01, -4.3780e-01, 1.8249e+00, -4.6913e-01, -6.2887e-01,
  26347. -6.3849e-01, 4.1039e-01, 6.2048e-02]]], device='cuda:0')
  26348. loss_train_step before backward: tensor(0.0242, device='cuda:0', grad_fn=<MseLossBackward>)
  26349. loss_train_step after backward: tensor(0.0242, device='cuda:0', grad_fn=<MseLossBackward>)
  26350. loss_train: 2.202410016208887
  26351. step: 78
  26352. running loss: 0.028236025848831885
  26353. Train Steps: 78/90 Loss: 0.0282 torch.Size([8, 600, 800])
  26354. torch.Size([8, 8])
  26355. tensor([[0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938],
  26356. [0.6218, 0.4098, 0.7238, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350],
  26357. [0.6201, 0.3970, 0.8413, 0.4950, 0.4413, 0.5183, 0.6088, 0.5400],
  26358. [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285],
  26359. [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795],
  26360. [ nan, nan, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633],
  26361. [0.6203, 0.4021, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405],
  26362. [0.6111, 0.3995, 0.8788, 0.4567, 0.3813, 0.4833, 0.5450, 0.5700]],
  26363. device='cuda:0', dtype=torch.float64)
  26364. predictions are: tensor([[ 0.6716, -0.3651, 1.6848, 0.1993, -0.4682, -0.3016, 0.3698, 0.0878],
  26365. [ 1.1126, -0.0634, 1.0590, -1.5005, -0.3099, -1.4083, 0.3902, 0.3436],
  26366. [ 0.4396, -0.4931, 1.6367, -0.1527, -0.2727, -0.1796, 0.3804, 0.3029],
  26367. [ 0.4806, -0.5013, 1.7506, -0.7712, -0.5522, -0.3615, 0.7104, 0.2110],
  26368. [ 0.7517, -0.3017, 1.6676, 0.0399, -0.4713, 0.0236, 0.7962, 0.3514],
  26369. [-1.4892, -1.7884, 1.1316, -1.2032, -0.4816, -1.2419, 0.0669, 0.3637],
  26370. [ 0.8437, -0.2534, 1.7568, -0.0860, -0.5708, -0.6748, 0.2525, 0.1682],
  26371. [ 0.6160, -0.3820, 1.7347, -0.3169, -0.5240, -0.3235, 0.1420, 0.2635]],
  26372. device='cuda:0', grad_fn=<AddmmBackward>)
  26373. landmarks are: tensor([[[ 6.3949e-01, -4.2132e-01, 1.7037e+00, 3.6228e-01, -4.3256e-01,
  26374. -1.0731e-01, 6.5602e-01, -4.8817e-03],
  26375. [ 5.8614e-01, -3.9292e-01, 1.0570e+00, -1.4314e+00, -3.2864e-01,
  26376. -1.1235e+00, 6.1824e-01, 1.8522e-01],
  26377. [ 5.7812e-01, -4.5219e-01, 1.5998e+00, 4.6189e-04, -2.4781e-01,
  26378. 1.0824e-01, 5.2587e-01, 2.0831e-01],
  26379. [ 6.1083e-01, -4.2731e-01, 1.8711e+00, -6.6159e-01, -5.7691e-01,
  26380. -1.9969e-01, 9.1557e-01, 1.5543e-01],
  26381. [ 6.3554e-01, -4.0805e-01, 1.6113e+00, 1.8522e-01, -4.7298e-01,
  26382. 1.4673e-01, 9.9965e-01, 3.9055e-01],
  26383. [-2.2859e+00, -2.2859e+00, 1.2030e+00, -1.0288e+00, -4.9607e-01,
  26384. -1.1081e+00, 8.1293e-02, 3.1609e-01],
  26385. [ 5.7910e-01, -4.2887e-01, 1.7694e+00, 3.7905e-02, -5.9233e-01,
  26386. -4.9270e-01, 4.1265e-01, 2.1070e-01],
  26387. [ 5.3684e-01, -4.4057e-01, 1.7730e+00, -1.7660e-01, -5.2494e-01,
  26388. -5.3426e-02, 2.3141e-01, 3.4688e-01]]], device='cuda:0')
  26389. loss_train_step before backward: tensor(0.0364, device='cuda:0', grad_fn=<MseLossBackward>)
  26390. loss_train_step after backward: tensor(0.0364, device='cuda:0', grad_fn=<MseLossBackward>)
  26391. loss_train: 2.2388419173657894
  26392. step: 79
  26393. running loss: 0.02833977110589607
  26394.  
  26395. Train Steps: 79/90 Loss: 0.0283 torch.Size([8, 600, 800])
  26396. torch.Size([8, 8])
  26397. tensor([[0.6219, 0.4097, 0.8738, 0.3400, 0.3563, 0.4117, 0.5975, 0.5683],
  26398. [0.6198, 0.4115, 0.7762, 0.2717, 0.3713, 0.3200, 0.5837, 0.5683],
  26399. [0.6199, 0.4071, 0.7600, 0.2117, 0.4037, 0.2767, 0.6138, 0.5550],
  26400. [0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367],
  26401. [0.6203, 0.4073, 0.8189, 0.2398, 0.4400, 0.2054, 0.5929, 0.5501],
  26402. [0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124],
  26403. [0.6199, 0.4093, 0.7913, 0.2533, 0.4288, 0.2467, 0.5975, 0.5700],
  26404. [0.6148, 0.4076, 0.8666, 0.4820, 0.4138, 0.5067, 0.5250, 0.5767]],
  26405. device='cuda:0', dtype=torch.float64)
  26406. predictions are: tensor([[ 0.4967, -0.4445, 1.6899, -0.5683, -0.6099, -0.5226, 0.4112, 0.2618],
  26407. [ 0.5527, -0.3992, 1.2878, -0.9653, -0.5312, -0.8909, 0.4131, 0.3071],
  26408. [ 0.5773, -0.3801, 1.1588, -1.2760, -0.3543, -1.0148, 0.4385, 0.3473],
  26409. [ 0.3141, -0.5629, 1.6904, -0.2649, -0.3561, -1.0912, 0.3390, 0.2800],
  26410. [ 0.3397, -0.5448, 1.3303, -1.0739, -0.2484, -1.2535, 0.4785, 0.2336],
  26411. [ 0.5512, -0.4214, 1.4524, -1.0067, -0.4021, -1.0468, 0.4744, 0.1392],
  26412. [ 0.3694, -0.5477, 1.2548, -1.1035, -0.3461, -1.0229, 0.5109, 0.2759],
  26413. [-0.1604, -0.9095, 1.6600, 0.0585, -0.4307, -0.0445, 0.3925, 0.2854]],
  26414. device='cuda:0', grad_fn=<AddmmBackward>)
  26415. landmarks are: tensor([[[ 0.5867, -0.3937, 1.7499, -0.7155, -0.6404, -0.3844, 0.4739,
  26416. 0.3392],
  26417. [ 0.5768, -0.3852, 1.2995, -1.0311, -0.5711, -0.8079, 0.4104,
  26418. 0.3392],
  26419. [ 0.5775, -0.4054, 1.2245, -1.3082, -0.4210, -1.0080, 0.5490,
  26420. 0.2776],
  26421. [ 0.5900, -0.3932, 1.8307, -0.3921, -0.4268, -1.1851, 0.3758,
  26422. 0.1929],
  26423. [ 0.5792, -0.4048, 1.4965, -1.1781, -0.2534, -1.3371, 0.4528,
  26424. 0.2549],
  26425. [ 0.5790, -0.4079, 1.5929, -1.0630, -0.4729, -1.0725, 0.4137,
  26426. 0.0807],
  26427. [ 0.5771, -0.3955, 1.3688, -1.1158, -0.3055, -1.1466, 0.4739,
  26428. 0.3469],
  26429. [ 0.5539, -0.4031, 1.7168, -0.0594, -0.3748, 0.0543, 0.1390,
  26430. 0.3777]]], device='cuda:0')
  26431. loss_train_step before backward: tensor(0.0219, device='cuda:0', grad_fn=<MseLossBackward>)
  26432. loss_train_step after backward: tensor(0.0219, device='cuda:0', grad_fn=<MseLossBackward>)
  26433. loss_train: 2.260784700512886
  26434. step: 80
  26435. running loss: 0.028259808756411077
  26436. Train Steps: 80/90 Loss: 0.0283 torch.Size([8, 600, 800])
  26437. torch.Size([8, 8])
  26438. tensor([[0.6224, 0.4097, 0.7438, 0.2267, 0.3850, 0.2850, 0.5988, 0.5250],
  26439. [0.6153, 0.4117, 0.8688, 0.5167, 0.4895, 0.5647, 0.5524, 0.5136],
  26440. [0.6117, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103],
  26441. [0.6231, 0.3973, 0.8650, 0.3950, 0.3625, 0.3183, 0.5837, 0.5167],
  26442. [0.6127, 0.4066, 0.8550, 0.5567, 0.4662, 0.5141, 0.5070, 0.5412],
  26443. [0.6109, 0.4041, 0.6975, 0.3167, 0.3513, 0.3383, 0.5153, 0.5319],
  26444. [0.6275, 0.4024, 0.8600, 0.2283, 0.5350, 0.1800, 0.7074, 0.5413],
  26445. [0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507]],
  26446. device='cuda:0', dtype=torch.float64)
  26447. predictions are: tensor([[ 0.5515, -0.4324, 1.2522, -1.1974, -0.5468, -0.9399, 0.5127, 0.2591],
  26448. [ 0.2839, -0.6032, 1.8354, 0.1435, -0.2102, 0.1414, 0.4120, 0.2700],
  26449. [ 0.4407, -0.4859, 0.9970, -1.3051, -0.5360, -1.1848, 0.2556, 0.2368],
  26450. [ 0.4679, -0.5184, 1.8603, -0.3844, -0.6523, -0.8006, 0.4449, 0.1568],
  26451. [ 0.4220, -0.5166, 1.7469, 0.3429, -0.2708, 0.0341, 0.1857, 0.3391],
  26452. [ 0.7520, -0.3054, 1.1945, -0.9319, -0.6896, -0.7861, 0.1203, 0.2432],
  26453. [-0.0073, -0.8187, 1.6968, -1.1910, 0.0458, -1.4126, 1.0480, 0.3479],
  26454. [ 0.5477, -0.4101, 1.0520, -1.3807, -0.5342, -1.2166, 0.3435, 0.2966]],
  26455. device='cuda:0', grad_fn=<AddmmBackward>)
  26456. landmarks are: tensor([[[ 0.5891, -0.3937, 1.1494, -1.2390, -0.5076, -0.9695, 0.4797,
  26457. 0.1390],
  26458. [ 0.5564, -0.3842, 1.7268, 0.1005, -0.0250, 0.3225, 0.2658,
  26459. 0.0862],
  26460. [ 0.5393, -0.4299, 0.7453, -1.3775, -0.5596, -1.1081, 0.1527,
  26461. 0.0712],
  26462. [ 0.5924, -0.4507, 1.7095, -0.4614, -0.6115, -0.8156, 0.4104,
  26463. 0.1005],
  26464. [ 0.5443, -0.4079, 1.6633, 0.2853, -0.1325, 0.0888, 0.0558,
  26465. 0.2138],
  26466. [ 0.5359, -0.4193, 0.9358, -0.8232, -0.6635, -0.7232, 0.0943,
  26467. 0.1710],
  26468. [ 0.6125, -0.4273, 1.6864, -1.2313, 0.1852, -1.4545, 0.9814,
  26469. 0.2142],
  26470. [ 0.5185, -0.4252, 0.9647, -1.2928, -0.4788, -1.2390, 0.2617,
  26471. 0.2576]]], device='cuda:0')
  26472. loss_train_step before backward: tensor(0.0207, device='cuda:0', grad_fn=<MseLossBackward>)
  26473. loss_train_step after backward: tensor(0.0207, device='cuda:0', grad_fn=<MseLossBackward>)
  26474. loss_train: 2.281465532258153
  26475. step: 81
  26476. running loss: 0.028166241138989544
  26477. Train Steps: 81/90 Loss: 0.0282 torch.Size([8, 600, 800])
  26478. torch.Size([8, 8])
  26479. tensor([[0.6111, 0.4033, 0.8300, 0.3267, 0.3588, 0.3333, 0.5444, 0.5637],
  26480. [0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033],
  26481. [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5363, 0.5550],
  26482. [0.6132, 0.4118, 0.8200, 0.3633, 0.3563, 0.5400, 0.5787, 0.5136],
  26483. [0.6197, 0.4090, 0.7825, 0.2500, 0.4200, 0.2483, 0.5988, 0.5667],
  26484. [0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719],
  26485. [0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166],
  26486. [0.6168, 0.4029, 0.8523, 0.3417, 0.3588, 0.5000, 0.6125, 0.5400]],
  26487. device='cuda:0', dtype=torch.float64)
  26488. predictions are: tensor([[ 0.7174, -0.3203, 1.5432, -0.7647, -0.5346, -0.9745, 0.1499, 0.2596],
  26489. [ 0.5553, -0.4361, 1.7861, 0.0788, -0.2853, 0.0368, 0.4463, 0.1984],
  26490. [ 0.7145, -0.2997, 0.9637, -1.2829, -0.4113, -1.4868, 0.1665, 0.3214],
  26491. [ 0.6634, -0.3140, 1.5424, -0.6475, -0.5232, -0.0573, 0.4547, 0.2503],
  26492. [ 0.4377, -0.4985, 1.2757, -1.2660, -0.3435, -1.3296, 0.5275, 0.3440],
  26493. [ 0.3779, -0.5464, 1.6157, -0.4544, -0.6177, -0.5650, 0.2407, 0.2846],
  26494. [ 0.2344, -0.6762, 1.8225, -0.0604, -0.4420, 0.0151, 0.8304, 0.2047],
  26495. [ 0.3008, -0.5890, 1.6560, -0.6949, -0.5658, -0.1531, 0.5472, 0.2596]],
  26496. device='cuda:0', grad_fn=<AddmmBackward>)
  26497. landmarks are: tensor([[[ 0.5366, -0.4232, 1.5478, -0.7771, -0.6289, -0.7463, 0.2288,
  26498. 0.3177],
  26499. [ 0.5668, -0.4306, 1.7730, 0.0697, -0.4037, 0.1390, 0.4450,
  26500. 0.0390],
  26501. [ 0.5448, -0.3859, 0.9242, -1.1466, -0.4152, -1.3005, 0.1910,
  26502. 0.2776],
  26503. [ 0.5466, -0.3840, 1.5016, -0.6077, -0.6404, 0.2083, 0.3871,
  26504. 0.0862],
  26505. [ 0.5764, -0.3969, 1.3284, -1.1312, -0.3460, -1.1389, 0.4797,
  26506. 0.3315],
  26507. [ 0.5367, -0.4294, 1.5709, -0.4999, -0.6693, -0.3075, 0.2455,
  26508. 0.3559],
  26509. [ 0.5944, -0.4490, 1.8643, -0.0659, -0.5147, 0.1235, 0.7684,
  26510. 0.1004],
  26511. [ 0.5631, -0.4249, 1.6509, -0.7078, -0.6289, 0.0236, 0.5432,
  26512. 0.2083]]], device='cuda:0')
  26513. loss_train_step before backward: tensor(0.0148, device='cuda:0', grad_fn=<MseLossBackward>)
  26514. loss_train_step after backward: tensor(0.0148, device='cuda:0', grad_fn=<MseLossBackward>)
  26515. loss_train: 2.2963146083056927
  26516. step: 82
  26517. running loss: 0.02800383668665479
  26518. Train Steps: 82/90 Loss: 0.0280 torch.Size([8, 600, 800])
  26519. torch.Size([8, 8])
  26520. tensor([[0.6275, 0.4024, 0.8600, 0.2283, 0.5350, 0.1800, 0.7074, 0.5413],
  26521. [0.6163, 0.4006, 0.8788, 0.4683, 0.3663, 0.4883, 0.5887, 0.5017],
  26522. [0.6053, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168],
  26523. [0.6170, 0.4102, 0.7468, 0.3695, 0.3463, 0.3767, 0.5238, 0.5823],
  26524. [0.6179, 0.4082, 0.6688, 0.2667, 0.3588, 0.3317, 0.5750, 0.5783],
  26525. [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5787, 0.5550],
  26526. [0.6022, 0.3994, 0.8025, 0.3350, 0.3350, 0.4400, 0.5565, 0.5025],
  26527. [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274]],
  26528. device='cuda:0', dtype=torch.float64)
  26529. predictions are: tensor([[ 0.1430, -0.7105, 1.6730, -1.2270, 0.1805, -1.4558, 0.9720, 0.2920],
  26530. [ 0.5546, -0.4364, 1.8704, -0.0628, -0.5260, 0.0190, 0.3254, 0.1272],
  26531. [-0.0603, -0.8157, 0.9096, -1.3470, -0.3919, -1.3662, 0.3169, 0.2475],
  26532. [ 0.4587, -0.4335, 1.3221, -0.6188, -0.6190, -0.6014, 0.0964, 0.4070],
  26533. [ 0.5745, -0.3822, 1.0003, -1.0263, -0.6209, -0.8339, 0.3180, 0.3165],
  26534. [ 1.1199, -0.0309, 1.5450, -0.9909, -0.6218, -0.6417, 0.3637, 0.2202],
  26535. [ 0.3321, -0.5572, 1.5391, -0.7018, -0.6428, -0.4123, 0.2566, 0.1486],
  26536. [ 0.4968, -0.4762, 1.9318, 0.1283, -0.5195, -0.2110, 0.7342, 0.2339]],
  26537. device='cuda:0', grad_fn=<AddmmBackward>)
  26538. landmarks are: tensor([[[ 0.6125, -0.4273, 1.6864, -1.2313, 0.1852, -1.4545, 0.9814,
  26539. 0.2142],
  26540. [ 0.5609, -0.4354, 1.7730, -0.1227, -0.5942, -0.0303, 0.4335,
  26541. 0.0313],
  26542. [ 0.5102, -0.4223, 0.8999, -1.4011, -0.4383, -1.3082, 0.2267,
  26543. 0.1013],
  26544. [ 0.5639, -0.3911, 1.1634, -0.5794, -0.6866, -0.5461, 0.1334,
  26545. 0.4036],
  26546. [ 0.5680, -0.4005, 0.8030, -1.0542, -0.6289, -0.7540, 0.3700,
  26547. 0.3854],
  26548. [ 0.5647, -0.4129, 1.4901, -1.0619, -0.6462, -0.5846, 0.3873,
  26549. 0.2776],
  26550. [ 0.4956, -0.4413, 1.4208, -0.7386, -0.7386, -0.2536, 0.2843,
  26551. 0.0351],
  26552. [ 0.6108, -0.4008, 1.9088, -0.0253, -0.5769, -0.3075, 0.7905,
  26553. 0.1499]]], device='cuda:0')
  26554. loss_train_step before backward: tensor(0.0251, device='cuda:0', grad_fn=<MseLossBackward>)
  26555. loss_train_step after backward: tensor(0.0251, device='cuda:0', grad_fn=<MseLossBackward>)
  26556. loss_train: 2.3214387968182564
  26557. step: 83
  26558. running loss: 0.02796914213034044
  26559.  
  26560. Train Steps: 83/90 Loss: 0.0280 torch.Size([8, 600, 800])
  26561. torch.Size([8, 8])
  26562. tensor([[ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600],
  26563. [0.6201, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044],
  26564. [0.6321, 0.4048, 0.8738, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927],
  26565. [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896],
  26566. [0.6336, 0.4154, 0.8900, 0.2767, 0.4988, 0.2867, 0.7422, 0.5540],
  26567. [0.6254, 0.3993, 0.8988, 0.4767, 0.3987, 0.5517, 0.6955, 0.5285],
  26568. [0.6102, 0.4020, 0.8638, 0.3717, 0.3625, 0.5017, 0.6038, 0.5500],
  26569. [0.6329, 0.4196, 0.9238, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748]],
  26570. device='cuda:0', dtype=torch.float64)
  26571. predictions are: tensor([[-2.2269, -2.2273, 1.1713, -1.1375, -0.4127, -1.0421, 0.1625, 0.2478],
  26572. [ 0.6665, -0.2975, 1.0809, -1.5144, -0.3767, -1.2959, 0.3798, 0.1838],
  26573. [ 0.8967, -0.1623, 1.6080, 0.1411, -0.5546, -0.2800, 0.2731, 0.0371],
  26574. [ 0.8213, -0.1644, 1.3793, -0.6185, -0.5996, -0.8791, -0.0651, 0.3671],
  26575. [ 0.6198, -0.3447, 1.5878, -1.1289, -0.0898, -1.0752, 0.8322, 0.2534],
  26576. [ 0.5938, -0.3939, 1.7633, -0.2117, -0.4702, 0.2779, 0.6836, 0.1326],
  26577. [ 0.6316, -0.3298, 1.6066, -0.6564, -0.6365, -0.0327, 0.4298, 0.2297],
  26578. [ 0.6697, -0.3118, 1.7474, -0.2772, -0.3892, -0.8977, 0.6521, 0.3015]],
  26579. device='cuda:0', grad_fn=<AddmmBackward>)
  26580. landmarks are: tensor([[[-2.2859, -2.2859, 1.4266, -1.1389, -0.4499, -1.1235, 0.2891,
  26581. 0.3007],
  26582. [ 0.5785, -0.4129, 1.2785, -1.3996, -0.3227, -1.3259, 0.4258,
  26583. 0.0438],
  26584. [ 0.6335, -0.4162, 1.7499, 0.3084, -0.4961, -0.2459, 0.6524,
  26585. -0.0102],
  26586. [ 0.5485, -0.3997, 1.4445, -0.4895, -0.6000, -0.8309, 0.1878,
  26587. 0.4374],
  26588. [ 0.6405, -0.3671, 1.8249, -1.0080, 0.0178, -0.9618, 1.1422,
  26589. 0.2730],
  26590. [ 0.6026, -0.4417, 1.8654, -0.0842, -0.4441, 0.2622, 0.9265,
  26591. 0.1554],
  26592. [ 0.5324, -0.4293, 1.7037, -0.5692, -0.6115, 0.0313, 0.5028,
  26593. 0.2545],
  26594. [ 0.6372, -0.3479, 1.9808, -0.1689, -0.3171, -0.9310, 1.1057,
  26595. 0.3692]]], device='cuda:0')
  26596. loss_train_step before backward: tensor(0.0214, device='cuda:0', grad_fn=<MseLossBackward>)
  26597. loss_train_step after backward: tensor(0.0214, device='cuda:0', grad_fn=<MseLossBackward>)
  26598. loss_train: 2.342796679586172
  26599. step: 84
  26600. running loss: 0.027890436661740143
  26601. Train Steps: 84/90 Loss: 0.0279 torch.Size([8, 600, 800])
  26602. torch.Size([8, 8])
  26603. tensor([[0.6199, 0.4093, 0.7913, 0.2533, 0.4288, 0.2467, 0.5975, 0.5700],
  26604. [0.6053, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168],
  26605. [0.6180, 0.4064, 0.7200, 0.2467, 0.4275, 0.2367, 0.5963, 0.6217],
  26606. [0.6275, 0.4157, 0.8337, 0.5800, 0.3763, 0.4200, 0.5547, 0.6125],
  26607. [0.6109, 0.4041, 0.6975, 0.3167, 0.3513, 0.3383, 0.5153, 0.5319],
  26608. [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617],
  26609. [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578],
  26610. [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366]],
  26611. device='cuda:0', dtype=torch.float64)
  26612. predictions are: tensor([[ 0.6604, -0.3537, 1.5051, -1.2975, -0.3695, -0.9548, 0.5446, 0.1394],
  26613. [ 0.1555, -0.6838, 1.0384, -1.4719, -0.4375, -1.2605, 0.3670, 0.1280],
  26614. [ 0.3765, -0.5204, 1.1939, -1.3233, -0.3674, -1.1210, 0.4890, 0.3674],
  26615. [ 0.7508, -0.2881, 1.7318, 0.2697, -0.5587, -0.2539, 0.3189, 0.3947],
  26616. [ 0.6070, -0.3943, 1.2620, -1.0188, -0.6195, -0.6756, 0.1259, 0.1423],
  26617. [ 0.6287, -0.3404, 1.9525, 0.1932, -0.5037, -0.1270, 0.3228, 0.3000],
  26618. [ 0.2278, -0.6186, 1.0869, -1.5289, -0.4159, -1.2903, 0.3593, 0.1703],
  26619. [ 0.2406, -0.6494, 1.6878, 0.1533, -0.4699, 0.1824, 0.8549, 0.1438]],
  26620. device='cuda:0', grad_fn=<AddmmBackward>)
  26621. landmarks are: tensor([[[ 0.5771, -0.3955, 1.3688, -1.1158, -0.3055, -1.1466, 0.4739,
  26622. 0.3469],
  26623. [ 0.5102, -0.4223, 0.8999, -1.4011, -0.4383, -1.3082, 0.2267,
  26624. 0.1013],
  26625. [ 0.5687, -0.4088, 1.0397, -1.1466, -0.3113, -1.1928, 0.4681,
  26626. 0.5855],
  26627. [ 0.6124, -0.3658, 1.5651, 0.3931, -0.5480, -0.3460, 0.2761,
  26628. 0.5431],
  26629. [ 0.5359, -0.4193, 0.9358, -0.8232, -0.6635, -0.7232, 0.0943,
  26630. 0.1710],
  26631. [ 0.5897, -0.3527, 1.8018, 0.2545, -0.5307, -0.3229, 0.3122,
  26632. 0.3084],
  26633. [ 0.5232, -0.4189, 0.9358, -1.4006, -0.3921, -1.3698, 0.2555,
  26634. 0.2906],
  26635. [ 0.6289, -0.4345, 1.3794, 0.3679, -0.4845, 0.0390, 0.9265,
  26636. 0.1928]]], device='cuda:0')
  26637. loss_train_step before backward: tensor(0.0225, device='cuda:0', grad_fn=<MseLossBackward>)
  26638. loss_train_step after backward: tensor(0.0225, device='cuda:0', grad_fn=<MseLossBackward>)
  26639. loss_train: 2.365304049104452
  26640. step: 85
  26641. running loss: 0.02782710646005238
  26642. Train Steps: 85/90 Loss: 0.0278 torch.Size([8, 600, 800])
  26643. torch.Size([8, 8])
  26644. tensor([[0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117],
  26645. [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586],
  26646. [0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183],
  26647. [0.6135, 0.4115, 0.8838, 0.4667, 0.4288, 0.6050, 0.5778, 0.5097],
  26648. [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600],
  26649. [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390],
  26650. [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220],
  26651. [0.6125, 0.4076, 0.8488, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505]],
  26652. device='cuda:0', dtype=torch.float64)
  26653. predictions are: tensor([[ 0.4752, -0.4915, 1.7250, -0.1743, -0.5882, -0.1805, 0.7140, 0.1793],
  26654. [ 0.5620, -0.3887, 1.4668, -1.3343, -0.3144, -1.1160, 0.8748, 0.2578],
  26655. [ 0.6195, -0.3832, 1.7568, -0.3389, -0.4534, 0.3030, 0.6783, 0.1351],
  26656. [ 0.3791, -0.4708, 1.6688, -0.3477, -0.2809, 0.3300, 0.3379, 0.1553],
  26657. [ 0.5733, -0.3845, 1.6611, 0.1772, -0.4800, -0.0565, 0.3584, 0.3798],
  26658. [ 0.5504, -0.4220, 1.6632, -0.1711, -0.4728, -0.0582, 0.2567, 0.2242],
  26659. [ 0.5528, -0.4051, 0.8821, -1.4736, -0.3661, -1.5189, 0.3338, 0.2473],
  26660. [ 0.6205, -0.3280, 1.5286, -0.6112, -0.6699, -0.7295, 0.0736, 0.2279]],
  26661. device='cuda:0', grad_fn=<AddmmBackward>)
  26662. landmarks are: tensor([[[ 0.6262, -0.4461, 1.8480, -0.0534, -0.5827, -0.1227, 0.6587,
  26663. 0.0774],
  26664. [ 0.6075, -0.4129, 1.6055, -1.0080, -0.2420, -1.0080, 0.9704,
  26665. 0.2944],
  26666. [ 0.5934, -0.4276, 1.8192, -0.1458, -0.4499, 0.3777, 0.6702,
  26667. 0.1082],
  26668. [ 0.5477, -0.3851, 1.7961, -0.1304, -0.3055, 0.5085, 0.3830,
  26669. 0.0682],
  26670. [ 0.5883, -0.3594, 1.7557, 0.2545, -0.4152, -0.0611, 0.3353,
  26671. 0.3007],
  26672. [ 0.5250, -0.4661, 1.7383, -0.0765, -0.4268, -0.0226, 0.2535,
  26673. 0.2035],
  26674. [ 0.5566, -0.4160, 0.9935, -1.3313, -0.2824, -1.5161, 0.2144,
  26675. 0.1253],
  26676. [ 0.5433, -0.4032, 1.6344, -0.4922, -0.5769, -0.5846, 0.0357,
  26677. 0.2567]]], device='cuda:0')
  26678. loss_train_step before backward: tensor(0.0094, device='cuda:0', grad_fn=<MseLossBackward>)
  26679. loss_train_step after backward: tensor(0.0094, device='cuda:0', grad_fn=<MseLossBackward>)
  26680. loss_train: 2.374746444635093
  26681. step: 86
  26682. running loss: 0.027613330751570853
  26683. Train Steps: 86/90 Loss: 0.0276 torch.Size([8, 600, 800])
  26684. torch.Size([8, 8])
  26685. tensor([[0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901],
  26686. [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309],
  26687. [0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274],
  26688. [0.6218, 0.4185, 0.7338, 0.2650, 0.4625, 0.1950, 0.5687, 0.5800],
  26689. [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400],
  26690. [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683],
  26691. [0.6311, 0.3998, 0.7975, 0.5767, 0.3838, 0.4850, 0.7327, 0.5343],
  26692. [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356]],
  26693. device='cuda:0', dtype=torch.float64)
  26694. predictions are: tensor([[ 0.3664, -0.5092, 1.3370, -1.0725, -0.5688, -0.9993, 0.1131, 0.0623],
  26695. [ 0.6189, -0.3521, 1.7721, -0.1010, -0.5002, -0.3470, 0.6110, 0.1564],
  26696. [ 0.2863, -0.5541, 1.6003, -0.9776, -0.1182, -1.1446, 0.6046, 0.1507],
  26697. [ 0.2970, -0.5251, 0.9866, -1.2467, -0.2565, -1.2284, 0.3042, 0.3866],
  26698. [ 0.7506, -0.2783, 1.5889, -0.8261, -0.6352, 0.0359, 0.5611, 0.2387],
  26699. [ 0.5821, -0.3698, 1.6028, 0.0332, -0.5533, -0.3493, 0.1958, 0.3500],
  26700. [ 0.2462, -0.6476, 1.4856, 0.0648, -0.5149, 0.1151, 0.7020, 0.1618],
  26701. [ 0.6343, -0.3339, 1.3000, -1.2645, -0.3201, -1.0795, 0.7160, 0.1657]],
  26702. device='cuda:0', grad_fn=<AddmmBackward>)
  26703. landmarks are: tensor([[[ 0.5515, -0.4129, 1.4785, -1.0080, -0.5192, -1.1004, 0.1034,
  26704. -0.0220],
  26705. [ 0.6520, -0.3623, 1.8885, 0.0313, -0.5538, -0.5384, 0.6926,
  26706. 0.1661],
  26707. [ 0.6537, -0.3719, 1.6979, -0.8617, -0.0169, -1.3621, 0.6926,
  26708. 0.1501],
  26709. [ 0.5862, -0.3530, 1.1032, -1.0619, -0.1497, -1.3852, 0.3411,
  26710. 0.3931],
  26711. [ 0.5631, -0.4008, 1.6113, -0.7309, -0.6693, -0.1304, 0.5836,
  26712. 0.2083],
  26713. [ 0.5777, -0.3915, 1.7961, 0.1698, -0.5192, -0.5307, 0.2141,
  26714. 0.3392],
  26715. [ 0.6289, -0.4393, 1.3977, 0.3777, -0.5134, -0.0457, 1.0984,
  26716. 0.1821],
  26717. [ 0.6174, -0.4225, 1.4975, -1.1709, -0.3174, -1.1806, 0.6539,
  26718. 0.1879]]], device='cuda:0')
  26719. loss_train_step before backward: tensor(0.0211, device='cuda:0', grad_fn=<MseLossBackward>)
  26720. loss_train_step after backward: tensor(0.0211, device='cuda:0', grad_fn=<MseLossBackward>)
  26721. loss_train: 2.3958609933033586
  26722. step: 87
  26723. running loss: 0.027538632106935156
  26724.  
  26725. Train Steps: 87/90 Loss: 0.0275 torch.Size([8, 600, 800])
  26726. torch.Size([8, 8])
  26727. tensor([[0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869],
  26728. [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366],
  26729. [0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433],
  26730. [0.6260, 0.4153, 0.9000, 0.4533, 0.4025, 0.2633, 0.6223, 0.4967],
  26731. [0.6108, 0.4011, 0.8037, 0.3400, 0.3700, 0.2933, 0.5658, 0.5617],
  26732. [0.6221, 0.4107, 0.7788, 0.3033, 0.3950, 0.2817, 0.6075, 0.5517],
  26733. [0.6204, 0.4007, 0.7838, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167],
  26734. [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356]],
  26735. device='cuda:0', dtype=torch.float64)
  26736. predictions are: tensor([[ 0.3245, -0.5463, 1.6416, 0.1805, -0.5145, -0.1114, 0.1358, 0.1085],
  26737. [ 0.4553, -0.4918, 1.4523, 0.0928, -0.5212, 0.1824, 0.9054, 0.2236],
  26738. [ 0.5183, -0.4307, 1.6151, -0.0479, -0.1344, 0.1879, 0.0953, 0.3011],
  26739. [ 0.4604, -0.4627, 1.8237, -0.4227, -0.4729, -0.9072, 0.5547, 0.1219],
  26740. [ 0.5760, -0.3800, 1.4301, -0.8783, -0.5904, -0.7605, 0.3787, 0.2626],
  26741. [ 0.4793, -0.4195, 1.2977, -1.1489, -0.4808, -0.8817, 0.4595, 0.2736],
  26742. [ 0.5724, -0.3816, 1.1668, -1.3779, -0.2650, -1.3689, 0.5275, 0.1317],
  26743. [ 0.6453, -0.3177, 1.3266, -1.2657, -0.3371, -1.0873, 0.7413, 0.1793]],
  26744. device='cuda:0', grad_fn=<AddmmBackward>)
  26745. landmarks are: tensor([[[ 0.5320, -0.4488, 1.6633, 0.3315, -0.5018, -0.2459, 0.0828,
  26746. -0.0370],
  26747. [ 0.6289, -0.4345, 1.3794, 0.3679, -0.4845, 0.0390, 0.9265,
  26748. 0.1928],
  26749. [ 0.5295, -0.4373, 1.6553, 0.1011, 0.0380, 0.0671, 0.0813,
  26750. 0.2237],
  26751. [ 0.6055, -0.3676, 1.8711, -0.1920, -0.4268, -1.0696, 0.5887,
  26752. 0.0081],
  26753. [ 0.5355, -0.4335, 1.4266, -0.7155, -0.5769, -0.9310, 0.3275,
  26754. 0.3087],
  26755. [ 0.5875, -0.3888, 1.3111, -0.8848, -0.4614, -0.9849, 0.5201,
  26756. 0.2622],
  26757. [ 0.5796, -0.4351, 1.3342, -1.3159, -0.2189, -1.4853, 0.4046,
  26758. 0.1005],
  26759. [ 0.6174, -0.4225, 1.4975, -1.1709, -0.3174, -1.1806, 0.6539,
  26760. 0.1879]]], device='cuda:0')
  26761. loss_train_step before backward: tensor(0.0118, device='cuda:0', grad_fn=<MseLossBackward>)
  26762. loss_train_step after backward: tensor(0.0118, device='cuda:0', grad_fn=<MseLossBackward>)
  26763. loss_train: 2.4076440930366516
  26764. step: 88
  26765. running loss: 0.027359591966325588
  26766. Train Steps: 88/90 Loss: 0.0274 torch.Size([8, 600, 800])
  26767. torch.Size([8, 8])
  26768. tensor([[0.6231, 0.3973, 0.8650, 0.3950, 0.3625, 0.3183, 0.5837, 0.5167],
  26769. [0.6204, 0.4007, 0.7838, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167],
  26770. [0.6250, 0.4146, 0.8838, 0.3933, 0.3588, 0.4283, 0.6162, 0.5367],
  26771. [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377],
  26772. [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100],
  26773. [0.6265, 0.4091, 0.8950, 0.3533, 0.3600, 0.3967, 0.6295, 0.4901],
  26774. [0.6175, 0.3957, 0.8700, 0.4817, 0.4662, 0.5133, 0.5800, 0.5517],
  26775. [ nan, nan, 0.8363, 0.3317, 0.3563, 0.3367, 0.5329, 0.5142]],
  26776. device='cuda:0', dtype=torch.float64)
  26777. predictions are: tensor([[ 0.8967, -0.1894, 1.6235, -0.5461, -0.5360, -0.7694, 0.4593, 0.1840],
  26778. [ 0.7258, -0.2661, 1.1216, -1.2929, -0.1953, -1.4731, 0.5593, 0.1530],
  26779. [ 0.6065, -0.3242, 1.6106, -0.5376, -0.5609, -0.3627, 0.5166, 0.2849],
  26780. [ 0.7895, -0.2600, 1.5678, 0.1344, -0.4056, -0.0342, 0.4855, 0.2049],
  26781. [ 0.5367, -0.4251, 1.5836, 0.1638, -0.4319, -0.0222, 0.6117, 0.1143],
  26782. [ 0.9586, -0.0897, 1.6710, -0.7051, -0.5443, -0.4784, 0.5907, 0.1210],
  26783. [ 0.7902, -0.2370, 1.5886, -0.1792, -0.1250, 0.1207, 0.3287, 0.1890],
  26784. [-1.6931, -1.8893, 1.3705, -0.9159, -0.5961, -0.7617, 0.1850, 0.1535]],
  26785. device='cuda:0', grad_fn=<AddmmBackward>)
  26786. landmarks are: tensor([[[ 0.5924, -0.4507, 1.7095, -0.4614, -0.6115, -0.8156, 0.4104,
  26787. 0.1005],
  26788. [ 0.5796, -0.4351, 1.3342, -1.3159, -0.2189, -1.4853, 0.4046,
  26789. 0.1005],
  26790. [ 0.6009, -0.3710, 1.7961, -0.4691, -0.6289, -0.3075, 0.5605,
  26791. 0.1929],
  26792. [ 0.5775, -0.4607, 1.6741, 0.1962, -0.4036, -0.1212, 0.4588,
  26793. 0.1979],
  26794. [ 0.6125, -0.4273, 1.6402, 0.2006, -0.4788, -0.0303, 0.6182,
  26795. 0.0697],
  26796. [ 0.6077, -0.3965, 1.8480, -0.6539, -0.6231, -0.4537, 0.6216,
  26797. -0.0220],
  26798. [ 0.5662, -0.4581, 1.7326, -0.0611, -0.1323, 0.0851, 0.3931,
  26799. 0.2622],
  26800. [-2.2859, -2.2859, 1.5767, -0.7540, -0.6404, -0.7309, 0.1753,
  26801. 0.0893]]], device='cuda:0')
  26802. loss_train_step before backward: tensor(0.0235, device='cuda:0', grad_fn=<MseLossBackward>)
  26803. loss_train_step after backward: tensor(0.0235, device='cuda:0', grad_fn=<MseLossBackward>)
  26804. loss_train: 2.4311892818659544
  26805. step: 89
  26806. running loss: 0.02731673350411185
  26807. Train Steps: 89/90 Loss: 0.0273 torch.Size([8, 600, 800])
  26808. torch.Size([8, 8])
  26809. tensor([[0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967],
  26810. [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600],
  26811. [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389],
  26812. [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650],
  26813. [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413],
  26814. [0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945],
  26815. [0.6250, 0.4054, 0.8770, 0.4723, 0.4662, 0.5367, 0.6162, 0.5433],
  26816. [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650]],
  26817. device='cuda:0', dtype=torch.float64)
  26818. predictions are: tensor([[ 0.5705, -0.3882, 1.6861, -0.0167, -0.5071, -0.7305, 0.5165, 0.0700],
  26819. [ 0.6534, -0.3226, 1.5963, -0.2239, -0.1139, 0.3164, 0.3441, 0.1890],
  26820. [ 0.5489, -0.3968, 1.0934, -1.3142, -0.5016, -1.0165, 0.5338, 0.1896],
  26821. [ 0.3663, -0.4814, 1.4985, -0.4219, -0.5752, -0.7880, 0.2258, 0.3580],
  26822. [ 0.4916, -0.5086, 1.8622, -0.3967, -0.3204, -0.9278, 0.9234, 0.0578],
  26823. [ 0.3684, -0.5708, 1.8041, -0.3877, -0.3142, -0.5238, 0.8517, 0.3063],
  26824. [ 0.8468, -0.2431, 1.6189, -0.1766, -0.1639, 0.3228, 0.4935, 0.1421],
  26825. [ 0.5632, -0.3607, 1.5821, -0.5251, -0.6298, -0.4061, 0.1537, 0.2595]],
  26826. device='cuda:0', grad_fn=<AddmmBackward>)
  26827. landmarks are: tensor([[[ 0.6054, -0.3767, 1.7788, 0.0774, -0.5711, -0.7694, 0.5887,
  26828. 0.0081],
  26829. [ 0.5652, -0.4325, 1.7388, -0.1429, -0.0688, 0.3469, 0.4970,
  26830. 0.3007],
  26831. [ 0.5697, -0.4466, 1.1973, -1.1871, -0.4571, -0.9965, 0.5219,
  26832. 0.2032],
  26833. [ 0.5894, -0.3503, 1.6402, -0.3614, -0.5827, -0.7925, 0.3238,
  26834. 0.3238],
  26835. [ 0.6224, -0.4345, 1.9404, -0.2921, -0.3171, -0.8771, 1.0655,
  26836. 0.2142],
  26837. [ 0.6471, -0.3719, 1.9866, -0.3921, -0.3806, -0.5538, 1.0070,
  26838. 0.4600],
  26839. [ 0.6009, -0.4135, 1.7651, -0.1043, -0.1323, 0.1929, 0.5605,
  26840. 0.2237],
  26841. [ 0.5757, -0.3917, 1.7095, -0.4768, -0.6346, -0.4229, 0.3931,
  26842. 0.3238]]], device='cuda:0')
  26843. loss_train_step before backward: tensor(0.0120, device='cuda:0', grad_fn=<MseLossBackward>)
  26844. loss_train_step after backward: tensor(0.0120, device='cuda:0', grad_fn=<MseLossBackward>)
  26845. loss_train: 2.443223439157009
  26846. step: 90
  26847. running loss: 0.027146927101744545
  26848. Valid Steps: 10/10 Loss: nan 6.5902
  26849. --------------------------------------------------
  26850. Epoch: 7 Train Loss: 0.0271 Valid Loss: nan
  26851. --------------------------------------------------
  26852. size of train loader is: 90
  26853. torch.Size([8, 600, 800])
  26854. torch.Size([8, 8])
  26855. tensor([[0.6199, 0.4112, 0.8475, 0.3717, 0.3550, 0.4350, 0.6063, 0.6083],
  26856. [0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626],
  26857. [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450],
  26858. [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567],
  26859. [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123],
  26860. [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650],
  26861. [ nan, nan, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702],
  26862. [0.6204, 0.4013, 0.8075, 0.2400, 0.4313, 0.2050, 0.5800, 0.5150]],
  26863. device='cuda:0', dtype=torch.float64)
  26864. predictions are: tensor([[ 0.7039, -0.2916, 1.6499, -0.5477, -0.5646, -0.1249, 0.4112, 0.3280],
  26865. [ 0.8252, -0.2680, 1.5895, 0.4821, -0.4184, 0.0452, 0.6871, -0.0390],
  26866. [ 0.7905, -0.2879, 1.7602, -0.0484, -0.4769, -0.2194, 0.5950, 0.1834],
  26867. [ 0.7472, -0.2969, 1.6772, -0.1715, -0.5247, -0.3710, 0.4660, 0.2601],
  26868. [ 0.8241, -0.2370, 1.0809, -1.1172, -0.4864, -1.0178, 0.3043, 0.0526],
  26869. [ 0.7743, -0.2381, 1.6707, -0.4007, -0.5739, -0.3339, 0.3342, 0.2430],
  26870. [-1.4220, -1.7229, 1.8686, -0.8128, 0.0823, -1.1515, 0.7699, 0.2080],
  26871. [ 0.5649, -0.4012, 1.3509, -1.1256, -0.2297, -1.2799, 0.5118, 0.0663]],
  26872. device='cuda:0', grad_fn=<AddmmBackward>)
  26873. landmarks are: tensor([[[ 0.5774, -0.3868, 1.6286, -0.5692, -0.6462, -0.2767, 0.5143,
  26874. 0.5239],
  26875. [ 0.6105, -0.4293, 1.5824, 0.5239, -0.4730, -0.0380, 0.5025,
  26876. -0.1492],
  26877. [ 0.5785, -0.4252, 1.7676, -0.1602, -0.5845, -0.3446, 0.4566,
  26878. 0.2314],
  26879. [ 0.5432, -0.4336, 1.7095, -0.1766, -0.5942, -0.4845, 0.3007,
  26880. 0.2853],
  26881. [ 0.5342, -0.4280, 1.0051, -1.0619, -0.5711, -1.0388, 0.2226,
  26882. 0.0802],
  26883. [ 0.5757, -0.3917, 1.7095, -0.4768, -0.6346, -0.4229, 0.3931,
  26884. 0.3238],
  26885. [-2.2859, -2.2859, 1.8942, -0.6693, 0.0871, -1.2236, 1.1130,
  26886. 0.3478],
  26887. [ 0.5796, -0.4326, 1.4439, -1.1774, -0.2940, -1.3390, 0.3931,
  26888. 0.0928]]], device='cuda:0')
  26889. loss_train_step before backward: tensor(0.0307, device='cuda:0', grad_fn=<MseLossBackward>)
  26890.  
  26891. loss_train_step after backward: tensor(0.0307, device='cuda:0', grad_fn=<MseLossBackward>)
  26892. loss_train: 0.030708475038409233
  26893. step: 1
  26894. running loss: 0.030708475038409233
  26895. Train Steps: 1/90 Loss: 0.0307 torch.Size([8, 600, 800])
  26896. torch.Size([8, 8])
  26897. tensor([[0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672],
  26898. [0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400],
  26899. [0.6138, 0.4054, 0.8750, 0.4750, 0.4363, 0.5017, 0.5086, 0.5822],
  26900. [0.6364, 0.4165, 0.9088, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297],
  26901. [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355],
  26902. [0.6189, 0.4033, 0.8650, 0.5267, 0.4487, 0.5150, 0.5925, 0.5050],
  26903. [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220],
  26904. [0.6142, 0.4127, 0.7575, 0.3067, 0.3438, 0.4383, 0.5778, 0.5207]],
  26905. device='cuda:0', dtype=torch.float64)
  26906. predictions are: tensor([[ 0.7814, -0.3283, 1.7350, 0.3717, -0.6610, -0.2527, 0.6901, 0.0986],
  26907. [ 0.4466, -0.4523, 1.5270, -0.9148, -0.1701, -1.3227, 0.4904, 0.1241],
  26908. [ 0.5207, -0.4192, 1.7305, 0.0228, -0.3488, 0.1860, 0.1492, 0.2329],
  26909. [ 0.1713, -0.6661, 1.9103, -0.2153, -0.4411, -0.7909, 0.6483, 0.1719],
  26910. [ 0.6538, -0.4094, 1.7988, -0.7230, -0.2640, -0.6816, 1.0283, 0.1449],
  26911. [ 0.6401, -0.3794, 1.7304, 0.2279, -0.2252, 0.2746, 0.4670, 0.1517],
  26912. [ 0.6683, -0.3525, 1.0400, -1.2745, -0.3444, -1.4572, 0.2691, 0.1660],
  26913. [ 0.5483, -0.4067, 1.2148, -0.8755, -0.6606, -0.2426, 0.4862, 0.2548]],
  26914. device='cuda:0', grad_fn=<AddmmBackward>)
  26915. landmarks are: tensor([[[ 0.6128, -0.4375, 1.7788, 0.2699, -0.6346, -0.2536, 0.5463,
  26916. -0.1278],
  26917. [ 0.5911, -0.3888, 1.4727, -0.9541, -0.0919, -1.4930, 0.3988,
  26918. 0.2083],
  26919. [ 0.5491, -0.4132, 1.7557, -0.0919, -0.2709, 0.0313, 0.0635,
  26920. 0.4032],
  26921. [ 0.6537, -0.3623, 1.9115, -0.2690, -0.4037, -0.8309, 0.6926,
  26922. 0.1608],
  26923. [ 0.6224, -0.4105, 1.9173, -0.7771, -0.1030, -0.7308, 1.1532,
  26924. 0.1875],
  26925. [ 0.5726, -0.4230, 1.7095, 0.1467, -0.2132, 0.0928, 0.4508,
  26926. 0.0467],
  26927. [ 0.5566, -0.4160, 0.9935, -1.3313, -0.2824, -1.5161, 0.2144,
  26928. 0.1253],
  26929. [ 0.5509, -0.3798, 1.2129, -0.8694, -0.6982, -0.2613, 0.3830,
  26930. 0.1193]]], device='cuda:0')
  26931. loss_train_step before backward: tensor(0.0126, device='cuda:0', grad_fn=<MseLossBackward>)
  26932. loss_train_step after backward: tensor(0.0126, device='cuda:0', grad_fn=<MseLossBackward>)
  26933. loss_train: 0.043306597508490086
  26934. step: 2
  26935. running loss: 0.021653298754245043
  26936. Train Steps: 2/90 Loss: 0.0217 torch.Size([8, 600, 800])
  26937. torch.Size([8, 8])
  26938. tensor([[0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869],
  26939. [0.6183, 0.4076, 0.8838, 0.4517, 0.3813, 0.4483, 0.5775, 0.5633],
  26940. [0.6204, 0.4007, 0.7838, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167],
  26941. [0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6262, 0.5167],
  26942. [0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200],
  26943. [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366],
  26944. [0.6250, 0.3993, 0.9138, 0.4333, 0.3763, 0.5217, 0.6995, 0.5320],
  26945. [0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274]],
  26946. device='cuda:0', dtype=torch.float64)
  26947. predictions are: tensor([[ 0.5350, -0.4089, 1.6639, -0.7590, -0.2530, -1.2740, 0.3201, 0.0325],
  26948. [ 0.6042, -0.3893, 1.6976, -0.0241, -0.5590, 0.0529, 0.2902, 0.2995],
  26949. [ 0.5105, -0.4591, 1.2283, -1.1165, -0.2299, -1.4138, 0.3845, 0.1057],
  26950. [ 0.5441, -0.4624, 1.8066, -0.1762, -0.5450, -0.5049, 0.5332, 0.1960],
  26951. [ 0.2679, -0.5802, 1.7151, -0.9117, -0.2376, -1.0972, 0.5032, 0.1522],
  26952. [ 0.6417, -0.3980, 1.3725, 0.3973, -0.4975, 0.1530, 0.8924, 0.2196],
  26953. [ 0.7965, -0.3336, 1.8087, -0.2137, -0.5701, 0.3298, 0.9831, 0.1608],
  26954. [ 0.3200, -0.5568, 1.6624, -0.7711, -0.0910, -1.2281, 0.5951, 0.1513]],
  26955. device='cuda:0', grad_fn=<AddmmBackward>)
  26956. landmarks are: tensor([[[ 0.6119, -0.3927, 1.6979, -0.7925, -0.2536, -1.3698, 0.4282,
  26957. -0.0368],
  26958. [ 0.5700, -0.4032, 1.7961, -0.1997, -0.5249, -0.2151, 0.3815,
  26959. 0.3161],
  26960. [ 0.5796, -0.4351, 1.3342, -1.3159, -0.2189, -1.4853, 0.4046,
  26961. 0.1005],
  26962. [ 0.6094, -0.3947, 1.8885, -0.2998, -0.5769, -0.6770, 0.6067,
  26963. 0.1005],
  26964. [ 0.6089, -0.3948, 1.7383, -0.8617, -0.2536, -1.2390, 0.6009,
  26965. 0.1159],
  26966. [ 0.6289, -0.4345, 1.3794, 0.3679, -0.4845, 0.0390, 0.9265,
  26967. 0.1928],
  26968. [ 0.6010, -0.4417, 1.9346, -0.2844, -0.5480, 0.1236, 0.9448,
  26969. 0.1715],
  26970. [ 0.6537, -0.3719, 1.6979, -0.8617, -0.0169, -1.3621, 0.6926,
  26971. 0.1501]]], device='cuda:0')
  26972. loss_train_step before backward: tensor(0.0125, device='cuda:0', grad_fn=<MseLossBackward>)
  26973. loss_train_step after backward: tensor(0.0125, device='cuda:0', grad_fn=<MseLossBackward>)
  26974. loss_train: 0.055825852788984776
  26975. step: 3
  26976. running loss: 0.01860861759632826
  26977. Train Steps: 3/90 Loss: 0.0186 torch.Size([8, 600, 800])
  26978. torch.Size([8, 8])
  26979. tensor([[0.6257, 0.4034, 0.8287, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901],
  26980. [0.6134, 0.4090, 0.6926, 0.2819, 0.3538, 0.3233, 0.5563, 0.5667],
  26981. [0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689],
  26982. [0.6148, 0.3996, 0.8488, 0.3867, 0.3488, 0.4067, 0.5863, 0.5000],
  26983. [0.6161, 0.4076, 0.8900, 0.4667, 0.4125, 0.5917, 0.6262, 0.5367],
  26984. [0.6102, 0.4005, 0.8688, 0.5100, 0.4813, 0.5400, 0.5404, 0.5064],
  26985. [0.6222, 0.3957, 0.8838, 0.5017, 0.3937, 0.4600, 0.5900, 0.5017],
  26986. [0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133]],
  26987. device='cuda:0', dtype=torch.float64)
  26988. predictions are: tensor([[ 0.5987, -0.4661, 1.6236, -0.9908, -0.4434, -1.1789, 0.7196, 0.0251],
  26989. [ 0.5112, -0.4525, 1.0903, -0.8345, -0.6445, -0.8166, 0.3535, 0.3133],
  26990. [ 0.5388, -0.4309, 1.4379, -0.4941, -0.6618, -0.5530, 0.2934, 0.3137],
  26991. [ 0.7562, -0.3317, 1.7788, -0.3304, -0.7112, -0.4528, 0.5880, 0.0941],
  26992. [ 0.3007, -0.6169, 1.8800, 0.0517, -0.2237, 0.4643, 0.7065, 0.1734],
  26993. [ 0.7015, -0.3517, 1.8053, 0.2233, 0.0085, 0.0399, 0.4138, 0.1027],
  26994. [ 0.3611, -0.5971, 1.8793, 0.2300, -0.4435, -0.2307, 0.6075, 0.0287],
  26995. [ 0.8464, -0.2693, 1.6988, -0.9977, -0.0583, -1.6439, 0.7758, 0.0953]],
  26996. device='cuda:0', grad_fn=<AddmmBackward>)
  26997. landmarks are: tensor([[[ 0.6042, -0.4225, 1.5420, -1.2082, -0.4730, -1.0311, 0.6380,
  26998. -0.0220],
  26999. [ 0.5473, -0.3966, 0.9131, -0.9838, -0.6520, -0.7925, 0.2834,
  27000. 0.3315],
  27001. [ 0.5327, -0.4381, 1.2880, -0.6308, -0.6866, -0.5307, 0.2658,
  27002. 0.3417],
  27003. [ 0.5538, -0.4400, 1.6344, -0.4999, -0.6751, -0.4075, 0.4219,
  27004. 0.0236],
  27005. [ 0.5598, -0.4032, 1.8249, -0.1304, -0.3806, 0.4470, 0.6067,
  27006. 0.1929],
  27007. [ 0.5328, -0.4361, 1.7268, 0.0697, -0.0630, 0.2083, 0.2103,
  27008. 0.0532],
  27009. [ 0.5879, -0.4583, 1.7961, 0.0313, -0.4672, -0.1612, 0.4393,
  27010. 0.0313],
  27011. [ 0.6085, -0.4084, 1.5536, -1.1466, -0.0746, -1.4853, 0.6298,
  27012. 0.0851]]], device='cuda:0')
  27013. loss_train_step before backward: tensor(0.0153, device='cuda:0', grad_fn=<MseLossBackward>)
  27014. loss_train_step after backward: tensor(0.0153, device='cuda:0', grad_fn=<MseLossBackward>)
  27015. loss_train: 0.07109592575579882
  27016. step: 4
  27017. running loss: 0.017773981438949704
  27018. Train Steps: 4/90 Loss: 0.0178 torch.Size([8, 600, 800])
  27019. torch.Size([8, 8])
  27020. tensor([[0.6109, 0.4041, 0.6975, 0.3167, 0.3513, 0.3383, 0.5153, 0.5319],
  27021. [ nan, nan, 0.7192, 0.2346, 0.4037, 0.2050, 0.5138, 0.5650],
  27022. [0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222],
  27023. [0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412],
  27024. [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098],
  27025. [ nan, nan, 0.7425, 0.2117, 0.3937, 0.2433, 0.5438, 0.5567],
  27026. [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600],
  27027. [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717]],
  27028. device='cuda:0', dtype=torch.float64)
  27029. predictions are: tensor([[ 0.7477, -0.3178, 1.2301, -0.7151, -0.6086, -0.7426, 0.3208, 0.1037],
  27030. [-1.8269, -1.9756, 1.2389, -1.0472, -0.3977, -1.2951, 0.2191, 0.1287],
  27031. [ 1.0299, -0.0680, 1.6474, -0.3865, -0.0985, -1.2672, 0.6018, 0.2814],
  27032. [ 0.8511, -0.2321, 1.9707, -0.4202, -0.4431, -1.0457, 0.6645, 0.0311],
  27033. [ 1.0978, -0.1226, 1.8701, 0.1610, -0.4615, 0.1962, 0.8286, 0.0466],
  27034. [-1.8291, -1.9821, 1.4092, -1.0083, -0.3946, -1.1683, 0.3364, 0.0987],
  27035. [ 0.9370, -0.1767, 1.8685, 0.0554, -0.0661, 0.2181, 0.7632, 0.1283],
  27036. [ 1.0451, -0.1336, 1.0969, -0.9064, -0.4676, -1.1103, 0.4340, 0.2015]],
  27037. device='cuda:0', grad_fn=<AddmmBackward>)
  27038. landmarks are: tensor([[[ 0.5359, -0.4193, 0.9358, -0.8232, -0.6635, -0.7232, 0.0943,
  27039. 0.1710],
  27040. [-2.2859, -2.2859, 1.0361, -1.2021, -0.4210, -1.3390, 0.0871,
  27041. 0.3238],
  27042. [ 0.6237, -0.2983, 1.3919, -0.4691, -0.0457, -1.2313, 0.2453,
  27043. 0.5882],
  27044. [ 0.5784, -0.4006, 1.7911, -0.5701, -0.5192, -1.0331, 0.4137,
  27045. 0.2139],
  27046. [ 0.5466, -0.4706, 1.7198, -0.0903, -0.5712, 0.1261, 0.4733,
  27047. 0.0688],
  27048. [-2.2859, -2.2859, 1.1436, -1.3082, -0.4672, -1.1620, 0.2256,
  27049. 0.2853],
  27050. [ 0.5652, -0.4325, 1.7388, -0.1429, -0.0688, 0.3469, 0.4970,
  27051. 0.3007],
  27052. [ 0.5633, -0.3883, 0.8648, -1.0288, -0.5192, -1.1081, 0.2776,
  27053. 0.3546]]], device='cuda:0')
  27054. loss_train_step before backward: tensor(0.0528, device='cuda:0', grad_fn=<MseLossBackward>)
  27055. loss_train_step after backward: tensor(0.0528, device='cuda:0', grad_fn=<MseLossBackward>)
  27056. loss_train: 0.12389781419187784
  27057. step: 5
  27058. running loss: 0.02477956283837557
  27059.  
  27060. Train Steps: 5/90 Loss: 0.0248 torch.Size([8, 600, 800])
  27061. torch.Size([8, 8])
  27062. tensor([[0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261],
  27063. [0.6079, 0.3964, 0.7420, 0.2958, 0.3563, 0.2917, 0.5351, 0.4980],
  27064. [0.6276, 0.4120, 0.8738, 0.3133, 0.4225, 0.2217, 0.6203, 0.4892],
  27065. [0.6185, 0.4129, 0.8900, 0.4567, 0.3937, 0.5417, 0.5734, 0.5110],
  27066. [ nan, nan, 0.7225, 0.2167, 0.3987, 0.2283, 0.5427, 0.5181],
  27067. [0.6250, 0.4146, 0.8838, 0.3933, 0.3588, 0.4283, 0.6162, 0.5367],
  27068. [0.6129, 0.4069, 0.8750, 0.5067, 0.3875, 0.4233, 0.5235, 0.5881],
  27069. [0.6259, 0.4133, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947]],
  27070. device='cuda:0', dtype=torch.float64)
  27071. predictions are: tensor([[ 0.4496, -0.4924, 1.7601, 0.0577, -0.3639, -0.1169, 0.4771, 0.1180],
  27072. [ 0.6491, -0.3763, 1.2254, -0.8365, -0.5457, -0.9526, 0.2617, 0.0583],
  27073. [ 0.7002, -0.3444, 1.8390, -0.6355, -0.3098, -1.3207, 0.7930, 0.0159],
  27074. [ 0.8954, -0.2209, 1.8416, -0.0108, -0.3952, 0.1701, 0.6371, 0.1106],
  27075. [-2.1383, -2.1947, 1.1317, -1.1560, -0.4116, -1.2603, 0.1670, 0.1438],
  27076. [ 0.7399, -0.3151, 1.7280, -0.3167, -0.5821, -0.4310, 0.7327, 0.2390],
  27077. [ 0.7173, -0.3291, 1.7100, 0.2690, -0.4947, -0.3467, 0.2674, 0.3007],
  27078. [ 0.7036, -0.3440, 1.5613, -1.0072, 0.0162, -1.4975, 0.7395, 0.0891]],
  27079. device='cuda:0', grad_fn=<AddmmBackward>)
  27080. landmarks are: tensor([[[ 5.7460e-01, -3.6231e-01, 1.7961e+00, -1.1501e-01, -3.6905e-01,
  27081. -3.8029e-02, 2.2079e-01, 1.4394e-01],
  27082. [ 5.2194e-01, -4.5504e-01, 1.1415e+00, -9.1962e-01, -6.4042e-01,
  27083. -9.3872e-01, 1.8562e-01, 1.4106e-02],
  27084. [ 6.1282e-01, -3.8283e-01, 1.7499e+00, -8.3865e-01, -3.3441e-01,
  27085. -1.2620e+00, 5.7925e-01, -2.6256e-02],
  27086. [ 5.7113e-01, -3.7875e-01, 1.8249e+00, -1.7660e-01, -4.6721e-01,
  27087. 2.1601e-01, 3.6246e-01, 7.4222e-02],
  27088. [-2.2859e+00, -2.2859e+00, 1.0513e+00, -1.2851e+00, -4.4411e-01,
  27089. -1.2313e+00, 2.2057e-01, 1.0729e-01],
  27090. [ 6.0092e-01, -3.7098e-01, 1.7961e+00, -4.6913e-01, -6.2887e-01,
  27091. -3.0747e-01, 5.6051e-01, 1.9292e-01],
  27092. [ 5.4515e-01, -4.0670e-01, 1.7557e+00, 5.4350e-02, -4.9607e-01,
  27093. -3.3056e-01, 1.3228e-01, 4.3063e-01],
  27094. [ 6.0514e-01, -3.7714e-01, 1.5016e+00, -1.2159e+00, 3.5104e-02,
  27095. -1.5777e+00, 6.0099e-01, -9.2270e-04]]], device='cuda:0')
  27096. loss_train_step before backward: tensor(0.0142, device='cuda:0', grad_fn=<MseLossBackward>)
  27097. loss_train_step after backward: tensor(0.0142, device='cuda:0', grad_fn=<MseLossBackward>)
  27098. loss_train: 0.13813432399183512
  27099. step: 6
  27100. running loss: 0.02302238733197252
  27101. Train Steps: 6/90 Loss: 0.0230 torch.Size([8, 600, 800])
  27102. torch.Size([8, 8])
  27103. tensor([[0.6234, 0.4023, 0.8888, 0.4633, 0.3975, 0.5767, 0.6400, 0.5183],
  27104. [0.6228, 0.4004, 0.8750, 0.5250, 0.3825, 0.5233, 0.6362, 0.5000],
  27105. [0.6126, 0.4039, 0.8237, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138],
  27106. [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5787, 0.5117],
  27107. [0.6265, 0.4088, 0.8025, 0.1850, 0.4163, 0.2500, 0.6290, 0.4947],
  27108. [0.6257, 0.4034, 0.8287, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901],
  27109. [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650],
  27110. [0.6307, 0.4029, 0.8650, 0.5200, 0.3763, 0.4017, 0.7311, 0.5366]],
  27111. device='cuda:0', dtype=torch.float64)
  27112. predictions are: tensor([[ 0.5177, -0.5142, 1.8768, -0.0287, -0.3660, 0.3563, 0.6098, 0.1759],
  27113. [ 0.6169, -0.4642, 1.7793, 0.2798, -0.4347, 0.0232, 0.5844, 0.0657],
  27114. [ 0.5446, -0.4551, 1.5762, -0.3746, -0.5773, -0.6849, 0.3518, 0.4858],
  27115. [ 0.5052, -0.5054, 1.2804, -1.1753, -0.3175, -1.4387, 0.2268, 0.0866],
  27116. [ 0.4783, -0.5226, 1.5799, -1.2029, -0.2695, -1.2095, 0.7283, 0.0610],
  27117. [ 0.5183, -0.5346, 1.5911, -0.9976, -0.3980, -1.1554, 0.4931, 0.0675],
  27118. [ 0.4365, -0.5086, 1.7467, -0.3318, -0.5918, -0.4954, 0.2639, 0.3172],
  27119. [ 0.5277, -0.5272, 1.8241, 0.2649, -0.4634, -0.5806, 0.9055, 0.0354]],
  27120. device='cuda:0', grad_fn=<AddmmBackward>)
  27121. landmarks are: tensor([[[ 5.9336e-01, -4.2756e-01, 1.8192e+00, -1.4580e-01, -4.4988e-01,
  27122. 3.7768e-01, 6.7021e-01, 1.0824e-01],
  27123. [ 5.9082e-01, -4.3664e-01, 1.7557e+00, 1.3903e-01, -5.1917e-01,
  27124. 1.3133e-01, 6.5289e-01, 2.3557e-02],
  27125. [ 5.4376e-01, -4.2055e-01, 1.5189e+00, -4.5373e-01, -6.1155e-01,
  27126. -6.2309e-01, 4.3649e-01, 5.4914e-01],
  27127. [ 5.5978e-01, -4.2008e-01, 1.1898e+00, -1.3005e+00, -3.8060e-01,
  27128. -1.3313e+00, 3.8730e-01, 7.7444e-02],
  27129. [ 6.0785e-01, -3.9761e-01, 1.4208e+00, -1.4314e+00, -3.6328e-01,
  27130. -1.1312e+00, 6.1950e-01, -9.2270e-04],
  27131. [ 6.0421e-01, -4.2248e-01, 1.5420e+00, -1.2082e+00, -4.7298e-01,
  27132. -1.0311e+00, 6.3800e-01, -2.1963e-02],
  27133. [ 5.7569e-01, -3.9169e-01, 1.7095e+00, -4.7683e-01, -6.3464e-01,
  27134. -4.2294e-01, 3.9307e-01, 3.2379e-01],
  27135. [ 6.2730e-01, -4.2490e-01, 1.7095e+00, 1.1594e-01, -5.4804e-01,
  27136. -4.3064e-01, 1.0910e+00, 1.9283e-01]]], device='cuda:0')
  27137. loss_train_step before backward: tensor(0.0106, device='cuda:0', grad_fn=<MseLossBackward>)
  27138. loss_train_step after backward: tensor(0.0106, device='cuda:0', grad_fn=<MseLossBackward>)
  27139. loss_train: 0.14870771393179893
  27140. step: 7
  27141. running loss: 0.021243959133114134
  27142. Train Steps: 7/90 Loss: 0.0212 torch.Size([8, 600, 800])
  27143. torch.Size([8, 8])
  27144. tensor([[0.6216, 0.4099, 0.7225, 0.2033, 0.4188, 0.2217, 0.5975, 0.5283],
  27145. [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834],
  27146. [0.6129, 0.4063, 0.8738, 0.5250, 0.4313, 0.4733, 0.5230, 0.5874],
  27147. [0.6353, 0.4128, 0.9138, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991],
  27148. [0.6264, 0.3972, 0.8853, 0.4771, 0.3853, 0.4511, 0.6293, 0.5334],
  27149. [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967],
  27150. [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609],
  27151. [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5637, 0.5633]],
  27152. device='cuda:0', dtype=torch.float64)
  27153. predictions are: tensor([[ 0.5176, -0.4837, 1.0740, -1.1601, -0.4477, -1.2307, 0.2159, 0.1664],
  27154. [ 0.6377, -0.4477, 1.5012, -1.2757, -0.1299, -1.4302, 0.5491, -0.0120],
  27155. [ 0.5548, -0.4542, 1.6038, 0.3398, -0.4500, 0.0341, 0.0801, 0.2882],
  27156. [-0.2789, -1.0308, 1.9276, -0.5532, -0.2452, -0.9053, 1.0819, 0.2952],
  27157. [ 0.6326, -0.4496, 1.7682, 0.0528, -0.6594, -0.1615, 0.5851, 0.0823],
  27158. [ 0.3568, -0.5675, 1.6265, -1.0354, -0.0946, -1.2957, 0.5937, 0.0173],
  27159. [ 0.4348, -0.5392, 1.7455, -0.7572, -0.4726, -0.8393, 0.7825, 0.1674],
  27160. [ 0.4264, -0.4988, 1.6926, 0.0959, -0.7095, -0.5125, 0.3129, 0.2892]],
  27161. device='cuda:0', grad_fn=<AddmmBackward>)
  27162. landmarks are: tensor([[[ 5.8505e-01, -3.9246e-01, 1.0513e+00, -1.3467e+00, -3.5173e-01,
  27163. -1.2620e+00, 4.7390e-01, 1.5443e-01],
  27164. [ 6.1264e-01, -4.0570e-01, 1.4439e+00, -1.3159e+00, -1.1501e-01,
  27165. -1.5777e+00, 5.5366e-01, -5.2974e-02],
  27166. [ 5.4527e-01, -4.0908e-01, 1.7499e+00, 1.3903e-01, -2.9400e-01,
  27167. -9.9615e-02, 1.2997e-01, 4.2725e-01],
  27168. [ 6.4871e-01, -3.7916e-01, 1.9346e+00, -6.5389e-01, -1.2079e-01,
  27169. -7.8476e-01, 1.0143e+00, 4.8139e-01],
  27170. [ 6.0754e-01, -4.5138e-01, 1.8032e+00, -8.2167e-02, -5.0606e-01,
  27171. -2.0228e-01, 6.2076e-01, 1.7788e-01],
  27172. [ 6.0479e-01, -3.7229e-01, 1.6517e+00, -1.0773e+00, 4.6189e-04,
  27173. -1.5161e+00, 5.8660e-01, 8.0947e-03],
  27174. [ 6.0260e-01, -4.0323e-01, 1.7326e+00, -7.7706e-01, -3.6905e-01,
  27175. -8.6174e-01, 9.7040e-01, 3.0505e-01],
  27176. [ 5.9024e-01, -3.4927e-01, 1.7961e+00, -7.2363e-03, -5.9423e-01,
  27177. -5.6151e-01, 3.1801e-01, 3.1609e-01]]], device='cuda:0')
  27178. loss_train_step before backward: tensor(0.0322, device='cuda:0', grad_fn=<MseLossBackward>)
  27179. loss_train_step after backward: tensor(0.0322, device='cuda:0', grad_fn=<MseLossBackward>)
  27180. loss_train: 0.18095311149954796
  27181. step: 8
  27182. running loss: 0.022619138937443495
  27183.  
  27184. Train Steps: 8/90 Loss: 0.0226 torch.Size([8, 600, 800])
  27185. torch.Size([8, 8])
  27186. tensor([[0.6201, 0.4017, 0.8871, 0.4621, 0.3517, 0.4675, 0.5999, 0.5106],
  27187. [0.6164, 0.4102, 0.8850, 0.4867, 0.4213, 0.5983, 0.5609, 0.5038],
  27188. [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617],
  27189. [0.6149, 0.4054, 0.6713, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695],
  27190. [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395],
  27191. [0.6218, 0.4098, 0.7238, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350],
  27192. [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284],
  27193. [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500]],
  27194. device='cuda:0', dtype=torch.float64)
  27195. predictions are: tensor([[ 0.3702, -0.6153, 1.9485, 0.0180, -0.6251, -0.1912, 0.5284, 0.1265],
  27196. [ 0.3924, -0.5669, 1.9019, 0.1812, -0.2794, 0.3535, 0.4786, 0.1891],
  27197. [ 0.4712, -0.5183, 1.9494, -0.0167, -0.5042, -0.3461, 0.3179, 0.1923],
  27198. [ 0.0712, -0.7806, 1.0954, -1.0432, -0.4323, -1.4276, 0.1083, 0.2082],
  27199. [ 0.5518, -0.4915, 1.3219, -1.1241, -0.4176, -1.2363, 0.5847, 0.1265],
  27200. [ 0.8444, -0.3112, 1.2868, -1.1265, -0.3504, -1.2792, 0.7130, 0.1989],
  27201. [ 0.5735, -0.4880, 1.4809, -1.0457, -0.1821, -1.5007, 0.5285, 0.0796],
  27202. [ 0.3346, -0.6325, 1.4708, -0.8016, -0.6968, -0.4213, 0.6137, 0.1977]],
  27203. device='cuda:0', grad_fn=<AddmmBackward>)
  27204. landmarks are: tensor([[[ 0.5783, -0.4306, 1.8114, -0.1515, -0.6617, -0.1268, 0.4851,
  27205. 0.0727],
  27206. [ 0.5614, -0.3912, 1.8018, -0.0380, -0.3402, 0.4778, 0.3049,
  27207. 0.0412],
  27208. [ 0.5488, -0.4221, 1.8018, -0.2459, -0.4845, -0.3075, 0.2309,
  27209. 0.3087],
  27210. [ 0.5544, -0.4133, 0.8145, -1.2082, -0.4268, -1.3544, 0.1221,
  27211. 0.3446],
  27212. [ 0.5713, -0.4538, 1.1028, -1.3659, -0.3831, -1.1273, 0.5340,
  27213. 0.2058],
  27214. [ 0.5861, -0.3929, 1.0570, -1.4314, -0.3286, -1.1235, 0.6182,
  27215. 0.1852],
  27216. [ 0.5761, -0.4070, 1.2452, -1.2541, -0.1725, -1.4835, 0.4511,
  27217. 0.1545],
  27218. [ 0.5763, -0.4147, 1.3226, -1.0619, -0.6635, -0.4152, 0.5374,
  27219. 0.2545]]], device='cuda:0')
  27220. loss_train_step before backward: tensor(0.0257, device='cuda:0', grad_fn=<MseLossBackward>)
  27221. loss_train_step after backward: tensor(0.0257, device='cuda:0', grad_fn=<MseLossBackward>)
  27222. loss_train: 0.2066050637513399
  27223. step: 9
  27224. running loss: 0.022956118194593325
  27225. Train Steps: 9/90 Loss: 0.0230 torch.Size([8, 600, 800])
  27226. torch.Size([8, 8])
  27227. tensor([[0.6243, 0.4128, 0.7762, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417],
  27228. [0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204],
  27229. [0.6202, 0.4066, 0.8746, 0.3376, 0.3717, 0.3090, 0.5842, 0.5165],
  27230. [0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904],
  27231. [0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272],
  27232. [0.6245, 0.4115, 0.8700, 0.4883, 0.4625, 0.5517, 0.6100, 0.5217],
  27233. [0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986],
  27234. [0.6186, 0.4013, 0.8191, 0.3188, 0.4279, 0.2060, 0.5767, 0.5411]],
  27235. device='cuda:0', dtype=torch.float64)
  27236. predictions are: tensor([[ 0.2651, -0.6157, 1.2570, -1.0635, -0.5154, -0.8834, 0.6192, 0.2747],
  27237. [ 0.5491, -0.4932, 1.5250, 0.1891, -0.4677, 0.1058, 0.6565, 0.1900],
  27238. [ 0.3616, -0.5737, 1.6377, -0.7714, -0.6248, -0.9259, 0.4174, 0.1688],
  27239. [ 0.2940, -0.6039, 1.6172, -1.1136, -0.1859, -1.2848, 0.6837, 0.0900],
  27240. [ 0.3842, -0.5400, 1.5422, -0.7308, -0.4851, -1.1094, 0.1827, 0.2126],
  27241. [ 0.5567, -0.4578, 1.6566, -0.1017, -0.2839, 0.2518, 0.5364, 0.2497],
  27242. [ 0.4010, -0.5515, 1.8669, -0.6154, -0.2421, -1.3174, 0.5998, 0.1543],
  27243. [ 0.3493, -0.5728, 1.4240, -0.8848, -0.3612, -1.3296, 0.3260, 0.1651]],
  27244. device='cuda:0', grad_fn=<AddmmBackward>)
  27245. landmarks are: tensor([[[ 0.5977, -0.3792, 1.2995, -1.0311, -0.5192, -0.8386, 0.5836,
  27246. 0.2160],
  27247. [ 0.6042, -0.4273, 1.7198, 0.2184, -0.3478, 0.1149, 0.8062,
  27248. 0.1176],
  27249. [ 0.5787, -0.4077, 1.7537, -0.7263, -0.5692, -0.8586, 0.4126,
  27250. 0.1000],
  27251. [ 0.6126, -0.3871, 1.6229, -1.0773, -0.2132, -1.3698, 0.5829,
  27252. -0.0209],
  27253. [ 0.5711, -0.4015, 1.6979, -0.6770, -0.5365, -1.0619, 0.1712,
  27254. 0.1494],
  27255. [ 0.5987, -0.3852, 1.7326, -0.0303, -0.1497, 0.2622, 0.5316,
  27256. 0.1236],
  27257. [ 0.6054, -0.3641, 1.8711, -0.5153, -0.2305, -1.3005, 0.5948,
  27258. 0.0171],
  27259. [ 0.5713, -0.4321, 1.4975, -0.8134, -0.3094, -1.3345, 0.3779,
  27260. 0.2134]]], device='cuda:0')
  27261. loss_train_step before backward: tensor(0.0138, device='cuda:0', grad_fn=<MseLossBackward>)
  27262. loss_train_step after backward: tensor(0.0138, device='cuda:0', grad_fn=<MseLossBackward>)
  27263. loss_train: 0.22040296904742718
  27264. step: 10
  27265. running loss: 0.02204029690474272
  27266. Train Steps: 10/90 Loss: 0.0220 torch.Size([8, 600, 800])
  27267. torch.Size([8, 8])
  27268. tensor([[0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901],
  27269. [ nan, nan, 0.7525, 0.2291, 0.3838, 0.3017, 0.6050, 0.5667],
  27270. [0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183],
  27271. [0.6264, 0.3972, 0.8853, 0.4771, 0.3853, 0.4511, 0.6293, 0.5334],
  27272. [0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131],
  27273. [0.6200, 0.4118, 0.8287, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799],
  27274. [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960],
  27275. [0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117]],
  27276. device='cuda:0', dtype=torch.float64)
  27277. predictions are: tensor([[ 0.5814, -0.4430, 1.4538, -1.1155, -0.4904, -1.2293, 0.2936, -0.0107],
  27278. [-1.1313, -1.5455, 1.3149, -1.3389, -0.4164, -1.0452, 0.6576, 0.2857],
  27279. [ 0.7803, -0.3256, 1.7643, -0.2083, -0.5666, -0.6007, 0.4089, 0.1928],
  27280. [ 0.7986, -0.3300, 1.8533, -0.2059, -0.5074, -0.3374, 0.7314, 0.1213],
  27281. [ 0.5316, -0.4786, 1.6618, -0.5652, -0.6384, -0.3324, 0.4430, 0.1096],
  27282. [ 0.5207, -0.4637, 1.5275, -0.6416, -0.4408, -1.1248, 0.3544, 0.2698],
  27283. [ 0.5309, -0.4870, 1.8894, -0.3099, -0.1636, -0.0802, 0.3456, 0.0679],
  27284. [ 0.7446, -0.3162, 1.5247, 0.2188, -0.4249, -0.1904, 0.4876, 0.4739]],
  27285. device='cuda:0', grad_fn=<AddmmBackward>)
  27286. landmarks are: tensor([[[ 0.5515, -0.4129, 1.4785, -1.0080, -0.5192, -1.1004, 0.1034,
  27287. -0.0220],
  27288. [-2.2859, -2.2859, 1.1898, -1.2278, -0.5134, -0.8925, 0.5085,
  27289. 0.3315],
  27290. [ 0.5944, -0.4543, 1.8018, 0.0082, -0.6058, -0.4306, 0.4162,
  27291. 0.1082],
  27292. [ 0.6075, -0.4514, 1.8032, -0.0822, -0.5061, -0.2023, 0.6208,
  27293. 0.1779],
  27294. [ 0.5581, -0.4586, 1.5586, -0.3747, -0.6792, -0.2391, 0.4455,
  27295. 0.0840],
  27296. [ 0.5779, -0.3840, 1.5420, -0.4306, -0.5423, -0.9772, 0.2041,
  27297. 0.3928],
  27298. [ 0.5205, -0.4577, 1.7326, -0.1304, -0.1862, 0.0447, 0.1301,
  27299. 0.0051],
  27300. [ 0.6118, -0.3983, 1.5824, 0.3469, -0.4268, -0.0688, 0.3469,
  27301. 0.5393]]], device='cuda:0')
  27302. loss_train_step before backward: tensor(0.0416, device='cuda:0', grad_fn=<MseLossBackward>)
  27303. loss_train_step after backward: tensor(0.0416, device='cuda:0', grad_fn=<MseLossBackward>)
  27304. loss_train: 0.26203702203929424
  27305. step: 11
  27306. running loss: 0.023821547458117657
  27307. Train Steps: 11/90 Loss: 0.0238 torch.Size([8, 600, 800])
  27308. torch.Size([8, 8])
  27309. tensor([[0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412],
  27310. [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600],
  27311. [0.6286, 0.4055, 0.9000, 0.4717, 0.3763, 0.4683, 0.7018, 0.5494],
  27312. [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450],
  27313. [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817],
  27314. [0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117],
  27315. [0.6127, 0.4119, 0.8650, 0.5300, 0.4512, 0.5250, 0.5260, 0.5461],
  27316. [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892]],
  27317. device='cuda:0', dtype=torch.float64)
  27318. predictions are: tensor([[ 0.6080, -0.3856, 1.7336, -0.8982, -0.5061, -1.2318, 0.4152, 0.1341],
  27319. [ 0.4990, -0.4607, 1.6878, -0.4426, -0.1015, 0.0150, 0.4344, 0.2551],
  27320. [ 0.6212, -0.4318, 1.8470, -0.4213, -0.5903, -0.3445, 0.9861, 0.1653],
  27321. [ 0.5750, -0.4459, 1.7188, -0.4567, -0.6006, -0.5663, 0.4482, 0.2204],
  27322. [ 0.3180, -0.5996, 1.5562, 0.1008, -0.2646, -0.2010, 0.1075, 0.1433],
  27323. [ 0.6339, -0.4288, 1.7481, -0.3449, -0.6115, -0.3539, 0.6958, 0.1332],
  27324. [ 0.5634, -0.4427, 1.6698, -0.0794, -0.2126, -0.0684, 0.2679, 0.2849],
  27325. [ 0.5064, -0.4542, 1.6120, -0.4523, -0.6879, -0.7564, 0.0842, 0.3415]],
  27326. device='cuda:0', grad_fn=<AddmmBackward>)
  27327. landmarks are: tensor([[[ 0.5784, -0.4006, 1.7911, -0.5701, -0.5192, -1.0331, 0.4137,
  27328. 0.2139],
  27329. [ 0.5652, -0.4325, 1.7388, -0.1429, -0.0688, 0.3469, 0.4970,
  27330. 0.3007],
  27331. [ 0.6174, -0.4129, 1.8711, -0.1073, -0.5480, -0.1227, 0.9558,
  27332. 0.2516],
  27333. [ 0.5785, -0.4252, 1.7676, -0.1602, -0.5845, -0.3446, 0.4566,
  27334. 0.2314],
  27335. [ 0.5193, -0.4699, 1.5767, 0.4008, -0.2420, 0.0774, 0.1178,
  27336. -0.0610],
  27337. [ 0.6262, -0.4461, 1.8480, -0.0534, -0.5827, -0.1227, 0.6587,
  27338. 0.0774],
  27339. [ 0.5443, -0.3831, 1.7095, 0.1621, -0.2016, 0.1390, 0.1437,
  27340. 0.2364],
  27341. [ 0.5454, -0.4053, 1.6633, -0.1766, -0.6058, -0.5923, 0.1577,
  27342. 0.4357]]], device='cuda:0')
  27343. loss_train_step before backward: tensor(0.0211, device='cuda:0', grad_fn=<MseLossBackward>)
  27344. loss_train_step after backward: tensor(0.0211, device='cuda:0', grad_fn=<MseLossBackward>)
  27345. loss_train: 0.2831164803355932
  27346. step: 12
  27347. running loss: 0.0235930400279661
  27348.  
  27349. Train Steps: 12/90 Loss: 0.0236 torch.Size([8, 600, 800])
  27350. torch.Size([8, 8])
  27351. tensor([[0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374],
  27352. [0.6325, 0.4165, 0.9000, 0.4617, 0.3813, 0.4900, 0.7485, 0.5447],
  27353. [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683],
  27354. [0.6284, 0.4127, 0.8538, 0.5867, 0.4363, 0.5083, 0.6038, 0.5433],
  27355. [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100],
  27356. [0.6201, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044],
  27357. [0.6250, 0.4236, 0.8638, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799],
  27358. [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600]],
  27359. device='cuda:0', dtype=torch.float64)
  27360. predictions are: tensor([[ 0.6903, -0.3651, 1.6474, -0.8374, -0.6341, -0.9663, 0.3582, 0.1343],
  27361. [ 0.7031, -0.3525, 1.8754, -0.3281, -0.5801, -0.0378, 0.9083, 0.2030],
  27362. [ 0.4434, -0.4866, 1.6338, -0.1192, -0.5832, -0.6506, 0.1753, 0.3616],
  27363. [ 0.3422, -0.5627, 1.6275, 0.1652, -0.3122, -0.0549, 0.3189, 0.1649],
  27364. [ 0.5406, -0.4587, 1.6440, -0.0238, -0.5029, -0.2293, 0.4639, 0.0663],
  27365. [ 0.3885, -0.5420, 1.1723, -1.5524, -0.3675, -1.4431, 0.3967, 0.1097],
  27366. [ 0.6008, -0.3660, 1.6981, -0.7082, -0.4588, -0.9460, 0.3118, 0.4068],
  27367. [ 0.5218, -0.4222, 1.7080, -0.3800, -0.1079, 0.0906, 0.3519, 0.2555]],
  27368. device='cuda:0', grad_fn=<AddmmBackward>)
  27369. landmarks are: tensor([[[ 0.5777, -0.4416, 1.7044, -0.5827, -0.5962, -0.8361, 0.4862,
  27370. 0.1963],
  27371. [ 0.6355, -0.3623, 1.8711, -0.1535, -0.5249, -0.0226, 1.1715,
  27372. 0.2302],
  27373. [ 0.5777, -0.3915, 1.7961, 0.1698, -0.5192, -0.5307, 0.2141,
  27374. 0.3392],
  27375. [ 0.6166, -0.3795, 1.6575, 0.4239, -0.2709, 0.0620, 0.5028,
  27376. 0.2237],
  27377. [ 0.6125, -0.4273, 1.6402, 0.2006, -0.4788, -0.0303, 0.6182,
  27378. 0.0697],
  27379. [ 0.5785, -0.4129, 1.2785, -1.3996, -0.3227, -1.3259, 0.4258,
  27380. 0.0438],
  27381. [ 0.6008, -0.3293, 1.7037, -0.5461, -0.4152, -0.8309, 0.3234,
  27382. 0.3928],
  27383. [ 0.5652, -0.4325, 1.7388, -0.1429, -0.0688, 0.3469, 0.4970,
  27384. 0.3007]]], device='cuda:0')
  27385. loss_train_step before backward: tensor(0.0163, device='cuda:0', grad_fn=<MseLossBackward>)
  27386. loss_train_step after backward: tensor(0.0163, device='cuda:0', grad_fn=<MseLossBackward>)
  27387. loss_train: 0.29939468391239643
  27388. step: 13
  27389. running loss: 0.02303036030095357
  27390. Train Steps: 13/90 Loss: 0.0230 torch.Size([8, 600, 800])
  27391. torch.Size([8, 8])
  27392. tensor([[0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285],
  27393. [0.6087, 0.3976, 0.8337, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300],
  27394. [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100],
  27395. [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767],
  27396. [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852],
  27397. [ nan, nan, 0.6688, 0.2513, 0.4113, 0.2117, 0.5193, 0.5933],
  27398. [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202],
  27399. [0.6197, 0.4118, 0.8688, 0.5517, 0.4037, 0.5233, 0.5875, 0.5600]],
  27400. device='cuda:0', dtype=torch.float64)
  27401. predictions are: tensor([[ 0.5391, -0.4051, 1.3651, -0.6155, -0.6337, -0.3972, 0.0932, 0.1569],
  27402. [ 0.8467, -0.2399, 1.5689, -0.7761, -0.5896, -0.9227, 0.5025, 0.1898],
  27403. [ 0.7796, -0.2896, 1.7108, -0.0121, -0.4627, -0.1891, 0.6232, 0.0323],
  27404. [ 0.7671, -0.2283, 1.6868, -0.3612, -0.5792, -0.8732, 0.2958, 0.2853],
  27405. [ 0.6036, -0.3718, 1.7765, -0.1932, -0.1691, -0.1180, 0.2454, 0.3723],
  27406. [-2.3431, -2.3347, 0.9550, -1.3480, -0.4412, -1.4053, 0.1967, 0.3362],
  27407. [ 0.7980, -0.2530, 1.7662, -0.1792, -0.2901, 0.0418, 0.4128, 0.0634],
  27408. [ 0.7255, -0.2844, 1.7484, 0.0438, -0.4412, 0.1162, 0.4451, 0.2396]],
  27409. device='cuda:0', grad_fn=<AddmmBackward>)
  27410. landmarks are: tensor([[[ 0.5253, -0.4296, 1.3084, -0.3098, -0.6115, -0.2767, 0.0928,
  27411. 0.1552],
  27412. [ 0.5255, -0.4495, 1.5651, -0.4999, -0.5711, -0.8463, 0.4566,
  27413. 0.1621],
  27414. [ 0.6125, -0.4273, 1.6402, 0.2006, -0.4788, -0.0303, 0.6182,
  27415. 0.0697],
  27416. [ 0.5781, -0.3848, 1.7441, -0.0996, -0.5769, -0.8002, 0.2021,
  27417. 0.3778],
  27418. [ 0.5500, -0.4060, 1.7326, 0.0236, -0.1554, -0.0226, 0.1438,
  27419. 0.4171],
  27420. [-2.2859, -2.2859, 0.8033, -1.1250, -0.3864, -1.3082, 0.1126,
  27421. 0.4543],
  27422. [ 0.5614, -0.4586, 1.7587, 0.0641, -0.2998, 0.1712, 0.4958,
  27423. 0.1170],
  27424. [ 0.5762, -0.3840, 1.7268, 0.2622, -0.4210, 0.1313, 0.4277,
  27425. 0.3007]]], device='cuda:0')
  27426. loss_train_step before backward: tensor(0.0169, device='cuda:0', grad_fn=<MseLossBackward>)
  27427. loss_train_step after backward: tensor(0.0169, device='cuda:0', grad_fn=<MseLossBackward>)
  27428. loss_train: 0.3162560146301985
  27429. step: 14
  27430. running loss: 0.022589715330728462
  27431. Train Steps: 14/90 Loss: 0.0226 torch.Size([8, 600, 800])
  27432. torch.Size([8, 8])
  27433. tensor([[ nan, nan, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729],
  27434. [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364],
  27435. [0.6113, 0.4088, 0.6859, 0.2208, 0.4363, 0.1700, 0.5188, 0.5533],
  27436. [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852],
  27437. [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467],
  27438. [0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400],
  27439. [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131],
  27440. [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672]],
  27441. device='cuda:0', dtype=torch.float64)
  27442. predictions are: tensor([[-1.8744, -2.0117, 0.9690, -1.2382, -0.4163, -1.3135, 0.1242, 0.3034],
  27443. [ 0.5949, -0.3956, 1.3967, -1.1633, -0.2750, -1.1070, 0.7447, 0.1486],
  27444. [ 0.5934, -0.3800, 0.9055, -1.2207, -0.3371, -1.3213, 0.0684, 0.2919],
  27445. [ 0.4608, -0.4651, 1.8125, 0.0842, -0.1831, 0.1602, 0.2000, 0.3755],
  27446. [ 0.6010, -0.3735, 1.6042, -0.6678, -0.7322, -0.2244, 0.4141, 0.2086],
  27447. [ 0.7164, -0.2757, 1.4874, -0.8664, -0.1305, -1.1824, 0.3852, 0.1847],
  27448. [ 0.8126, -0.2737, 1.7947, -0.4520, -0.7088, -0.3828, 0.3791, 0.0978],
  27449. [ 0.6742, -0.3727, 1.8206, 0.3258, -0.6526, -0.1677, 0.5933, 0.0655]],
  27450. device='cuda:0', grad_fn=<AddmmBackward>)
  27451. landmarks are: tensor([[[-2.2859, -2.2859, 0.8824, -1.2727, -0.3691, -1.4237, 0.0943,
  27452. 0.3604],
  27453. [ 0.6158, -0.3960, 1.4092, -1.2774, -0.2074, -1.1851, 0.8491,
  27454. 0.1917],
  27455. [ 0.5377, -0.3978, 0.8824, -1.2663, -0.2709, -1.5007, 0.1102,
  27456. 0.2699],
  27457. [ 0.5500, -0.4060, 1.7326, 0.0236, -0.1554, -0.0226, 0.1438,
  27458. 0.4171],
  27459. [ 0.5484, -0.4591, 1.5940, -0.7771, -0.6693, -0.3460, 0.4219,
  27460. 0.2391],
  27461. [ 0.5911, -0.3888, 1.4727, -0.9541, -0.0919, -1.4930, 0.3988,
  27462. 0.2083],
  27463. [ 0.5783, -0.4216, 1.6847, -0.5078, -0.6732, -0.5377, 0.4752,
  27464. 0.0839],
  27465. [ 0.6128, -0.4375, 1.7788, 0.2699, -0.6346, -0.2536, 0.5463,
  27466. -0.1278]]], device='cuda:0')
  27467. loss_train_step before backward: tensor(0.0122, device='cuda:0', grad_fn=<MseLossBackward>)
  27468. loss_train_step after backward: tensor(0.0122, device='cuda:0', grad_fn=<MseLossBackward>)
  27469. loss_train: 0.32841287832707167
  27470. step: 15
  27471. running loss: 0.021894191888471446
  27472. Train Steps: 15/90 Loss: 0.0219 torch.Size([8, 600, 800])
  27473. torch.Size([8, 8])
  27474. tensor([[0.6257, 0.4167, 0.8775, 0.3433, 0.3563, 0.4133, 0.6200, 0.5667],
  27475. [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012],
  27476. [0.6164, 0.4119, 0.7913, 0.2650, 0.3538, 0.3500, 0.5614, 0.5038],
  27477. [0.6262, 0.4052, 0.8888, 0.4700, 0.3675, 0.5117, 0.6350, 0.5233],
  27478. [0.6249, 0.4142, 0.8350, 0.3283, 0.3613, 0.3700, 0.6188, 0.5400],
  27479. [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879],
  27480. [0.6151, 0.4085, 0.8750, 0.4367, 0.3887, 0.4367, 0.5066, 0.5846],
  27481. [0.6161, 0.4076, 0.8900, 0.4667, 0.4125, 0.5917, 0.6262, 0.5367]],
  27482. device='cuda:0', dtype=torch.float64)
  27483. predictions are: tensor([[ 0.6738, -0.3040, 1.6709, -0.5974, -0.5665, -0.4662, 0.4484, 0.4006],
  27484. [ 0.6564, -0.3728, 1.6862, 0.2819, -0.4988, -0.4484, 0.4865, 0.1666],
  27485. [ 0.4000, -0.5246, 1.2383, -1.0922, -0.6371, -0.8293, 0.3081, 0.1241],
  27486. [ 0.6967, -0.3275, 1.8110, -0.0835, -0.4345, -0.0513, 0.5109, 0.1657],
  27487. [ 0.5529, -0.3873, 1.5040, -0.7385, -0.5744, -0.7136, 0.4337, 0.3210],
  27488. [ 0.6056, -0.3954, 1.7734, -0.2286, -0.3551, 0.1743, 0.3885, 0.1289],
  27489. [ 0.3163, -0.5174, 1.6173, -0.2200, -0.4569, -0.3970, -0.0269, 0.4105],
  27490. [ 0.4139, -0.4945, 1.7536, -0.1140, -0.1944, 0.3604, 0.4320, 0.2440]],
  27491. device='cuda:0', grad_fn=<AddmmBackward>)
  27492. landmarks are: tensor([[[ 0.6040, -0.3614, 1.7672, -0.7001, -0.6404, -0.3768, 0.5778,
  27493. 0.3315],
  27494. [ 0.6249, -0.4352, 1.8018, 0.2545, -0.6173, -0.1997, 0.6401,
  27495. 0.0291],
  27496. [ 0.5614, -0.3834, 1.3688, -1.0619, -0.6520, -0.6693, 0.3069,
  27497. 0.0412],
  27498. [ 0.6063, -0.4142, 1.8192, -0.1150, -0.5885, 0.0774, 0.6471,
  27499. 0.1313],
  27500. [ 0.6006, -0.3728, 1.5709, -0.7694, -0.6173, -0.5769, 0.5721,
  27501. 0.2083],
  27502. [ 0.5676, -0.4327, 1.8252, -0.2095, -0.4886, 0.3084, 0.4727,
  27503. -0.0322],
  27504. [ 0.5553, -0.3992, 1.7557, -0.2690, -0.4903, -0.2690, 0.0542,
  27505. 0.4145],
  27506. [ 0.5598, -0.4032, 1.8249, -0.1304, -0.3806, 0.4470, 0.6067,
  27507. 0.1929]]], device='cuda:0')
  27508. loss_train_step before backward: tensor(0.0115, device='cuda:0', grad_fn=<MseLossBackward>)
  27509. loss_train_step after backward: tensor(0.0115, device='cuda:0', grad_fn=<MseLossBackward>)
  27510. loss_train: 0.3399082263931632
  27511. step: 16
  27512. running loss: 0.0212442641495727
  27513.  
  27514. Train Steps: 16/90 Loss: 0.0212 torch.Size([8, 600, 800])
  27515. torch.Size([8, 8])
  27516. tensor([[0.6179, 0.4008, 0.7505, 0.2678, 0.4368, 0.1891, 0.5831, 0.5263],
  27517. [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679],
  27518. [0.6289, 0.4032, 0.8419, 0.5446, 0.4075, 0.5017, 0.6312, 0.5117],
  27519. [0.6263, 0.4233, 0.7924, 0.4626, 0.3788, 0.2883, 0.5573, 0.6047],
  27520. [0.6336, 0.4191, 0.8938, 0.5167, 0.3937, 0.3517, 0.7343, 0.5748],
  27521. [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650],
  27522. [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450],
  27523. [ nan, nan, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533]],
  27524. device='cuda:0', dtype=torch.float64)
  27525. predictions are: tensor([[ 0.7029, -0.2821, 1.0760, -1.2492, -0.3319, -1.2570, 0.1479, 0.1191],
  27526. [ 0.7587, -0.2442, 1.6849, -0.0106, -0.3610, 0.4631, 0.6869, 0.2252],
  27527. [ 0.5850, -0.3844, 1.6199, 0.0817, -0.3890, 0.0821, 0.4641, 0.0563],
  27528. [ 0.7050, -0.2352, 1.3946, -0.2707, -0.4803, -0.7746, 0.0691, 0.4567],
  27529. [ 0.6798, -0.3020, 1.7941, 0.0049, -0.3509, -0.4692, 0.7393, 0.2167],
  27530. [ 0.6755, -0.2483, 1.5387, -0.4930, -0.5451, -0.6314, 0.1540, 0.3930],
  27531. [ 0.6301, -0.3319, 1.7693, -0.2778, -0.5249, -0.2802, 0.2878, 0.1862],
  27532. [-2.3278, -2.2892, 1.3048, -0.9943, -0.5037, -0.9370, 0.2053, 0.2479]],
  27533. device='cuda:0', grad_fn=<AddmmBackward>)
  27534. landmarks are: tensor([[[ 0.5680, -0.4345, 1.1806, -1.0490, -0.2683, -1.4127, 0.4074,
  27535. 0.1449],
  27536. [ 0.6257, -0.4249, 1.5998, 0.1236, -0.3806, 0.3084, 0.9887,
  27537. 0.3371],
  27538. [ 0.6189, -0.4238, 1.6026, 0.2295, -0.4037, 0.0313, 0.6298,
  27539. 0.0774],
  27540. [ 0.6069, -0.3309, 1.3742, -0.1493, -0.5365, -0.9541, 0.2884,
  27541. 0.5071],
  27542. [ 0.6405, -0.3503, 1.8423, 0.1005, -0.4672, -0.6616, 1.1057,
  27543. 0.3692],
  27544. [ 0.5894, -0.3503, 1.6402, -0.3614, -0.5827, -0.7925, 0.3238,
  27545. 0.3238],
  27546. [ 0.5785, -0.4252, 1.7676, -0.1602, -0.5845, -0.3446, 0.4566,
  27547. 0.2314],
  27548. [-2.2859, -2.2859, 1.5074, -1.0388, -0.5423, -0.9849, 0.2199,
  27549. 0.2699]]], device='cuda:0')
  27550. loss_train_step before backward: tensor(0.0158, device='cuda:0', grad_fn=<MseLossBackward>)
  27551. loss_train_step after backward: tensor(0.0158, device='cuda:0', grad_fn=<MseLossBackward>)
  27552. loss_train: 0.3556927992030978
  27553. step: 17
  27554. running loss: 0.020923105835476342
  27555. Train Steps: 17/90 Loss: 0.0209 torch.Size([8, 600, 800])
  27556. torch.Size([8, 8])
  27557. tensor([[0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494],
  27558. [0.6329, 0.4196, 0.9238, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748],
  27559. [0.6200, 0.4071, 0.7338, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517],
  27560. [0.6043, 0.4022, 0.6887, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136],
  27561. [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567],
  27562. [0.6203, 0.4096, 0.8862, 0.4267, 0.3538, 0.4117, 0.6025, 0.5650],
  27563. [0.6251, 0.4163, 0.8662, 0.4467, 0.3625, 0.3567, 0.6038, 0.5533],
  27564. [0.6109, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117]],
  27565. device='cuda:0', dtype=torch.float64)
  27566. predictions are: tensor([[ 0.6595, -0.3308, 1.7053, 0.3498, -0.4550, -0.0557, 0.7336, 0.1286],
  27567. [ 0.5503, -0.4199, 1.9664, -0.1062, -0.2386, -0.7404, 0.8984, 0.3366],
  27568. [ 0.4786, -0.4499, 1.0389, -1.3838, -0.2911, -0.9671, 0.3889, 0.2742],
  27569. [ 0.3249, -0.5281, 0.7924, -1.2884, -0.4708, -0.9557, 0.1264, 0.1307],
  27570. [ 0.3706, -0.5244, 1.7032, -0.2031, -0.5283, -0.3190, 0.2407, 0.3382],
  27571. [ 0.5839, -0.3540, 1.7904, -0.3297, -0.4904, -0.1844, 0.3168, 0.3530],
  27572. [ 0.2467, -0.5829, 1.6835, -0.3035, -0.5309, -0.5152, 0.2587, 0.3660],
  27573. [ 0.6686, -0.3094, 1.7843, -0.1293, -0.5499, -0.2107, 0.1485, 0.0952]],
  27574. device='cuda:0', grad_fn=<AddmmBackward>)
  27575. landmarks are: tensor([[[ 0.6224, -0.4032, 1.5940, 0.2930, -0.5769, -0.2690, 0.8900,
  27576. 0.2516],
  27577. [ 0.6372, -0.3479, 1.9808, -0.1689, -0.3171, -0.9310, 1.1057,
  27578. 0.3692],
  27579. [ 0.5779, -0.4054, 1.1032, -1.4006, -0.3460, -1.1543, 0.5547,
  27580. 0.2622],
  27581. [ 0.5053, -0.4281, 0.8954, -1.3698, -0.5423, -1.1389, 0.2453,
  27582. 0.0862],
  27583. [ 0.5432, -0.4336, 1.7095, -0.1766, -0.5942, -0.4845, 0.3007,
  27584. 0.2853],
  27585. [ 0.5790, -0.3940, 1.8076, -0.3152, -0.6520, -0.3844, 0.4970,
  27586. 0.3238],
  27587. [ 0.6016, -0.3633, 1.7152, -0.2228, -0.6115, -0.6385, 0.5028,
  27588. 0.2699],
  27589. [ 0.5356, -0.4343, 1.7441, -0.1150, -0.6462, -0.3614, 0.3238,
  27590. 0.0774]]], device='cuda:0')
  27591. loss_train_step before backward: tensor(0.0152, device='cuda:0', grad_fn=<MseLossBackward>)
  27592. loss_train_step after backward: tensor(0.0152, device='cuda:0', grad_fn=<MseLossBackward>)
  27593. loss_train: 0.3708633789792657
  27594. step: 18
  27595. running loss: 0.020603521054403648
  27596. Train Steps: 18/90 Loss: 0.0206 torch.Size([8, 600, 800])
  27597. torch.Size([8, 8])
  27598. tensor([[0.6236, 0.3977, 0.8985, 0.4806, 0.3835, 0.5216, 0.6613, 0.5166],
  27599. [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609],
  27600. [0.6196, 0.4068, 0.7645, 0.2234, 0.4575, 0.1737, 0.5926, 0.5284],
  27601. [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133],
  27602. [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285],
  27603. [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398],
  27604. [0.6263, 0.4039, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804],
  27605. [0.6202, 0.4054, 0.8892, 0.4428, 0.3642, 0.3249, 0.5840, 0.5426]],
  27606. device='cuda:0', dtype=torch.float64)
  27607. predictions are: tensor([[ 0.3771, -0.5383, 1.7638, -0.0992, -0.5666, 0.2643, 0.5956, 0.2325],
  27608. [ 0.7077, -0.2939, 1.6108, -1.1808, 0.1599, -1.2673, 0.9221, 0.3429],
  27609. [ 0.5240, -0.3895, 1.0684, -1.2673, -0.2932, -1.2565, 0.1783, 0.2553],
  27610. [ 0.3992, -0.4817, 1.6507, 0.1151, -0.2505, 0.2378, 0.2202, 0.2257],
  27611. [ 0.4623, -0.4594, 1.6011, 0.3008, -0.5909, 0.1249, 0.5551, 0.2070],
  27612. [ 0.6439, -0.3487, 1.6420, 0.2801, -0.6249, -0.1156, 0.3183, 0.2343],
  27613. [ 0.3152, -0.5551, 1.8351, -0.2781, -0.6664, -0.0461, 0.4377, 0.0753],
  27614. [ 0.6804, -0.2900, 1.7184, -0.2531, -0.6444, -0.5595, 0.2237, 0.3050]],
  27615. device='cuda:0', grad_fn=<AddmmBackward>)
  27616. landmarks are: tensor([[[ 5.9436e-01, -4.4897e-01, 1.8643e+00, -6.5918e-02, -5.1472e-01,
  27617. 1.2348e-01, 7.6842e-01, 1.0043e-01],
  27618. [ 6.3718e-01, -3.5749e-01, 1.6633e+00, -1.2082e+00, 2.1986e-01,
  27619. -1.2467e+00, 1.1313e+00, 3.0505e-01],
  27620. [ 5.7610e-01, -4.0701e-01, 1.2452e+00, -1.2541e+00, -1.7255e-01,
  27621. -1.4835e+00, 4.5107e-01, 1.5453e-01],
  27622. [ 5.7864e-01, -4.1409e-01, 1.7037e+00, 1.5443e-01, -1.8624e-01,
  27623. 7.3556e-02, 4.3926e-01, 8.5142e-02],
  27624. [ 6.5365e-01, -3.9601e-01, 1.6517e+00, 3.1609e-01, -4.9607e-01,
  27625. 4.6189e-04, 7.6203e-01, 1.5543e-01],
  27626. [ 5.7864e-01, -4.4627e-01, 1.6655e+00, 2.2157e-01, -5.1146e-01,
  27627. -2.6752e-01, 4.2362e-01, 2.0749e-01],
  27628. [ 6.0716e-01, -4.2055e-01, 1.8711e+00, -2.5358e-01, -6.1155e-01,
  27629. -1.3041e-01, 6.8119e-01, -6.7050e-02],
  27630. [ 5.7875e-01, -4.1347e-01, 1.8214e+00, -2.4075e-01, -6.0389e-01,
  27631. -7.8543e-01, 4.1155e-01, 2.2033e-01]]], device='cuda:0')
  27632. loss_train_step before backward: tensor(0.0161, device='cuda:0', grad_fn=<MseLossBackward>)
  27633. loss_train_step after backward: tensor(0.0161, device='cuda:0', grad_fn=<MseLossBackward>)
  27634. loss_train: 0.3869306379929185
  27635. step: 19
  27636. running loss: 0.020364770420679922
  27637. Train Steps: 19/90 Loss: 0.0204 torch.Size([8, 600, 800])
  27638. torch.Size([8, 8])
  27639. tensor([[0.6205, 0.4016, 0.8350, 0.2717, 0.3987, 0.2550, 0.5787, 0.5133],
  27640. [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750],
  27641. [0.6095, 0.3970, 0.8688, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940],
  27642. [0.6022, 0.3994, 0.8025, 0.3350, 0.3350, 0.4400, 0.5565, 0.5025],
  27643. [0.6250, 0.4131, 0.8688, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367],
  27644. [0.6200, 0.4098, 0.8237, 0.2917, 0.4012, 0.2967, 0.6000, 0.5683],
  27645. [0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494],
  27646. [0.6278, 0.4253, 0.8875, 0.5017, 0.4113, 0.2750, 0.5413, 0.6196]],
  27647. device='cuda:0', dtype=torch.float64)
  27648. predictions are: tensor([[ 0.4445, -0.4425, 1.4694, -0.9243, -0.4568, -0.9045, 0.4037, 0.1216],
  27649. [ 0.5492, -0.4187, 1.6414, 0.1282, -0.3174, 0.0859, 0.3000, 0.2823],
  27650. [ 0.4324, -0.4828, 1.6951, 0.0454, -0.1453, 0.1048, 0.2725, 0.0768],
  27651. [ 0.2279, -0.6074, 1.3656, -0.6036, -0.7623, -0.1955, 0.3766, 0.1503],
  27652. [ 0.1854, -0.5919, 1.6488, -0.8114, -0.3160, -1.0299, 0.6023, 0.2372],
  27653. [ 0.5083, -0.4121, 1.3570, -0.8879, -0.4907, -0.6800, 0.4704, 0.3468],
  27654. [ 0.5510, -0.4482, 1.7670, -0.4521, -0.4118, -0.3523, 1.0814, 0.2116],
  27655. [ 0.7426, -0.2122, 1.5960, 0.2323, -0.3715, -0.8716, 0.3507, 0.4529]],
  27656. device='cuda:0', grad_fn=<AddmmBackward>)
  27657. landmarks are: tensor([[[ 0.5800, -0.4312, 1.5709, -1.0311, -0.4441, -1.1081, 0.3873,
  27658. 0.0851],
  27659. [ 0.5525, -0.4463, 1.7326, 0.0313, -0.2536, -0.0688, 0.1968,
  27660. 0.3700],
  27661. [ 0.5292, -0.4520, 1.7268, -0.0842, -0.0413, -0.0324, 0.1116,
  27662. -0.0039],
  27663. [ 0.4956, -0.4413, 1.4208, -0.7386, -0.7386, -0.2536, 0.2843,
  27664. 0.0351],
  27665. [ 0.6011, -0.3778, 1.7268, -0.9079, -0.3113, -1.1928, 0.5605,
  27666. 0.1929],
  27667. [ 0.5778, -0.3930, 1.5189, -0.9387, -0.4326, -0.9156, 0.4855,
  27668. 0.3392],
  27669. [ 0.6388, -0.3792, 1.9635, -0.6616, -0.2536, -0.5153, 1.1605,
  27670. 0.2516],
  27671. [ 0.6139, -0.3216, 1.8134, 0.0313, -0.3864, -1.0157, 0.2144,
  27672. 0.5762]]], device='cuda:0')
  27673. loss_train_step before backward: tensor(0.0167, device='cuda:0', grad_fn=<MseLossBackward>)
  27674. loss_train_step after backward: tensor(0.0167, device='cuda:0', grad_fn=<MseLossBackward>)
  27675. loss_train: 0.40361075196415186
  27676. step: 20
  27677. running loss: 0.020180537598207593
  27678.  
  27679. Train Steps: 20/90 Loss: 0.0202 torch.Size([8, 600, 800])
  27680. torch.Size([8, 8])
  27681. tensor([[0.6203, 0.4073, 0.8189, 0.2398, 0.4400, 0.2054, 0.5929, 0.5501],
  27682. [ nan, nan, 0.8850, 0.2817, 0.5112, 0.2183, 0.7184, 0.5436],
  27683. [0.6162, 0.4134, 0.6700, 0.2467, 0.3962, 0.2533, 0.5737, 0.5467],
  27684. [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103],
  27685. [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896],
  27686. [0.6111, 0.3995, 0.8788, 0.4567, 0.3813, 0.4833, 0.5450, 0.5700],
  27687. [0.6145, 0.4008, 0.8750, 0.5383, 0.3975, 0.4650, 0.5563, 0.5533],
  27688. [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973]],
  27689. device='cuda:0', dtype=torch.float64)
  27690. predictions are: tensor([[ 0.3955, -0.4749, 1.3898, -0.9947, -0.2276, -1.1466, 0.4303, 0.2292],
  27691. [-0.9126, -1.3647, 1.7853, -1.0556, 0.1312, -1.1487, 1.0912, 0.3533],
  27692. [ 0.5603, -0.3804, 0.8494, -1.0202, -0.4774, -1.0772, 0.3581, 0.2918],
  27693. [ 0.5956, -0.3725, 1.5091, -0.6703, -0.6470, -0.3083, 0.4570, 0.1711],
  27694. [ 0.4852, -0.4004, 1.4856, -0.3499, -0.6665, -0.7114, 0.0906, 0.3698],
  27695. [ 0.6508, -0.3468, 1.8328, -0.0150, -0.5481, 0.1108, 0.3329, 0.1858],
  27696. [ 0.6672, -0.3458, 1.7533, 0.3656, -0.3711, 0.0669, 0.4311, 0.2554],
  27697. [ 0.7912, -0.3052, 1.8702, 0.3306, -0.6130, -0.0232, 0.8078, -0.0553]],
  27698. device='cuda:0', grad_fn=<AddmmBackward>)
  27699. landmarks are: tensor([[[ 0.5792, -0.4048, 1.4965, -1.1781, -0.2534, -1.3371, 0.4528,
  27700. 0.2549],
  27701. [-2.2859, -2.2859, 1.8018, -0.9849, 0.0755, -1.2774, 1.0326,
  27702. 0.2249],
  27703. [ 0.5603, -0.3764, 0.8088, -1.1466, -0.4557, -1.1158, 0.3642,
  27704. 0.2391],
  27705. [ 0.5886, -0.3784, 1.4554, -0.9079, -0.6577, -0.4845, 0.3440,
  27706. 0.0712],
  27707. [ 0.5485, -0.3997, 1.4445, -0.4895, -0.6000, -0.8309, 0.1878,
  27708. 0.4374],
  27709. [ 0.5368, -0.4406, 1.7730, -0.1766, -0.5249, -0.0534, 0.2314,
  27710. 0.3469],
  27711. [ 0.5526, -0.4347, 1.7557, 0.2006, -0.4499, -0.1381, 0.2834,
  27712. 0.2699],
  27713. [ 0.6240, -0.4321, 1.8423, 0.1852, -0.5885, -0.1612, 0.6962,
  27714. 0.0111]]], device='cuda:0')
  27715. loss_train_step before backward: tensor(0.0536, device='cuda:0', grad_fn=<MseLossBackward>)
  27716. loss_train_step after backward: tensor(0.0536, device='cuda:0', grad_fn=<MseLossBackward>)
  27717. loss_train: 0.45722213480621576
  27718. step: 21
  27719. running loss: 0.021772482609819798
  27720. Train Steps: 21/90 Loss: 0.0218 torch.Size([8, 600, 800])
  27721. torch.Size([8, 8])
  27722. tensor([[0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268],
  27723. [0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433],
  27724. [0.6112, 0.4029, 0.8638, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567],
  27725. [0.6199, 0.4102, 0.8950, 0.4417, 0.4012, 0.5367, 0.6112, 0.5967],
  27726. [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633],
  27727. [0.6222, 0.3957, 0.8838, 0.5017, 0.3937, 0.4600, 0.5900, 0.5017],
  27728. [0.6111, 0.3995, 0.8788, 0.4567, 0.3813, 0.4833, 0.5450, 0.5700],
  27729. [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690]],
  27730. device='cuda:0', dtype=torch.float64)
  27731. predictions are: tensor([[ 0.4032, -0.5113, 1.7258, -0.0226, -0.2910, -0.1014, 0.5768, 0.1624],
  27732. [ 0.6011, -0.3635, 1.2942, -1.0630, -0.6989, -0.7143, 0.5331, 0.1748],
  27733. [ 0.5156, -0.4315, 1.7901, -0.0675, -0.0420, -0.1152, 0.4514, 0.2772],
  27734. [ 0.4480, -0.4534, 1.7835, -0.2277, -0.3825, 0.1414, 0.4593, 0.3638],
  27735. [ 0.6085, -0.3451, 1.7743, 0.1849, -0.5596, -0.3062, 0.3344, 0.4080],
  27736. [ 0.6167, -0.3827, 1.8204, 0.0671, -0.5119, -0.2906, 0.4843, -0.0131],
  27737. [ 0.6249, -0.3598, 1.8184, -0.1034, -0.5868, -0.1451, 0.3213, 0.2177],
  27738. [ 0.6213, -0.3791, 1.5676, 0.2199, -0.4263, 0.1815, 1.0220, 0.3422]],
  27739. device='cuda:0', grad_fn=<AddmmBackward>)
  27740. landmarks are: tensor([[[ 0.5828, -0.4682, 1.7031, -0.0497, -0.2458, 0.0818, 0.6381,
  27741. 0.1474],
  27742. [ 0.5665, -0.3963, 1.3457, -0.9926, -0.6058, -0.6308, 0.5721,
  27743. 0.2237],
  27744. [ 0.5371, -0.4252, 1.7037, -0.0688, -0.0342, 0.0620, 0.3758,
  27745. 0.2853],
  27746. [ 0.5772, -0.3913, 1.8480, -0.2459, -0.4326, 0.1929, 0.5374,
  27747. 0.4701],
  27748. [ 0.5879, -0.3591, 1.8018, 0.1236, -0.5596, -0.1612, 0.3469,
  27749. 0.3161],
  27750. [ 0.5879, -0.4583, 1.7961, 0.0313, -0.4672, -0.1612, 0.4393,
  27751. 0.0313],
  27752. [ 0.5368, -0.4406, 1.7730, -0.1766, -0.5249, -0.0534, 0.2314,
  27753. 0.3469],
  27754. [ 0.6306, -0.4153, 1.5141, 0.2224, -0.3691, 0.2622, 1.0033,
  27755. 0.3425]]], device='cuda:0')
  27756. loss_train_step before backward: tensor(0.0053, device='cuda:0', grad_fn=<MseLossBackward>)
  27757. loss_train_step after backward: tensor(0.0053, device='cuda:0', grad_fn=<MseLossBackward>)
  27758. loss_train: 0.46247611194849014
  27759. step: 22
  27760. running loss: 0.021021641452204098
  27761. Train Steps: 22/90 Loss: 0.0210 torch.Size([8, 600, 800])
  27762. torch.Size([8, 8])
  27763. tensor([[0.6275, 0.4024, 0.8600, 0.2283, 0.5350, 0.1800, 0.7074, 0.5413],
  27764. [0.6053, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168],
  27765. [0.6048, 0.3928, 0.8538, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719],
  27766. [0.6166, 0.4008, 0.8563, 0.5667, 0.4388, 0.4933, 0.5575, 0.5567],
  27767. [0.6128, 0.4118, 0.8638, 0.5333, 0.4625, 0.5267, 0.5193, 0.5475],
  27768. [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261],
  27769. [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274],
  27770. [0.6182, 0.3967, 0.8263, 0.3065, 0.3526, 0.4161, 0.6192, 0.5290]],
  27771. device='cuda:0', dtype=torch.float64)
  27772. predictions are: tensor([[ 0.4703, -0.4666, 1.7351, -1.2481, 0.1270, -1.5093, 1.1119, 0.2799],
  27773. [ 0.5133, -0.4259, 0.8617, -1.3169, -0.4766, -1.3048, 0.2593, 0.1520],
  27774. [ 0.2934, -0.5855, 1.6849, 0.3353, -0.5495, 0.0145, 0.3481, 0.2117],
  27775. [ 0.6970, -0.3230, 1.6839, 0.4214, -0.3112, 0.0768, 0.3962, 0.3051],
  27776. [ 0.5945, -0.3530, 1.7217, 0.2182, -0.2332, 0.1234, 0.2404, 0.2736],
  27777. [ 0.4804, -0.4319, 1.8376, -0.0268, -0.4507, -0.0504, 0.3362, 0.2036],
  27778. [ 0.5193, -0.4420, 1.8981, 0.0182, -0.6593, -0.2885, 0.9374, 0.2386],
  27779. [ 0.4919, -0.4813, 1.6119, -0.7349, -0.7653, -0.3712, 0.6983, 0.1223]],
  27780. device='cuda:0', grad_fn=<AddmmBackward>)
  27781. landmarks are: tensor([[[ 0.6125, -0.4273, 1.6864, -1.2313, 0.1852, -1.4545, 0.9814,
  27782. 0.2142],
  27783. [ 0.5102, -0.4223, 0.8999, -1.4011, -0.4383, -1.3082, 0.2267,
  27784. 0.1013],
  27785. [ 0.5079, -0.4714, 1.6575, 0.2237, -0.4961, 0.0774, 0.1465,
  27786. -0.1061],
  27787. [ 0.5620, -0.4346, 1.6691, 0.3315, -0.2594, -0.0072, 0.2891,
  27788. 0.2853],
  27789. [ 0.5447, -0.3838, 1.7037, 0.1775, -0.1497, 0.1467, 0.1128,
  27790. 0.2431],
  27791. [ 0.5746, -0.3623, 1.7961, -0.1150, -0.3691, -0.0380, 0.2208,
  27792. 0.1439],
  27793. [ 0.6108, -0.4008, 1.9088, -0.0253, -0.5769, -0.3075, 0.7905,
  27794. 0.1499],
  27795. [ 0.5697, -0.4538, 1.5308, -0.8703, -0.6572, -0.3639, 0.5739,
  27796. 0.1576]]], device='cuda:0')
  27797. loss_train_step before backward: tensor(0.0083, device='cuda:0', grad_fn=<MseLossBackward>)
  27798. loss_train_step after backward: tensor(0.0083, device='cuda:0', grad_fn=<MseLossBackward>)
  27799. loss_train: 0.4708037283271551
  27800. step: 23
  27801. running loss: 0.02046972731857196
  27802. Train Steps: 23/90 Loss: 0.0205 torch.Size([8, 600, 800])
  27803. torch.Size([8, 8])
  27804. tensor([[0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904],
  27805. [0.6265, 0.4071, 0.8875, 0.3367, 0.3975, 0.3350, 0.6312, 0.5250],
  27806. [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012],
  27807. [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167],
  27808. [0.6137, 0.4038, 0.8563, 0.4050, 0.3813, 0.2550, 0.5106, 0.4954],
  27809. [0.6207, 0.4110, 0.8738, 0.5000, 0.4800, 0.5633, 0.6300, 0.5433],
  27810. [0.6307, 0.4029, 0.8988, 0.4817, 0.3937, 0.3500, 0.7311, 0.5378],
  27811. [0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6138, 0.5450]],
  27812. device='cuda:0', dtype=torch.float64)
  27813. predictions are: tensor([[ 0.3197, -0.5203, 1.6434, -1.0400, -0.1443, -1.1808, 0.6363, 0.1232],
  27814. [ 0.2566, -0.6155, 1.8100, -0.6931, -0.4654, -0.6712, 0.7653, 0.2250],
  27815. [ 0.5813, -0.4408, 1.7087, 0.2765, -0.6172, -0.2340, 0.6644, 0.1122],
  27816. [ 0.4842, -0.4296, 1.1465, -0.9483, -0.4555, -1.0081, 0.4722, 0.4620],
  27817. [ 0.5570, -0.3612, 1.5991, -0.3948, -0.5087, -1.0276, 0.1089, 0.1086],
  27818. [ 0.5946, -0.3962, 1.6840, -0.0339, -0.1180, 0.3206, 0.4458, 0.3277],
  27819. [ 0.5608, -0.4686, 1.8304, -0.0690, -0.4417, -0.6189, 1.0276, 0.1999],
  27820. [ 0.3944, -0.5490, 1.7384, -0.1617, -0.3495, 0.2295, 0.4583, 0.2216]],
  27821. device='cuda:0', grad_fn=<AddmmBackward>)
  27822. landmarks are: tensor([[[ 0.6126, -0.3871, 1.6229, -1.0773, -0.2132, -1.3698, 0.5829,
  27823. -0.0209],
  27824. [ 0.6078, -0.4057, 1.8134, -0.7309, -0.4499, -0.7386, 0.6298,
  27825. 0.1390],
  27826. [ 0.6249, -0.4352, 1.8018, 0.2545, -0.6173, -0.1997, 0.6401,
  27827. 0.0291],
  27828. [ 0.5631, -0.4129, 1.2129, -0.9233, -0.4152, -1.0311, 0.4566,
  27829. 0.5624],
  27830. [ 0.5485, -0.4209, 1.6691, -0.4152, -0.5249, -1.1081, 0.0725,
  27831. 0.0021],
  27832. [ 0.5811, -0.3878, 1.7499, 0.0236, -0.0688, 0.3161, 0.6240,
  27833. 0.2237],
  27834. [ 0.6273, -0.4249, 1.8654, -0.0611, -0.4672, -0.6693, 1.0910,
  27835. 0.1982],
  27836. [ 0.5960, -0.4102, 1.8018, -0.1612, -0.3344, 0.1159, 0.5490,
  27837. 0.2314]]], device='cuda:0')
  27838. loss_train_step before backward: tensor(0.0093, device='cuda:0', grad_fn=<MseLossBackward>)
  27839. loss_train_step after backward: tensor(0.0093, device='cuda:0', grad_fn=<MseLossBackward>)
  27840. loss_train: 0.4800864914432168
  27841. step: 24
  27842. running loss: 0.020003603810134035
  27843.  
  27844. Train Steps: 24/90 Loss: 0.0200 torch.Size([8, 600, 800])
  27845. torch.Size([8, 8])
  27846. tensor([[0.6329, 0.4196, 0.9238, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748],
  27847. [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760],
  27848. [0.6280, 0.4101, 0.9050, 0.4533, 0.3775, 0.3217, 0.6338, 0.4915],
  27849. [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283],
  27850. [0.6200, 0.4070, 0.8938, 0.4183, 0.3538, 0.4567, 0.6175, 0.5400],
  27851. [0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407],
  27852. [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103],
  27853. [0.6107, 0.4050, 0.8700, 0.4850, 0.4470, 0.4848, 0.5043, 0.5431]],
  27854. device='cuda:0', dtype=torch.float64)
  27855. predictions are: tensor([[ 0.2778, -0.6069, 1.9192, -0.2346, -0.2136, -0.9693, 1.0890, 0.4038],
  27856. [ 0.6852, -0.3778, 1.5858, 0.2068, -0.3921, -0.0105, 1.0698, 0.2575],
  27857. [ 0.4798, -0.4860, 1.8392, -0.2396, -0.4881, -0.8281, 0.6195, 0.1020],
  27858. [ 0.6741, -0.3266, 1.7744, -0.3613, -0.5465, -0.4756, 0.5234, 0.2343],
  27859. [ 0.3450, -0.5909, 1.8070, -0.4339, -0.5138, -0.2004, 0.6149, 0.1366],
  27860. [ 0.2770, -0.5442, 1.5616, -0.4305, -0.5066, -0.8888, 0.0636, 0.2526],
  27861. [ 0.6650, -0.3418, 1.4621, -0.8721, -0.5571, -0.4428, 0.4900, 0.1761],
  27862. [ 0.5881, -0.3967, 1.6592, -0.0824, -0.1299, 0.0190, 0.0928, 0.1560]],
  27863. device='cuda:0', grad_fn=<AddmmBackward>)
  27864. landmarks are: tensor([[[ 0.6372, -0.3479, 1.9808, -0.1689, -0.3171, -0.9310, 1.1057,
  27865. 0.3692],
  27866. [ 0.6421, -0.4008, 1.6055, 0.2160, -0.5076, -0.0534, 1.1020,
  27867. 0.3745],
  27868. [ 0.6148, -0.3918, 1.8942, -0.1920, -0.5423, -0.8002, 0.6414,
  27869. -0.0156],
  27870. [ 0.5880, -0.3887, 1.8423, -0.3306, -0.6231, -0.5230, 0.4046,
  27871. 0.1544],
  27872. [ 0.5778, -0.4061, 1.8423, -0.3537, -0.6520, -0.1766, 0.5663,
  27873. 0.2083],
  27874. [ 0.5470, -0.3952, 1.6377, -0.4253, -0.6289, -0.8079, 0.0249,
  27875. 0.2116],
  27876. [ 0.5886, -0.3784, 1.4554, -0.9079, -0.6577, -0.4845, 0.3440,
  27877. 0.0712],
  27878. [ 0.5351, -0.4153, 1.7326, -0.0457, -0.2214, -0.0466, 0.0434,
  27879. 0.2228]]], device='cuda:0')
  27880. loss_train_step before backward: tensor(0.0106, device='cuda:0', grad_fn=<MseLossBackward>)
  27881. loss_train_step after backward: tensor(0.0106, device='cuda:0', grad_fn=<MseLossBackward>)
  27882. loss_train: 0.49067305121570826
  27883. step: 25
  27884. running loss: 0.01962692204862833
  27885. Train Steps: 25/90 Loss: 0.0196 torch.Size([8, 600, 800])
  27886. torch.Size([8, 8])
  27887. tensor([[0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5787, 0.5600],
  27888. [0.6339, 0.4149, 0.8800, 0.5000, 0.3900, 0.5283, 0.7541, 0.5424],
  27889. [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202],
  27890. [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5637, 0.5633],
  27891. [0.6260, 0.4214, 0.8538, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983],
  27892. [0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058],
  27893. [ nan, nan, 0.8938, 0.2850, 0.4662, 0.3117, 0.7406, 0.5528],
  27894. [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285]],
  27895. device='cuda:0', dtype=torch.float64)
  27896. predictions are: tensor([[ 0.6417, -0.3648, 1.6992, -0.2418, -0.0921, -0.0855, 0.3235, 0.2546],
  27897. [ 0.6611, -0.3901, 1.7823, -0.1001, -0.4637, 0.1033, 1.0566, 0.1444],
  27898. [ 0.7788, -0.2877, 1.6923, -0.0969, -0.2912, -0.0103, 0.4486, 0.0481],
  27899. [ 0.7775, -0.2282, 1.7290, -0.1277, -0.6033, -0.6896, 0.4119, 0.3582],
  27900. [ 0.5787, -0.3499, 1.5670, 0.1464, -0.5430, -0.6805, 0.3149, 0.4437],
  27901. [ 0.7507, -0.3013, 1.6025, -0.4169, -0.5600, -0.2327, 0.2022, 0.0216],
  27902. [-1.5653, -1.8225, 1.9168, -1.0289, -0.1499, -1.0868, 1.0415, 0.2593],
  27903. [ 0.8505, -0.2690, 1.6382, 0.1561, -0.4971, -0.1405, 0.8296, 0.0649]],
  27904. device='cuda:0', grad_fn=<AddmmBackward>)
  27905. landmarks are: tensor([[[ 5.5052e-01, -4.2071e-01, 1.7095e+00, -5.3426e-02, -5.0936e-02,
  27906. 1.0502e-01, 3.8730e-01, 3.0069e-01],
  27907. [ 6.4212e-01, -3.6953e-01, 1.7788e+00, 2.3557e-02, -4.8453e-01,
  27908. 1.5443e-01, 1.1971e+00, 2.1955e-01],
  27909. [ 5.6143e-01, -4.5860e-01, 1.7587e+00, 6.4079e-02, -2.9982e-01,
  27910. 1.7122e-01, 4.9584e-01, 1.1701e-01],
  27911. [ 5.9024e-01, -3.4927e-01, 1.7961e+00, -7.2363e-03, -5.9423e-01,
  27912. -5.6151e-01, 3.1801e-01, 3.1609e-01],
  27913. [ 6.0554e-01, -3.3934e-01, 1.6575e+00, 2.5450e-01, -5.9423e-01,
  27914. -5.4611e-01, 2.9492e-01, 4.7775e-01],
  27915. [ 5.2448e-01, -4.3610e-01, 1.5940e+00, -2.9207e-01, -5.4804e-01,
  27916. -9.1917e-02, 2.4319e-01, 5.0176e-02],
  27917. [-2.2859e+00, -2.2859e+00, 1.8423e+00, -9.6952e-01, -1.3233e-01,
  27918. -8.4634e-01, 1.1349e+00, 2.6764e-01],
  27919. [ 6.5365e-01, -3.9601e-01, 1.6517e+00, 3.1609e-01, -4.9607e-01,
  27920. 4.6189e-04, 7.6203e-01, 1.5543e-01]]], device='cuda:0')
  27921. loss_train_step before backward: tensor(0.0227, device='cuda:0', grad_fn=<MseLossBackward>)
  27922. loss_train_step after backward: tensor(0.0227, device='cuda:0', grad_fn=<MseLossBackward>)
  27923. loss_train: 0.5133614921942353
  27924. step: 26
  27925. running loss: 0.019744672776701357
  27926. Train Steps: 26/90 Loss: 0.0197 torch.Size([8, 600, 800])
  27927. torch.Size([8, 8])
  27928. tensor([[0.6265, 0.4071, 0.8875, 0.3367, 0.3975, 0.3350, 0.6312, 0.5250],
  27929. [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583],
  27930. [0.6311, 0.3998, 0.7975, 0.5767, 0.3838, 0.4850, 0.7327, 0.5343],
  27931. [0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117],
  27932. [0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567],
  27933. [0.6289, 0.4081, 0.8720, 0.3487, 0.3900, 0.3183, 0.6703, 0.5376],
  27934. [0.6151, 0.4085, 0.8750, 0.4367, 0.3887, 0.4367, 0.5066, 0.5846],
  27935. [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967]],
  27936. device='cuda:0', dtype=torch.float64)
  27937. predictions are: tensor([[ 0.4348, -0.5345, 1.8946, -0.7295, -0.4771, -0.7699, 0.7855, 0.1025],
  27938. [ 0.5123, -0.4724, 1.6877, -0.4641, -0.6370, -0.5252, 0.5698, 0.2832],
  27939. [ 0.5497, -0.4999, 1.5042, 0.1221, -0.5073, -0.1399, 0.8437, 0.0930],
  27940. [ 0.3297, -0.5854, 1.5203, 0.3082, -0.4449, -0.1237, 0.4385, 0.4445],
  27941. [ 0.6769, -0.3657, 1.7477, -0.1668, -0.0879, 0.0667, 0.3535, 0.1672],
  27942. [ 0.7711, -0.3086, 1.7944, -0.5775, -0.4388, -0.8960, 0.9241, 0.1388],
  27943. [ 0.2592, -0.5935, 1.7154, -0.2854, -0.5463, -0.3459, 0.0919, 0.3221],
  27944. [ 0.2919, -0.5843, 1.6723, -1.1340, -0.0278, -1.4489, 0.6213, 0.0186]],
  27945. device='cuda:0', grad_fn=<AddmmBackward>)
  27946. landmarks are: tensor([[[ 6.0779e-01, -4.0570e-01, 1.8134e+00, -7.3087e-01, -4.4988e-01,
  27947. -7.3857e-01, 6.2979e-01, 1.3903e-01],
  27948. [ 5.7610e-01, -3.9661e-01, 1.6171e+00, -4.8453e-01, -6.3464e-01,
  27949. -4.6913e-01, 4.7390e-01, 2.9299e-01],
  27950. [ 6.2895e-01, -4.3934e-01, 1.3977e+00, 3.7768e-01, -5.1339e-01,
  27951. -4.5727e-02, 1.0984e+00, 1.8214e-01],
  27952. [ 6.1184e-01, -3.9831e-01, 1.5824e+00, 3.4688e-01, -4.2679e-01,
  27953. -6.8822e-02, 3.4688e-01, 5.3934e-01],
  27954. [ 5.5381e-01, -4.1386e-01, 1.7557e+00, -1.8430e-01, -4.5897e-02,
  27955. 1.2417e-01, 4.2194e-01, 2.8530e-01],
  27956. [ 6.1907e-01, -4.0082e-01, 1.7420e+00, -6.7528e-01, -4.8453e-01,
  27957. -8.1555e-01, 8.1006e-01, 1.9744e-01],
  27958. [ 5.5525e-01, -3.9923e-01, 1.7557e+00, -2.6898e-01, -4.9030e-01,
  27959. -2.6898e-01, 5.4227e-02, 4.1446e-01],
  27960. [ 6.0479e-01, -3.7229e-01, 1.6517e+00, -1.0773e+00, 4.6189e-04,
  27961. -1.5161e+00, 5.8660e-01, 8.0947e-03]]], device='cuda:0')
  27962. loss_train_step before backward: tensor(0.0129, device='cuda:0', grad_fn=<MseLossBackward>)
  27963. loss_train_step after backward: tensor(0.0129, device='cuda:0', grad_fn=<MseLossBackward>)
  27964. loss_train: 0.5262837493792176
  27965. step: 27
  27966. running loss: 0.019491990717748802
  27967. Train Steps: 27/90 Loss: 0.0195 torch.Size([8, 600, 800])
  27968. torch.Size([8, 8])
  27969. tensor([[0.6079, 0.3964, 0.7420, 0.2958, 0.3563, 0.2917, 0.5351, 0.4980],
  27970. [0.6204, 0.4007, 0.7838, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167],
  27971. [0.6230, 0.4113, 0.7213, 0.1983, 0.4325, 0.2367, 0.6262, 0.5400],
  27972. [0.6126, 0.4067, 0.8638, 0.5383, 0.4188, 0.4850, 0.5016, 0.5392],
  27973. [0.6257, 0.4034, 0.8287, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901],
  27974. [0.6125, 0.3999, 0.8750, 0.4883, 0.4750, 0.4700, 0.5533, 0.5617],
  27975. [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350],
  27976. [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869]],
  27977. device='cuda:0', dtype=torch.float64)
  27978. predictions are: tensor([[ 5.9047e-01, -4.3015e-01, 1.2829e+00, -8.2089e-01, -6.3280e-01,
  27979. -8.4250e-01, 3.8176e-01, -1.2932e-03],
  27980. [ 4.8052e-01, -4.9887e-01, 1.3317e+00, -1.0921e+00, -2.2815e-01,
  27981. -1.4085e+00, 6.0268e-01, 1.1726e-01],
  27982. [ 3.5900e-01, -5.5937e-01, 1.1690e+00, -1.2250e+00, -3.4386e-01,
  27983. -1.1243e+00, 8.3362e-01, 2.8056e-01],
  27984. [ 5.8892e-01, -4.3297e-01, 1.7712e+00, 4.5259e-01, -3.8018e-01,
  27985. 1.1718e-01, 3.4146e-01, 2.2360e-01],
  27986. [-5.7707e-02, -8.8170e-01, 1.5893e+00, -9.8783e-01, -4.7708e-01,
  27987. -9.3724e-01, 8.6913e-01, 4.6581e-02],
  27988. [ 6.0177e-01, -4.4187e-01, 1.8321e+00, 1.1511e-01, -1.2282e-01,
  27989. 4.9209e-02, 3.9132e-01, 3.2597e-01],
  27990. [ 3.7568e-01, -5.4407e-01, 1.9095e+00, -1.6609e-01, -6.2486e-01,
  27991. -3.4410e-01, 4.6614e-01, 3.1886e-01],
  27992. [ 8.9065e-01, -2.2668e-01, 1.8704e+00, -6.2750e-01, -2.4321e-01,
  27993. -1.2563e+00, 6.0711e-01, 7.1932e-02]], device='cuda:0',
  27994. grad_fn=<AddmmBackward>)
  27995. landmarks are: tensor([[[ 0.5219, -0.4550, 1.1415, -0.9196, -0.6404, -0.9387, 0.1856,
  27996. 0.0141],
  27997. [ 0.5796, -0.4351, 1.3342, -1.3159, -0.2189, -1.4853, 0.4046,
  27998. 0.1005],
  27999. [ 0.5917, -0.3861, 1.0455, -1.3698, -0.2882, -1.1928, 0.6067,
  28000. 0.2083],
  28001. [ 0.5436, -0.4076, 1.7037, 0.2006, -0.3517, -0.0457, 0.0311,
  28002. 0.2048],
  28003. [ 0.6042, -0.4225, 1.5420, -1.2082, -0.4730, -1.0311, 0.6380,
  28004. -0.0220],
  28005. [ 0.5432, -0.4388, 1.7557, -0.0303, -0.0919, -0.1150, 0.2699,
  28006. 0.3087],
  28007. [ 0.6072, -0.3239, 1.8423, -0.3537, -0.4961, -0.3921, 0.2083,
  28008. 0.1852],
  28009. [ 0.6119, -0.3927, 1.6979, -0.7925, -0.2536, -1.3698, 0.4282,
  28010. -0.0368]]], device='cuda:0')
  28011. loss_train_step before backward: tensor(0.0293, device='cuda:0', grad_fn=<MseLossBackward>)
  28012. loss_train_step after backward: tensor(0.0293, device='cuda:0', grad_fn=<MseLossBackward>)
  28013. loss_train: 0.5556024992838502
  28014. step: 28
  28015. running loss: 0.019842946402994648
  28016.  
  28017. Train Steps: 28/90 Loss: 0.0198 torch.Size([8, 600, 800])
  28018. torch.Size([8, 8])
  28019. tensor([[0.6160, 0.4093, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808],
  28020. [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114],
  28021. [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374],
  28022. [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817],
  28023. [ nan, nan, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552],
  28024. [ nan, nan, 0.6688, 0.2513, 0.4113, 0.2117, 0.5193, 0.5933],
  28025. [0.6083, 0.3957, 0.8638, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980],
  28026. [0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507]],
  28027. device='cuda:0', dtype=torch.float64)
  28028. predictions are: tensor([[ 0.9783, -0.1393, 1.6980, -0.1290, -0.5966, -0.2572, 0.4469, 0.3263],
  28029. [ 1.0109, -0.1533, 1.7835, 0.1568, -0.2468, 0.1617, 0.6071, 0.1030],
  28030. [ 1.0725, -0.1292, 1.8002, -0.4632, -0.6134, -0.8617, 0.7768, 0.0539],
  28031. [ 1.0948, -0.0996, 1.6784, 0.3902, -0.3171, -0.0440, 0.3689, 0.1001],
  28032. [-2.1774, -2.1930, 1.7331, -1.1181, 0.0143, -1.2074, 0.8685, 0.2866],
  28033. [-2.1275, -2.1613, 0.9271, -1.1838, -0.4288, -1.2942, 0.1119, 0.3530],
  28034. [ 1.1153, -0.1008, 1.8017, 0.0431, -0.3620, 0.0071, 0.4366, -0.0337],
  28035. [ 0.8060, -0.2648, 1.0662, -1.2371, -0.4700, -1.2497, 0.5248, 0.1688]],
  28036. device='cuda:0', grad_fn=<AddmmBackward>)
  28037. landmarks are: tensor([[[ 5.5924e-01, -3.9561e-01, 1.5543e+00, -2.4557e-01, -5.8845e-01,
  28038. -1.6890e-01, 1.3922e-01, 3.9681e-01],
  28039. [ 5.4496e-01, -4.7305e-01, 1.7420e+00, 1.3720e-01, -1.9186e-01,
  28040. 2.6139e-01, 4.9757e-01, 7.6435e-02],
  28041. [ 5.7771e-01, -4.4157e-01, 1.7044e+00, -5.8275e-01, -5.9618e-01,
  28042. -8.3610e-01, 4.8621e-01, 1.9626e-01],
  28043. [ 5.1928e-01, -4.6990e-01, 1.5767e+00, 4.0077e-01, -2.4203e-01,
  28044. 7.7444e-02, 1.1776e-01, -6.1038e-02],
  28045. [-2.2859e+00, -2.2859e+00, 1.7557e+00, -1.1466e+00, 8.7067e-02,
  28046. -1.0773e+00, 1.1239e+00, 2.7833e-01],
  28047. [-2.2859e+00, -2.2859e+00, 8.0331e-01, -1.1250e+00, -3.8637e-01,
  28048. -1.3082e+00, 1.1262e-01, 4.5430e-01],
  28049. [ 5.2379e-01, -4.5797e-01, 1.7037e+00, 4.6189e-04, -2.7090e-01,
  28050. 6.2048e-02, 1.8356e-01, 1.4106e-02],
  28051. [ 5.1853e-01, -4.2517e-01, 9.6467e-01, -1.2928e+00, -4.7875e-01,
  28052. -1.2390e+00, 2.6170e-01, 2.5757e-01]]], device='cuda:0')
  28053. loss_train_step before backward: tensor(0.0428, device='cuda:0', grad_fn=<MseLossBackward>)
  28054. loss_train_step after backward: tensor(0.0428, device='cuda:0', grad_fn=<MseLossBackward>)
  28055. loss_train: 0.5983591759577394
  28056. step: 29
  28057. running loss: 0.020633075033025496
  28058. Train Steps: 29/90 Loss: 0.0206 torch.Size([8, 600, 800])
  28059. torch.Size([8, 8])
  28060. tensor([[0.6095, 0.3970, 0.8688, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940],
  28061. [0.6275, 0.4024, 0.8600, 0.2283, 0.5350, 0.1800, 0.7074, 0.5413],
  28062. [0.6034, 0.4011, 0.7350, 0.2533, 0.3438, 0.3367, 0.5516, 0.5084],
  28063. [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767],
  28064. [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092],
  28065. [0.6336, 0.4086, 0.8900, 0.3950, 0.3900, 0.2950, 0.6504, 0.5066],
  28066. [0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742],
  28067. [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679]],
  28068. device='cuda:0', dtype=torch.float64)
  28069. predictions are: tensor([[ 0.5075, -0.5088, 1.7788, -0.1351, -0.1265, -0.0524, 0.2179, 0.0959],
  28070. [-0.2226, -0.9488, 1.6794, -1.3423, 0.1735, -1.5009, 1.0534, 0.2728],
  28071. [ 0.4128, -0.5544, 1.1306, -1.1477, -0.6543, -0.7419, 0.3301, 0.0285],
  28072. [ 0.5527, -0.3928, 1.7188, -0.1665, -0.5780, -0.7781, 0.2299, 0.3728],
  28073. [ 0.6418, -0.4208, 1.8075, 0.0507, -0.4922, 0.0938, 0.3359, 0.1123],
  28074. [ 0.7886, -0.3359, 1.8696, -0.5041, -0.4656, -0.9213, 0.8079, 0.0993],
  28075. [ 0.9940, -0.1880, 1.8234, 0.1831, -0.6006, -0.5426, 0.5605, 0.0328],
  28076. [ 0.3124, -0.6318, 1.6706, 0.0678, -0.3578, 0.2999, 0.9073, 0.3424]],
  28077. device='cuda:0', grad_fn=<AddmmBackward>)
  28078. landmarks are: tensor([[[ 0.5292, -0.4520, 1.7268, -0.0842, -0.0413, -0.0324, 0.1116,
  28079. -0.0039],
  28080. [ 0.6125, -0.4273, 1.6864, -1.2313, 0.1852, -1.4545, 0.9814,
  28081. 0.2142],
  28082. [ 0.5009, -0.4333, 1.1090, -1.1158, -0.6982, -0.7309, 0.2617,
  28083. 0.0622],
  28084. [ 0.5781, -0.3848, 1.7441, -0.0996, -0.5769, -0.8002, 0.2021,
  28085. 0.3778],
  28086. [ 0.5783, -0.4363, 1.7724, 0.0432, -0.5153, 0.0871, 0.4840,
  28087. 0.0663],
  28088. [ 0.6405, -0.3984, 1.8249, -0.4614, -0.4845, -0.9233, 0.7182,
  28089. 0.0539],
  28090. [ 0.6113, -0.4057, 1.7499, 0.3007, -0.5885, -0.5384, 0.4513,
  28091. -0.0957],
  28092. [ 0.6257, -0.4249, 1.5998, 0.1236, -0.3806, 0.3084, 0.9887,
  28093. 0.3371]]], device='cuda:0')
  28094. loss_train_step before backward: tensor(0.0240, device='cuda:0', grad_fn=<MseLossBackward>)
  28095. loss_train_step after backward: tensor(0.0240, device='cuda:0', grad_fn=<MseLossBackward>)
  28096. loss_train: 0.6223869854584336
  28097. step: 30
  28098. running loss: 0.020746232848614453
  28099. Train Steps: 30/90 Loss: 0.0207 torch.Size([8, 600, 800])
  28100. torch.Size([8, 8])
  28101. tensor([[0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350],
  28102. [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960],
  28103. [0.6339, 0.4081, 0.8425, 0.5417, 0.3850, 0.4833, 0.7335, 0.5760],
  28104. [0.6226, 0.4103, 0.8575, 0.3450, 0.4388, 0.2067, 0.5787, 0.5383],
  28105. [0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5988, 0.5700],
  28106. [0.6097, 0.3988, 0.8650, 0.5250, 0.4213, 0.5200, 0.5675, 0.5050],
  28107. [0.6163, 0.4006, 0.8788, 0.4683, 0.3663, 0.4883, 0.5887, 0.5017],
  28108. [0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220]],
  28109. device='cuda:0', dtype=torch.float64)
  28110. predictions are: tensor([[ 0.5228, -0.4469, 1.4481, -1.1311, -0.2062, -1.3879, 0.5930, 0.1343],
  28111. [ 0.7055, -0.3875, 1.8804, -0.0432, -0.2867, 0.0243, 0.3167, 0.0265],
  28112. [ 0.6936, -0.4046, 1.7164, 0.3293, -0.4862, -0.0240, 1.1612, 0.2557],
  28113. [ 0.7628, -0.2640, 1.8340, -0.6504, -0.2326, -1.2022, 0.4564, 0.2166],
  28114. [-0.6807, -1.2229, 1.4377, -0.8805, -0.5891, -0.6684, 0.4366, 0.2792],
  28115. [ 0.5615, -0.4705, 1.7807, 0.2131, -0.3258, 0.1174, 0.3444, 0.0610],
  28116. [ 0.6153, -0.4303, 1.8371, -0.0601, -0.5939, -0.0201, 0.5271, 0.1234],
  28117. [ 0.4692, -0.5014, 1.0312, -1.2795, -0.2998, -1.5233, 0.2746, 0.1054]],
  28118. device='cuda:0', grad_fn=<AddmmBackward>)
  28119. landmarks are: tensor([[[ 0.5768, -0.4031, 1.3838, -1.1527, -0.2188, -1.4216, 0.4379,
  28120. 0.1850],
  28121. [ 0.5205, -0.4577, 1.7326, -0.1304, -0.1862, 0.0447, 0.1301,
  28122. 0.0051],
  28123. [ 0.6421, -0.4008, 1.6055, 0.2160, -0.5076, -0.0534, 1.1020,
  28124. 0.3745],
  28125. [ 0.5898, -0.3908, 1.6748, -0.6924, -0.2594, -1.3313, 0.3873,
  28126. 0.2006],
  28127. [ 0.5960, -0.3888, 1.4840, -1.0095, -0.6115, -0.6231, 0.4797,
  28128. 0.3469],
  28129. [ 0.5303, -0.4440, 1.7095, 0.1390, -0.3402, 0.1159, 0.3353,
  28130. 0.0467],
  28131. [ 0.5609, -0.4354, 1.7730, -0.1227, -0.5942, -0.0303, 0.4335,
  28132. 0.0313],
  28133. [ 0.5566, -0.4160, 0.9935, -1.3313, -0.2824, -1.5161, 0.2144,
  28134. 0.1253]]], device='cuda:0')
  28135. loss_train_step before backward: tensor(0.0424, device='cuda:0', grad_fn=<MseLossBackward>)
  28136. loss_train_step after backward: tensor(0.0424, device='cuda:0', grad_fn=<MseLossBackward>)
  28137. loss_train: 0.6647640662267804
  28138. step: 31
  28139. running loss: 0.021444002136347756
  28140. Train Steps: 31/90 Loss: 0.0214 torch.Size([8, 600, 800])
  28141. torch.Size([8, 8])
  28142. tensor([[0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650],
  28143. [0.6245, 0.4115, 0.8700, 0.4883, 0.4625, 0.5517, 0.6100, 0.5217],
  28144. [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817],
  28145. [0.6264, 0.4055, 0.8425, 0.2767, 0.4425, 0.2767, 0.7050, 0.5586],
  28146. [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478],
  28147. [0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719],
  28148. [0.6193, 0.4034, 0.7757, 0.2347, 0.3733, 0.2919, 0.5930, 0.4926],
  28149. [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467]],
  28150. device='cuda:0', dtype=torch.float64)
  28151. predictions are: tensor([[ 0.4506, -0.5275, 1.7686, -0.2940, -0.5330, -0.4536, 0.3733, 0.2598],
  28152. [ 0.5407, -0.5005, 1.7940, 0.1318, -0.0543, 0.2764, 0.6289, 0.1438],
  28153. [ 0.3794, -0.5269, 1.5145, -0.7211, -0.1765, -1.1368, 0.3719, 0.3084],
  28154. [ 0.4169, -0.5643, 1.6879, -0.9827, -0.1545, -1.1282, 0.9710, 0.1931],
  28155. [ 0.3583, -0.5941, 1.9093, -0.4784, -0.4752, -0.8796, 0.5574, 0.1085],
  28156. [ 0.6627, -0.3939, 1.6628, -0.2187, -0.5922, -0.3117, 0.2554, 0.1885],
  28157. [ 0.7224, -0.3879, 1.3228, -1.0733, -0.4802, -1.0500, 0.5118, -0.1316],
  28158. [ 0.6318, -0.4254, 1.6359, -0.5419, -0.6135, -0.3646, 0.5188, 0.1461]],
  28159. device='cuda:0', grad_fn=<AddmmBackward>)
  28160. landmarks are: tensor([[[ 0.5757, -0.3917, 1.7095, -0.4768, -0.6346, -0.4229, 0.3931,
  28161. 0.3238],
  28162. [ 0.5987, -0.3852, 1.7326, -0.0303, -0.1497, 0.2622, 0.5316,
  28163. 0.1236],
  28164. [ 0.5964, -0.3380, 1.4785, -0.8386, -0.2420, -1.0619, 0.3238,
  28165. 0.4008],
  28166. [ 0.6075, -0.4129, 1.6055, -1.0080, -0.2420, -1.0080, 0.9704,
  28167. 0.2944],
  28168. [ 0.5792, -0.4052, 1.8214, -0.6587, -0.5384, -0.8924, 0.4381,
  28169. 0.2442],
  28170. [ 0.5367, -0.4294, 1.5709, -0.4999, -0.6693, -0.3075, 0.2455,
  28171. 0.3559],
  28172. [ 0.5746, -0.4225, 1.2968, -1.2019, -0.5615, -0.9374, 0.4533,
  28173. -0.0108],
  28174. [ 0.5484, -0.4591, 1.5940, -0.7771, -0.6693, -0.3460, 0.4219,
  28175. 0.2391]]], device='cuda:0')
  28176. loss_train_step before backward: tensor(0.0130, device='cuda:0', grad_fn=<MseLossBackward>)
  28177. loss_train_step after backward: tensor(0.0130, device='cuda:0', grad_fn=<MseLossBackward>)
  28178. loss_train: 0.6778084067627788
  28179. step: 32
  28180. running loss: 0.021181512711336836
  28181.  
  28182. Train Steps: 32/90 Loss: 0.0212 torch.Size([8, 600, 800])
  28183. torch.Size([8, 8])
  28184. tensor([[0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6188, 0.5283],
  28185. [0.6222, 0.3957, 0.8838, 0.5017, 0.3937, 0.4600, 0.5900, 0.5017],
  28186. [0.6273, 0.4143, 0.8750, 0.5700, 0.3987, 0.4717, 0.6013, 0.5467],
  28187. [0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124],
  28188. [0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507],
  28189. [0.6175, 0.3957, 0.8700, 0.4817, 0.4662, 0.5133, 0.5800, 0.5517],
  28190. [ nan, nan, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533],
  28191. [0.6278, 0.4253, 0.8875, 0.5017, 0.4113, 0.2750, 0.5413, 0.6196]],
  28192. device='cuda:0', dtype=torch.float64)
  28193. predictions are: tensor([[ 0.7188, -0.3367, 1.8920, -0.6707, -0.4313, -1.0907, 0.6629, 0.0770],
  28194. [ 0.6263, -0.4435, 1.7797, 0.0100, -0.4408, -0.1267, 0.4634, -0.0449],
  28195. [ 0.7959, -0.2640, 1.6776, 0.2334, -0.4097, -0.1032, 0.5011, 0.1118],
  28196. [ 0.5462, -0.4585, 1.5470, -1.0173, -0.4050, -1.0583, 0.4912, -0.0234],
  28197. [ 0.5802, -0.4418, 0.9874, -1.2851, -0.3965, -1.1792, 0.4044, 0.1651],
  28198. [ 0.6268, -0.4222, 1.7376, -0.0928, -0.1368, 0.0947, 0.3983, 0.1281],
  28199. [-2.0925, -2.1734, 1.2965, -0.9992, -0.5074, -0.9629, 0.3409, 0.2234],
  28200. [ 0.7433, -0.2898, 1.7182, -0.0200, -0.2876, -0.9544, 0.3056, 0.4590]],
  28201. device='cuda:0', grad_fn=<AddmmBackward>)
  28202. landmarks are: tensor([[[ 0.6062, -0.3778, 1.8423, -0.6462, -0.4383, -1.0465, 0.5721,
  28203. 0.1544],
  28204. [ 0.5879, -0.4583, 1.7961, 0.0313, -0.4672, -0.1612, 0.4393,
  28205. 0.0313],
  28206. [ 0.6115, -0.3724, 1.7557, 0.3469, -0.4441, -0.1073, 0.4912,
  28207. 0.2391],
  28208. [ 0.5790, -0.4079, 1.5929, -1.0630, -0.4729, -1.0725, 0.4137,
  28209. 0.0807],
  28210. [ 0.5185, -0.4252, 0.9647, -1.2928, -0.4788, -1.2390, 0.2617,
  28211. 0.2576],
  28212. [ 0.5662, -0.4581, 1.7326, -0.0611, -0.1323, 0.0851, 0.3931,
  28213. 0.2622],
  28214. [-2.2859, -2.2859, 1.5074, -1.0388, -0.5423, -0.9849, 0.2199,
  28215. 0.2699],
  28216. [ 0.6139, -0.3216, 1.8134, 0.0313, -0.3864, -1.0157, 0.2144,
  28217. 0.5762]]], device='cuda:0')
  28218. loss_train_step before backward: tensor(0.0063, device='cuda:0', grad_fn=<MseLossBackward>)
  28219. loss_train_step after backward: tensor(0.0063, device='cuda:0', grad_fn=<MseLossBackward>)
  28220. loss_train: 0.6841176045127213
  28221. step: 33
  28222. running loss: 0.020730836500385492
  28223. Train Steps: 33/90 Loss: 0.0207 torch.Size([8, 600, 800])
  28224. torch.Size([8, 8])
  28225. tensor([[0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350],
  28226. [ nan, nan, 0.8525, 0.2217, 0.5413, 0.2367, 0.7367, 0.5482],
  28227. [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967],
  28228. [0.6336, 0.4154, 0.8900, 0.2767, 0.4988, 0.2867, 0.7422, 0.5540],
  28229. [0.6311, 0.3998, 0.7975, 0.5767, 0.3838, 0.4850, 0.7327, 0.5343],
  28230. [0.6246, 0.4028, 0.8738, 0.4867, 0.4088, 0.5667, 0.6362, 0.5200],
  28231. [0.6268, 0.4052, 0.8175, 0.2250, 0.4688, 0.1917, 0.6375, 0.5267],
  28232. [0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019]],
  28233. device='cuda:0', dtype=torch.float64)
  28234. predictions are: tensor([[ 0.6255, -0.3592, 1.6984, -0.2594, -0.6196, -0.3773, 0.0322, 0.2342],
  28235. [-1.9203, -2.0523, 1.4849, -1.2326, 0.1044, -1.1637, 0.8986, 0.2863],
  28236. [ 0.6367, -0.3641, 1.5570, -1.0202, -0.1333, -1.3515, 0.3457, 0.0102],
  28237. [ 0.5406, -0.4691, 1.7121, -0.8162, -0.1013, -0.9956, 0.8345, 0.1929],
  28238. [ 0.6415, -0.4127, 1.3976, 0.2055, -0.5793, 0.0214, 0.5662, 0.1494],
  28239. [ 0.6824, -0.3714, 1.7048, 0.0315, -0.4587, 0.4051, 0.4384, 0.1293],
  28240. [ 0.6387, -0.3973, 1.3418, -1.1451, -0.2160, -1.4084, 0.4027, 0.0928],
  28241. [ 0.6806, -0.3924, 1.7546, -0.0103, -0.6495, -0.3794, 0.3367, 0.0941]],
  28242. device='cuda:0', grad_fn=<AddmmBackward>)
  28243. landmarks are: tensor([[[ 6.0722e-01, -3.2394e-01, 1.8423e+00, -3.5366e-01, -4.9607e-01,
  28244. -3.9215e-01, 2.0831e-01, 1.8522e-01],
  28245. [-2.2859e+00, -2.2859e+00, 1.6517e+00, -1.2620e+00, 2.1409e-01,
  28246. -1.1928e+00, 1.1166e+00, 2.4627e-01],
  28247. [ 6.0479e-01, -3.7229e-01, 1.6517e+00, -1.0773e+00, 4.6189e-04,
  28248. -1.5161e+00, 5.8660e-01, 8.0947e-03],
  28249. [ 6.4048e-01, -3.6712e-01, 1.8249e+00, -1.0080e+00, 1.7783e-02,
  28250. -9.6182e-01, 1.1422e+00, 2.7299e-01],
  28251. [ 6.2895e-01, -4.3934e-01, 1.3977e+00, 3.7768e-01, -5.1339e-01,
  28252. -4.5727e-02, 1.0984e+00, 1.8214e-01],
  28253. [ 5.9902e-01, -4.2556e-01, 1.7499e+00, -3.8029e-02, -3.9792e-01,
  28254. 3.3149e-01, 6.5289e-01, 1.1594e-01],
  28255. [ 6.0918e-01, -4.1432e-01, 1.4901e+00, -1.2467e+00, -1.2079e-01,
  28256. -1.4006e+00, 6.5866e-01, 1.4673e-01],
  28257. [ 6.3718e-01, -4.1286e-01, 1.8942e+00, -7.6520e-02, -6.1732e-01,
  28258. -4.7683e-01, 6.9989e-01, 3.2524e-02]]], device='cuda:0')
  28259. loss_train_step before backward: tensor(0.0209, device='cuda:0', grad_fn=<MseLossBackward>)
  28260. loss_train_step after backward: tensor(0.0209, device='cuda:0', grad_fn=<MseLossBackward>)
  28261. loss_train: 0.7050670147873461
  28262. step: 34
  28263. running loss: 0.020737265140804297
  28264. Train Steps: 34/90 Loss: 0.0207 torch.Size([8, 600, 800])
  28265. torch.Size([8, 8])
  28266. tensor([[0.6069, 0.3975, 0.8625, 0.5083, 0.4388, 0.5483, 0.5650, 0.4967],
  28267. [0.6198, 0.4130, 0.8762, 0.4117, 0.3650, 0.4900, 0.5707, 0.5103],
  28268. [0.6213, 0.4001, 0.7712, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183],
  28269. [0.6093, 0.3990, 0.8400, 0.4333, 0.3688, 0.4633, 0.5560, 0.5656],
  28270. [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617],
  28271. [0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904],
  28272. [0.6300, 0.4133, 0.8538, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413],
  28273. [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167]],
  28274. device='cuda:0', dtype=torch.float64)
  28275. predictions are: tensor([[ 0.4029, -0.5921, 1.6903, 0.1829, -0.3016, 0.2530, 0.2826, 0.0848],
  28276. [ 0.5061, -0.4869, 1.8188, -0.2548, -0.6397, -0.0794, 0.3478, 0.0932],
  28277. [ 0.5058, -0.5134, 1.2198, -1.2538, -0.3336, -1.4177, 0.3494, 0.0629],
  28278. [ 0.5262, -0.4614, 1.6594, -0.1469, -0.6424, -0.1302, 0.2515, 0.2958],
  28279. [ 0.5180, -0.4978, 1.7588, 0.0278, -0.3378, 0.0622, 0.4041, 0.2559],
  28280. [ 0.5641, -0.4334, 1.6918, -0.9860, -0.2200, -1.3162, 0.6009, -0.0315],
  28281. [ 0.4836, -0.5410, 1.7605, -1.1247, 0.1414, -1.2232, 1.1395, 0.1591],
  28282. [ 0.5724, -0.4363, 1.2401, -0.8151, -0.4982, -0.9966, 0.3776, 0.4530]],
  28283. device='cuda:0', grad_fn=<AddmmBackward>)
  28284. landmarks are: tensor([[[ 0.5174, -0.4497, 1.6979, 0.0620, -0.2594, 0.2468, 0.3238,
  28285. 0.0082],
  28286. [ 0.5771, -0.3784, 1.7614, -0.3844, -0.6000, -0.0226, 0.3501,
  28287. 0.0712],
  28288. [ 0.5838, -0.4379, 1.2764, -1.3082, -0.2824, -1.4545, 0.4162,
  28289. 0.1082],
  28290. [ 0.5283, -0.4429, 1.5940, -0.2844, -0.5827, -0.1458, 0.2823,
  28291. 0.3267],
  28292. [ 0.5637, -0.4143, 1.7519, -0.0787, -0.3055, -0.0149, 0.3758,
  28293. 0.3084],
  28294. [ 0.6126, -0.3871, 1.6229, -1.0773, -0.2132, -1.3698, 0.5829,
  28295. -0.0209],
  28296. [ 0.6240, -0.3768, 1.6575, -1.2851, 0.2949, -1.2467, 1.1276,
  28297. 0.2142],
  28298. [ 0.5631, -0.4129, 1.2129, -0.9233, -0.4152, -1.0311, 0.4566,
  28299. 0.5624]]], device='cuda:0')
  28300. loss_train_step before backward: tensor(0.0056, device='cuda:0', grad_fn=<MseLossBackward>)
  28301. loss_train_step after backward: tensor(0.0056, device='cuda:0', grad_fn=<MseLossBackward>)
  28302. loss_train: 0.7106758425943553
  28303. step: 35
  28304. running loss: 0.020305024074124437
  28305. Train Steps: 35/90 Loss: 0.0203 torch.Size([8, 600, 800])
  28306. torch.Size([8, 8])
  28307. tensor([[0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869],
  28308. [0.6219, 0.3934, 0.8688, 0.5267, 0.4313, 0.4967, 0.5988, 0.4983],
  28309. [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167],
  28310. [ nan, nan, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733],
  28311. [0.6286, 0.4078, 0.8063, 0.2267, 0.4788, 0.1533, 0.5953, 0.4913],
  28312. [0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6262, 0.5167],
  28313. [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250],
  28314. [ nan, nan, 0.7335, 0.2569, 0.3788, 0.2667, 0.5066, 0.5578]],
  28315. device='cuda:0', dtype=torch.float64)
  28316. predictions are: tensor([[ 0.7175, -0.3182, 1.5601, -1.1531, -0.1840, -1.2262, 0.5677, -0.0161],
  28317. [ 0.7583, -0.3484, 1.7435, 0.2148, -0.4061, 0.0851, 0.4845, -0.0334],
  28318. [ 0.7261, -0.2570, 1.2147, -0.6802, -0.0970, -1.2256, 0.2552, 0.4634],
  28319. [-1.5970, -1.8209, 0.9663, -1.2563, -0.2124, -1.3566, 0.3545, 0.4310],
  28320. [ 0.6751, -0.3353, 1.3610, -1.0378, -0.1904, -1.4201, 0.3632, 0.0626],
  28321. [ 0.9350, -0.1898, 1.9036, -0.2485, -0.6079, -0.5402, 0.5454, 0.1273],
  28322. [ 0.9440, -0.1748, 1.8448, 0.0834, -0.4474, 0.3213, 0.6020, 0.1024],
  28323. [-1.9158, -2.0474, 1.1735, -1.1139, -0.5103, -1.0042, 0.2081, 0.2710]],
  28324. device='cuda:0', grad_fn=<AddmmBackward>)
  28325. landmarks are: tensor([[[ 0.6127, -0.3944, 1.5189, -1.2467, -0.1323, -1.4622, 0.5646,
  28326. -0.0369],
  28327. [ 0.5864, -0.4690, 1.7268, 0.1467, -0.2940, 0.0082, 0.4797,
  28328. 0.0159],
  28329. [ 0.6161, -0.3075, 1.1678, -0.6500, 0.0813, -1.4006, 0.2545,
  28330. 0.5624],
  28331. [-2.2859, -2.2859, 0.7106, -1.4468, -0.2882, -1.4237, 0.2430,
  28332. 0.3623],
  28333. [ 0.6177, -0.4022, 1.4381, -1.2390, -0.0746, -1.5777, 0.4639,
  28334. -0.0168],
  28335. [ 0.6094, -0.3947, 1.8885, -0.2998, -0.5769, -0.6770, 0.6067,
  28336. 0.1005],
  28337. [ 0.5991, -0.3803, 1.8018, -0.0534, -0.3460, 0.1852, 0.5374,
  28338. 0.1390],
  28339. [-2.2859, -2.2859, 1.1020, -1.0994, -0.5365, -1.0542, 0.0542,
  28340. 0.2905]]], device='cuda:0')
  28341. loss_train_step before backward: tensor(0.0280, device='cuda:0', grad_fn=<MseLossBackward>)
  28342. loss_train_step after backward: tensor(0.0280, device='cuda:0', grad_fn=<MseLossBackward>)
  28343. loss_train: 0.7386775049380958
  28344. step: 36
  28345. running loss: 0.02051881958161377
  28346.  
  28347. Train Steps: 36/90 Loss: 0.0205 torch.Size([8, 600, 800])
  28348. torch.Size([8, 8])
  28349. tensor([[0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400],
  28350. [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617],
  28351. [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5787, 0.5550],
  28352. [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366],
  28353. [0.6189, 0.4029, 0.8375, 0.5767, 0.4745, 0.4829, 0.5551, 0.5598],
  28354. [0.6346, 0.4144, 0.9088, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899],
  28355. [0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5837, 0.5583],
  28356. [ nan, nan, 0.8488, 0.2300, 0.5563, 0.2100, 0.7390, 0.5679]],
  28357. device='cuda:0', dtype=torch.float64)
  28358. predictions are: tensor([[ 0.7026, -0.3475, 1.7143, -0.5126, -0.6330, -0.2998, 0.4079, 0.1759],
  28359. [ 0.4452, -0.4884, 1.7904, -0.2999, -0.4580, -0.4563, 0.1168, 0.2451],
  28360. [ 0.8208, -0.2664, 1.4383, -1.0448, -0.6861, -0.8196, 0.1715, 0.0891],
  28361. [ 0.6975, -0.3649, 1.3709, 0.1070, -0.4919, -0.1339, 0.7419, 0.1754],
  28362. [ 0.6489, -0.3556, 1.5818, 0.2273, -0.1254, -0.3050, 0.1784, 0.2450],
  28363. [ 0.7435, -0.3313, 1.8172, -0.2670, -0.4164, -0.4652, 0.8118, 0.1959],
  28364. [ 0.4413, -0.4983, 1.7393, -0.3508, -0.1120, -0.1085, 0.3081, 0.1528],
  28365. [-1.9452, -2.0874, 1.6159, -1.3266, 0.2288, -1.4092, 1.0428, 0.2691]],
  28366. device='cuda:0', grad_fn=<AddmmBackward>)
  28367. landmarks are: tensor([[[ 0.5778, -0.4393, 1.8018, -0.4614, -0.6693, -0.1381, 0.5490,
  28368. 0.2083],
  28369. [ 0.5488, -0.4221, 1.8018, -0.2459, -0.4845, -0.3075, 0.2309,
  28370. 0.3087],
  28371. [ 0.5647, -0.4129, 1.4901, -1.0619, -0.6462, -0.5846, 0.3873,
  28372. 0.2776],
  28373. [ 0.6289, -0.4345, 1.3794, 0.3679, -0.4845, 0.0390, 0.9265,
  28374. 0.1928],
  28375. [ 0.5726, -0.4249, 1.5824, 0.3777, -0.0942, -0.0556, 0.2782,
  28376. 0.2997],
  28377. [ 0.6454, -0.3719, 1.9115, -0.1304, -0.5076, -0.2844, 1.0033,
  28378. 0.4386],
  28379. [ 0.5223, -0.4336, 1.7557, -0.2074, -0.0573, 0.3084, 0.4104,
  28380. 0.2930],
  28381. [-2.2859, -2.2859, 1.6344, -1.2236, 0.2834, -1.3159, 1.1276,
  28382. 0.3371]]], device='cuda:0')
  28383. loss_train_step before backward: tensor(0.0191, device='cuda:0', grad_fn=<MseLossBackward>)
  28384. loss_train_step after backward: tensor(0.0191, device='cuda:0', grad_fn=<MseLossBackward>)
  28385. loss_train: 0.7578194118104875
  28386. step: 37
  28387. running loss: 0.02048160572460777
  28388. Train Steps: 37/90 Loss: 0.0205 torch.Size([8, 600, 800])
  28389. torch.Size([8, 8])
  28390. tensor([[ nan, nan, 0.7725, 0.2611, 0.3675, 0.2733, 0.5413, 0.5167],
  28391. [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116],
  28392. [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892],
  28393. [0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5737, 0.5350],
  28394. [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389],
  28395. [0.6145, 0.4007, 0.8775, 0.4533, 0.4562, 0.5533, 0.6088, 0.5533],
  28396. [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896],
  28397. [0.6148, 0.3996, 0.8488, 0.3867, 0.3488, 0.4067, 0.5863, 0.5000]],
  28398. device='cuda:0', dtype=torch.float64)
  28399. predictions are: tensor([[-2.1001, -2.2408, 1.2620, -1.2134, -0.3332, -1.1288, 0.4092, 0.1548],
  28400. [ 0.6518, -0.3524, 1.6998, -0.3598, -0.2870, 0.3224, 0.6301, 0.1420],
  28401. [ 0.7167, -0.2884, 1.6898, -0.2260, -0.4900, -0.6974, 0.2223, 0.3375],
  28402. [ 0.6266, -0.3632, 1.8030, -0.1496, -0.3881, -1.0460, 0.4562, 0.1622],
  28403. [ 0.4956, -0.4900, 1.2467, -1.3024, -0.3474, -1.1994, 0.6318, 0.1830],
  28404. [ 0.4585, -0.5079, 1.7557, -0.2600, -0.0559, 0.1161, 0.5975, 0.2690],
  28405. [ 0.6890, -0.3344, 1.5034, -0.6130, -0.4778, -1.0306, 0.1694, 0.3261],
  28406. [ 0.6072, -0.4055, 1.6764, -0.5149, -0.5742, -0.4923, 0.5001, 0.1026]],
  28407. device='cuda:0', grad_fn=<AddmmBackward>)
  28408. landmarks are: tensor([[[-2.2859, -2.2859, 1.2820, -1.0801, -0.5885, -1.0234, 0.2141,
  28409. 0.1005],
  28410. [ 0.5376, -0.3903, 1.7095, -0.3229, -0.4730, 0.4701, 0.3871,
  28411. 0.0772],
  28412. [ 0.5454, -0.4053, 1.6633, -0.1766, -0.6058, -0.5923, 0.1577,
  28413. 0.4357],
  28414. [ 0.5901, -0.4000, 1.8423, -0.0688, -0.5307, -0.9233, 0.3642,
  28415. 0.1852],
  28416. [ 0.5697, -0.4466, 1.1973, -1.1871, -0.4571, -0.9965, 0.5219,
  28417. 0.2032],
  28418. [ 0.5526, -0.4351, 1.7672, -0.1920, -0.1785, 0.2699, 0.5259,
  28419. 0.2699],
  28420. [ 0.5485, -0.3997, 1.4445, -0.4895, -0.6000, -0.8309, 0.1878,
  28421. 0.4374],
  28422. [ 0.5538, -0.4400, 1.6344, -0.4999, -0.6751, -0.4075, 0.4219,
  28423. 0.0236]]], device='cuda:0')
  28424. loss_train_step before backward: tensor(0.0116, device='cuda:0', grad_fn=<MseLossBackward>)
  28425. loss_train_step after backward: tensor(0.0116, device='cuda:0', grad_fn=<MseLossBackward>)
  28426. loss_train: 0.7694539255462587
  28427. step: 38
  28428. running loss: 0.020248787514375227
  28429. Train Steps: 38/90 Loss: 0.0202 torch.Size([8, 600, 800])
  28430. torch.Size([8, 8])
  28431. tensor([[0.6125, 0.3999, 0.8750, 0.4883, 0.4750, 0.4700, 0.5533, 0.5617],
  28432. [0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367],
  28433. [0.6205, 0.4016, 0.8350, 0.2717, 0.3987, 0.2550, 0.5787, 0.5133],
  28434. [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6038, 0.4833],
  28435. [0.6104, 0.4029, 0.8738, 0.4900, 0.4088, 0.4533, 0.5070, 0.5510],
  28436. [0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5737, 0.5350],
  28437. [ nan, nan, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621],
  28438. [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209]],
  28439. device='cuda:0', dtype=torch.float64)
  28440. predictions are: tensor([[ 0.3345, -0.5815, 1.6559, -0.2002, -0.0839, -0.0443, 0.3847, 0.3240],
  28441. [ 0.6685, -0.3828, 1.5534, -1.0486, -0.4268, -0.9877, 0.7333, 0.1445],
  28442. [ 0.6851, -0.3588, 1.4897, -1.1548, -0.4079, -1.0997, 0.4530, 0.0514],
  28443. [ 0.4126, -0.5595, 1.6417, -0.2154, -0.5463, -0.0265, 0.4602, 0.1108],
  28444. [ 0.5710, -0.4018, 1.6016, -0.0976, -0.4236, -0.0688, 0.2664, 0.2464],
  28445. [ 0.6039, -0.3627, 1.6929, -0.2127, -0.5257, -0.8801, 0.3550, 0.1962],
  28446. [-1.8773, -2.0152, 1.4880, -1.3270, 0.1654, -1.2752, 0.8369, 0.2394],
  28447. [ 0.5513, -0.3570, 1.5899, -0.3356, -0.2179, -1.0885, 0.2876, 0.5012]],
  28448. device='cuda:0', grad_fn=<AddmmBackward>)
  28449. landmarks are: tensor([[[ 0.5432, -0.4388, 1.7557, -0.0303, -0.0919, -0.1150, 0.2699,
  28450. 0.3087],
  28451. [ 0.6174, -0.4201, 1.7309, -0.8784, -0.4735, -0.9524, 0.6242,
  28452. 0.1931],
  28453. [ 0.5800, -0.4312, 1.5709, -1.0311, -0.4441, -1.1081, 0.3873,
  28454. 0.0851],
  28455. [ 0.5726, -0.4795, 1.7788, -0.0149, -0.5711, -0.0688, 0.5028,
  28456. -0.0534],
  28457. [ 0.5334, -0.4252, 1.7499, -0.0226, -0.3979, -0.1920, 0.0558,
  28458. 0.2589],
  28459. [ 0.5901, -0.4000, 1.8423, -0.0688, -0.5307, -0.9233, 0.3642,
  28460. 0.1852],
  28461. [-2.2859, -2.2859, 1.7210, -0.9772, 0.1852, -1.3698, 0.7986,
  28462. 0.3104],
  28463. [ 0.6174, -0.3118, 1.6402, -0.2074, -0.1958, -1.0927, 0.2267,
  28464. 0.5822]]], device='cuda:0')
  28465. loss_train_step before backward: tensor(0.0153, device='cuda:0', grad_fn=<MseLossBackward>)
  28466. loss_train_step after backward: tensor(0.0153, device='cuda:0', grad_fn=<MseLossBackward>)
  28467. loss_train: 0.7847062242217362
  28468. step: 39
  28469. running loss: 0.020120672415941954
  28470. Train Steps: 39/90 Loss: 0.0201 torch.Size([8, 600, 800])
  28471. torch.Size([8, 8])
  28472. tensor([[0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742],
  28473. [ nan, nan, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563],
  28474. [ nan, nan, 0.6900, 0.1917, 0.3937, 0.2367, 0.5240, 0.5246],
  28475. [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617],
  28476. [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413],
  28477. [0.6293, 0.4024, 0.8750, 0.5000, 0.4012, 0.5733, 0.7121, 0.5633],
  28478. [0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528],
  28479. [0.6104, 0.4029, 0.8738, 0.4900, 0.4088, 0.4533, 0.5070, 0.5510]],
  28480. device='cuda:0', dtype=torch.float64)
  28481. predictions are: tensor([[ 0.7944, -0.2597, 1.6941, -0.7817, -0.3957, -1.1214, 0.3668, 0.0797],
  28482. [-1.2229, -1.5579, 1.2096, -0.8862, -0.5949, -0.8335, 0.0194, 0.3173],
  28483. [-1.9269, -2.0668, 1.0180, -1.3759, -0.3612, -1.2341, 0.1503, 0.2716],
  28484. [ 0.6871, -0.2951, 1.6941, 0.2017, -0.4402, -0.3049, 0.2864, 0.4204],
  28485. [ 0.6057, -0.3936, 1.8279, -0.9332, 0.0218, -1.2719, 0.9551, 0.2071],
  28486. [ 0.9779, -0.1799, 1.6772, 0.0201, -0.3361, 0.4483, 0.8435, 0.2974],
  28487. [ 0.6400, -0.3831, 1.8871, -0.6828, -0.3191, -0.5734, 0.8844, 0.1991],
  28488. [ 0.7648, -0.2619, 1.6307, 0.0055, -0.3844, -0.1027, 0.1083, 0.2771]],
  28489. device='cuda:0', grad_fn=<AddmmBackward>)
  28490. landmarks are: tensor([[[ 0.6134, -0.4218, 1.7268, -0.6154, -0.4730, -1.0850, 0.5463,
  28491. -0.0957],
  28492. [-2.2859, -2.2859, 1.4006, -0.8105, -0.6115, -0.8232, 0.0419,
  28493. 0.2837],
  28494. [-2.2859, -2.2859, 0.9012, -1.4006, -0.4672, -1.1928, 0.1342,
  28495. 0.1373],
  28496. [ 0.5897, -0.3527, 1.8018, 0.2545, -0.5307, -0.3229, 0.3122,
  28497. 0.3084],
  28498. [ 0.6108, -0.4201, 1.8711, -0.7848, -0.0053, -1.2236, 1.0362,
  28499. 0.2142],
  28500. [ 0.6207, -0.4273, 1.7557, 0.0236, -0.4326, 0.3623, 1.0033,
  28501. 0.3157],
  28502. [ 0.6388, -0.3623, 1.9173, -0.7386, -0.3517, -0.5846, 1.1495,
  28503. 0.2676],
  28504. [ 0.5334, -0.4252, 1.7499, -0.0226, -0.3979, -0.1920, 0.0558,
  28505. 0.2589]]], device='cuda:0')
  28506. loss_train_step before backward: tensor(0.0405, device='cuda:0', grad_fn=<MseLossBackward>)
  28507. loss_train_step after backward: tensor(0.0405, device='cuda:0', grad_fn=<MseLossBackward>)
  28508. loss_train: 0.8252143119461834
  28509. step: 40
  28510. running loss: 0.020630357798654585
  28511.  
  28512. Train Steps: 40/90 Loss: 0.0206 torch.Size([8, 600, 800])
  28513. torch.Size([8, 8])
  28514. tensor([[0.6109, 0.4041, 0.6975, 0.3167, 0.3513, 0.3383, 0.5153, 0.5319],
  28515. [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290],
  28516. [0.6185, 0.4098, 0.8838, 0.4900, 0.4537, 0.5800, 0.6288, 0.5400],
  28517. [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750],
  28518. [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268],
  28519. [0.6203, 0.4073, 0.8189, 0.2398, 0.4400, 0.2054, 0.5929, 0.5501],
  28520. [0.6115, 0.4081, 0.6725, 0.2433, 0.4088, 0.1933, 0.5167, 0.5544],
  28521. [0.6267, 0.4080, 0.8438, 0.2633, 0.4763, 0.1800, 0.6259, 0.5240]],
  28522. device='cuda:0', dtype=torch.float64)
  28523. predictions are: tensor([[ 0.2722, -0.6165, 1.1779, -0.8024, -0.6699, -0.6848, 0.0763, 0.2715],
  28524. [ 0.2924, -0.6367, 1.3922, -1.1740, -0.2910, -1.2150, 0.6039, 0.1700],
  28525. [ 0.3281, -0.5799, 1.9128, 0.0207, -0.1557, 0.4681, 0.6649, 0.3074],
  28526. [ 0.3481, -0.5868, 1.1737, -1.0323, -0.5273, -0.8910, 0.5458, 0.4523],
  28527. [ 0.2317, -0.6822, 1.8328, 0.0113, -0.2837, 0.2063, 0.6388, 0.2003],
  28528. [ 0.4696, -0.4915, 1.5757, -1.0361, -0.2425, -1.3144, 0.4999, 0.2187],
  28529. [ 0.2441, -0.6215, 1.0434, -1.0948, -0.4172, -1.3396, -0.0088, 0.3534],
  28530. [ 0.6824, -0.3667, 1.7853, -0.9849, -0.1111, -1.4141, 0.6500, 0.1578]],
  28531. device='cuda:0', grad_fn=<AddmmBackward>)
  28532. landmarks are: tensor([[[ 0.5359, -0.4193, 0.9358, -0.8232, -0.6635, -0.7232, 0.0943,
  28533. 0.1710],
  28534. [ 0.6125, -0.4273, 1.2807, -1.3253, -0.2574, -1.2542, 0.6864,
  28535. 0.1575],
  28536. [ 0.5709, -0.3933, 1.7961, -0.0226, -0.1901, 0.3931, 0.6182,
  28537. 0.2083],
  28538. [ 0.5805, -0.3818, 1.0282, -1.1774, -0.4903, -0.9310, 0.5894,
  28539. 0.3700],
  28540. [ 0.5828, -0.4682, 1.7031, -0.0497, -0.2458, 0.0818, 0.6381,
  28541. 0.1474],
  28542. [ 0.5792, -0.4048, 1.4965, -1.1781, -0.2534, -1.3371, 0.4528,
  28543. 0.2549],
  28544. [ 0.5385, -0.4010, 0.8203, -1.1620, -0.3979, -1.3929, 0.1005,
  28545. 0.2747],
  28546. [ 0.6088, -0.4015, 1.6113, -1.0696, -0.0861, -1.4545, 0.6051,
  28547. 0.1343]]], device='cuda:0')
  28548. loss_train_step before backward: tensor(0.0182, device='cuda:0', grad_fn=<MseLossBackward>)
  28549. loss_train_step after backward: tensor(0.0182, device='cuda:0', grad_fn=<MseLossBackward>)
  28550. loss_train: 0.843462482560426
  28551. step: 41
  28552. running loss: 0.02057225567220551
  28553. Train Steps: 41/90 Loss: 0.0206 torch.Size([8, 600, 800])
  28554. torch.Size([8, 8])
  28555. tensor([[0.6109, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567],
  28556. [0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117],
  28557. [0.6208, 0.4082, 0.8538, 0.3067, 0.3588, 0.3717, 0.6112, 0.5517],
  28558. [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617],
  28559. [0.6147, 0.4026, 0.6600, 0.2467, 0.4088, 0.2150, 0.5489, 0.5773],
  28560. [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869],
  28561. [0.6261, 0.4045, 0.8865, 0.5369, 0.3895, 0.4859, 0.6683, 0.5249],
  28562. [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882]],
  28563. device='cuda:0', dtype=torch.float64)
  28564. predictions are: tensor([[ 0.4012, -0.5602, 1.1020, -1.4905, -0.3912, -1.1780, 0.3697, 0.3340],
  28565. [ 0.3519, -0.6377, 1.8958, -0.1771, -0.4805, -0.0822, 0.7601, 0.2015],
  28566. [ 0.7661, -0.3157, 1.7028, -0.8933, -0.5319, -0.5710, 0.5506, 0.2958],
  28567. [ 0.3787, -0.5236, 1.8381, 0.0948, -0.3877, -0.3379, 0.4612, 0.4892],
  28568. [ 0.3725, -0.5542, 0.9727, -1.3297, -0.2899, -1.4031, 0.1775, 0.5426],
  28569. [ 0.1261, -0.7136, 1.7445, 0.1966, -0.3572, -0.2723, 0.2821, 0.1150],
  28570. [ 0.5290, -0.4819, 1.8134, 0.0249, -0.4198, 0.0279, 0.8776, 0.1686],
  28571. [ 0.4397, -0.5016, 1.8540, -0.5343, -0.5032, -0.5813, 0.1569, 0.1158]],
  28572. device='cuda:0', grad_fn=<AddmmBackward>)
  28573. landmarks are: tensor([[[ 0.5356, -0.4217, 1.0339, -1.4776, -0.5076, -1.1081, 0.4219,
  28574. 0.2853],
  28575. [ 0.6262, -0.4461, 1.8480, -0.0534, -0.5827, -0.1227, 0.6587,
  28576. 0.0774],
  28577. [ 0.5814, -0.4003, 1.6575, -0.8694, -0.6289, -0.5692, 0.5374,
  28578. 0.2622],
  28579. [ 0.5897, -0.3527, 1.8018, 0.2545, -0.5307, -0.3229, 0.3122,
  28580. 0.3084],
  28581. [ 0.5532, -0.4264, 0.7626, -1.1466, -0.3979, -1.2928, 0.2494,
  28582. 0.3808],
  28583. [ 0.5320, -0.4488, 1.6633, 0.3315, -0.5018, -0.2459, 0.0828,
  28584. -0.0370],
  28585. [ 0.6059, -0.4177, 1.8087, 0.1941, -0.4868, -0.0414, 0.8010,
  28586. 0.1385],
  28587. [ 0.5595, -0.3988, 1.7672, -0.4460, -0.5538, -0.5384, 0.0828,
  28588. -0.0310]]], device='cuda:0')
  28589. loss_train_step before backward: tensor(0.0174, device='cuda:0', grad_fn=<MseLossBackward>)
  28590. loss_train_step after backward: tensor(0.0174, device='cuda:0', grad_fn=<MseLossBackward>)
  28591. loss_train: 0.860898629296571
  28592. step: 42
  28593. running loss: 0.02049758641182312
  28594. Train Steps: 42/90 Loss: 0.0205 torch.Size([8, 600, 800])
  28595. torch.Size([8, 8])
  28596. tensor([[ nan, nan, 0.8525, 0.2217, 0.5413, 0.2367, 0.7367, 0.5482],
  28597. [0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869],
  28598. [ nan, nan, 0.7335, 0.2569, 0.3788, 0.2667, 0.5066, 0.5578],
  28599. [0.6034, 0.4011, 0.7350, 0.2533, 0.3438, 0.3367, 0.5516, 0.5084],
  28600. [0.6204, 0.4055, 0.8438, 0.5733, 0.4574, 0.4801, 0.5487, 0.5617],
  28601. [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283],
  28602. [0.6197, 0.4050, 0.7527, 0.2000, 0.4042, 0.2249, 0.5895, 0.4995],
  28603. [0.6132, 0.4037, 0.6963, 0.2217, 0.4100, 0.1950, 0.5395, 0.5175]],
  28604. device='cuda:0', dtype=torch.float64)
  28605. predictions are: tensor([[-1.6744, -1.8164, 1.6428, -1.1464, 0.1531, -1.0890, 1.0504, 0.3297],
  28606. [ 0.6633, -0.3424, 1.5168, -1.0733, -0.2207, -1.2959, 0.5642, 0.1678],
  28607. [-2.1848, -2.1812, 1.1882, -1.0478, -0.5552, -0.9504, 0.1036, 0.2840],
  28608. [ 0.5811, -0.4098, 1.1966, -1.0334, -0.6815, -0.6616, 0.3493, 0.1752],
  28609. [ 0.7715, -0.2316, 1.6568, 0.3646, -0.2515, 0.0227, 0.3323, 0.3487],
  28610. [ 0.6553, -0.3458, 1.7052, 0.3154, -0.5779, -0.2111, 0.3768, 0.2283],
  28611. [ 0.7520, -0.3007, 1.2283, -1.2327, -0.4260, -1.1571, 0.4222, 0.2272],
  28612. [ 0.6980, -0.3132, 1.0704, -1.2022, -0.4169, -1.3390, 0.2616, 0.2387]],
  28613. device='cuda:0', grad_fn=<AddmmBackward>)
  28614. landmarks are: tensor([[[-2.2859, -2.2859, 1.6517, -1.2620, 0.2141, -1.1928, 1.1166,
  28615. 0.2463],
  28616. [ 0.6058, -0.3892, 1.4208, -1.0927, -0.1843, -1.4237, 0.6154,
  28617. -0.0370],
  28618. [-2.2859, -2.2859, 1.1020, -1.0994, -0.5365, -1.0542, 0.0542,
  28619. 0.2905],
  28620. [ 0.5009, -0.4333, 1.1090, -1.1158, -0.6982, -0.7309, 0.2617,
  28621. 0.0622],
  28622. [ 0.5795, -0.4129, 1.6113, 0.3623, -0.1733, -0.0684, 0.2487,
  28623. 0.3084],
  28624. [ 0.5900, -0.4377, 1.6113, 0.3623, -0.5018, -0.3229, 0.3700,
  28625. 0.1544],
  28626. [ 0.5762, -0.4153, 1.1908, -1.3622, -0.4190, -1.2471, 0.4368,
  28627. 0.0213],
  28628. [ 0.5465, -0.4214, 0.9300, -1.2620, -0.3921, -1.3852, 0.2062,
  28629. 0.1043]]], device='cuda:0')
  28630. loss_train_step before backward: tensor(0.0178, device='cuda:0', grad_fn=<MseLossBackward>)
  28631. loss_train_step after backward: tensor(0.0178, device='cuda:0', grad_fn=<MseLossBackward>)
  28632. loss_train: 0.8787088342942297
  28633. step: 43
  28634. running loss: 0.02043508916963325
  28635. Train Steps: 43/90 Loss: 0.0204 torch.Size([8, 600, 800])
  28636. torch.Size([8, 8])
  28637. tensor([[0.6147, 0.4107, 0.8137, 0.3333, 0.3750, 0.2683, 0.5006, 0.5412],
  28638. [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837],
  28639. [0.6200, 0.4112, 0.8862, 0.4100, 0.3638, 0.4917, 0.6088, 0.6050],
  28640. [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896],
  28641. [0.6265, 0.4071, 0.8875, 0.3367, 0.3975, 0.3350, 0.6312, 0.5250],
  28642. [0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240],
  28643. [0.6339, 0.4149, 0.8800, 0.5000, 0.3900, 0.5283, 0.7541, 0.5424],
  28644. [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852]],
  28645. device='cuda:0', dtype=torch.float64)
  28646. predictions are: tensor([[ 0.0593, -0.7209, 1.4025, -0.7395, -0.5557, -1.0149, -0.0363, 0.1739],
  28647. [ 0.2105, -0.6398, 0.9159, -0.9487, -0.5745, -1.0085, 0.1048, 0.4227],
  28648. [ 0.5317, -0.4285, 1.8629, -0.3572, -0.4762, 0.0316, 0.7137, 0.3804],
  28649. [ 0.4177, -0.4962, 1.5345, -0.5597, -0.5811, -0.9179, 0.0925, 0.3951],
  28650. [ 0.6118, -0.4042, 1.9377, -0.8004, -0.4052, -0.8034, 0.7940, 0.1437],
  28651. [ 0.0240, -0.7626, 1.1786, -1.3303, -0.2676, -1.4906, 0.2640, 0.2023],
  28652. [ 0.5820, -0.4623, 1.8527, 0.0597, -0.4232, 0.3058, 1.2145, 0.1657],
  28653. [ 0.2763, -0.6095, 1.8244, 0.0415, -0.1050, -0.0060, 0.3189, 0.3642]],
  28654. device='cuda:0', grad_fn=<AddmmBackward>)
  28655. landmarks are: tensor([[[ 0.5532, -0.3888, 1.4727, -0.7463, -0.5538, -1.0465, 0.0265,
  28656. 0.2138],
  28657. [ 0.5629, -0.3871, 0.7243, -0.9581, -0.5827, -0.9849, 0.1288,
  28658. 0.4103],
  28659. [ 0.5777, -0.3868, 1.8076, -0.3921, -0.6058, -0.0149, 0.5259,
  28660. 0.5085],
  28661. [ 0.5485, -0.3997, 1.4445, -0.4895, -0.6000, -0.8309, 0.1878,
  28662. 0.4374],
  28663. [ 0.6078, -0.4057, 1.8134, -0.7309, -0.4499, -0.7386, 0.6298,
  28664. 0.1390],
  28665. [ 0.5676, -0.4112, 1.1898, -1.2467, -0.2940, -1.4622, 0.2103,
  28666. 0.1343],
  28667. [ 0.6421, -0.3695, 1.7788, 0.0236, -0.4845, 0.1544, 1.1971,
  28668. 0.2196],
  28669. [ 0.5500, -0.4060, 1.7326, 0.0236, -0.1554, -0.0226, 0.1438,
  28670. 0.4171]]], device='cuda:0')
  28671. loss_train_step before backward: tensor(0.0223, device='cuda:0', grad_fn=<MseLossBackward>)
  28672. loss_train_step after backward: tensor(0.0223, device='cuda:0', grad_fn=<MseLossBackward>)
  28673. loss_train: 0.9010163084603846
  28674. step: 44
  28675. running loss: 0.02047764337409965
  28676.  
  28677. Train Steps: 44/90 Loss: 0.0205 torch.Size([8, 600, 800])
  28678. torch.Size([8, 8])
  28679. tensor([[0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217],
  28680. [0.6117, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103],
  28681. [ nan, nan, 0.8938, 0.2850, 0.4662, 0.3117, 0.7406, 0.5528],
  28682. [0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901],
  28683. [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117],
  28684. [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100],
  28685. [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098],
  28686. [0.6151, 0.4125, 0.8738, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483]],
  28687. device='cuda:0', dtype=torch.float64)
  28688. predictions are: tensor([[ 0.8081, -0.2567, 1.7158, -0.0621, -0.3863, 0.1884, 0.6553, 0.2262],
  28689. [ 0.3466, -0.5509, 0.7081, -1.3252, -0.4496, -1.2155, 0.2184, 0.2836],
  28690. [-2.5157, -2.3895, 1.8121, -0.7981, -0.1205, -0.9275, 0.9323, 0.2574],
  28691. [ 0.4411, -0.4603, 1.7228, -0.2251, -0.3915, -0.2788, 0.1090, 0.1176],
  28692. [ 0.5141, -0.4182, 1.6298, -0.8154, -0.5195, -0.9728, 0.3624, 0.1820],
  28693. [ 0.7026, -0.3435, 1.6917, -0.4188, -0.5994, -0.5678, 0.6849, 0.2348],
  28694. [ 0.5526, -0.4045, 1.6131, -0.0275, -0.4721, 0.1363, 0.4688, 0.2287],
  28695. [ 0.6117, -0.3064, 1.5351, -0.2352, -0.5716, -0.6021, 0.0366, 0.4231]],
  28696. device='cuda:0', grad_fn=<AddmmBackward>)
  28697. landmarks are: tensor([[[ 0.6047, -0.4209, 1.7557, -0.0303, -0.4845, 0.2545, 0.6587,
  28698. 0.1236],
  28699. [ 0.5393, -0.4299, 0.7453, -1.3775, -0.5596, -1.1081, 0.1527,
  28700. 0.0712],
  28701. [-2.2859, -2.2859, 1.8423, -0.9695, -0.1323, -0.8463, 1.1349,
  28702. 0.2676],
  28703. [ 0.5663, -0.3955, 1.7788, -0.2382, -0.4037, -0.2690, 0.0828,
  28704. -0.0220],
  28705. [ 0.5799, -0.4329, 1.7210, -0.7694, -0.5711, -0.8771, 0.3988,
  28706. 0.0774],
  28707. [ 0.6220, -0.4436, 1.8711, -0.3691, -0.6173, -0.4999, 0.6702,
  28708. 0.0697],
  28709. [ 0.5466, -0.4706, 1.7198, -0.0903, -0.5712, 0.1261, 0.4733,
  28710. 0.0688],
  28711. [ 0.5551, -0.3807, 1.7499, -0.2459, -0.6346, -0.5384, 0.0871,
  28712. 0.2468]]], device='cuda:0')
  28713. loss_train_step before backward: tensor(0.0107, device='cuda:0', grad_fn=<MseLossBackward>)
  28714. loss_train_step after backward: tensor(0.0107, device='cuda:0', grad_fn=<MseLossBackward>)
  28715. loss_train: 0.9117077724076807
  28716. step: 45
  28717. running loss: 0.020260172720170683
  28718. Train Steps: 45/90 Loss: 0.0203 torch.Size([8, 600, 800])
  28719. torch.Size([8, 8])
  28720. tensor([[0.6275, 0.4081, 0.8063, 0.2017, 0.4825, 0.1583, 0.6156, 0.4869],
  28721. [0.6265, 0.4088, 0.8025, 0.1850, 0.4163, 0.2500, 0.6290, 0.4947],
  28722. [0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733],
  28723. [0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266],
  28724. [0.6311, 0.3998, 0.7975, 0.5767, 0.3838, 0.4850, 0.7327, 0.5343],
  28725. [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131],
  28726. [0.6131, 0.4037, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680],
  28727. [0.6082, 0.4042, 0.6975, 0.1917, 0.4100, 0.1983, 0.5502, 0.5578]],
  28728. device='cuda:0', dtype=torch.float64)
  28729. predictions are: tensor([[ 0.3473, -0.5753, 1.5471, -1.2589, -0.0220, -1.4570, 0.6313, 0.0374],
  28730. [ 0.3961, -0.5188, 1.5376, -1.2642, -0.3802, -1.0928, 0.6439, 0.0683],
  28731. [ 0.5461, -0.3509, 1.9025, 0.1118, -0.6093, -0.2064, 0.1833, 0.4084],
  28732. [ 0.5049, -0.4197, 1.1840, -1.0966, -0.3444, -1.0405, 0.4707, 0.5814],
  28733. [ 0.3431, -0.6076, 1.5917, 0.3166, -0.5506, 0.0278, 0.8716, 0.1111],
  28734. [ 0.4011, -0.5269, 1.8692, -0.4701, -0.7028, -0.4832, 0.3813, 0.0916],
  28735. [-0.0361, -0.7732, 0.9775, -0.9902, -0.5846, -0.9177, 0.0684, 0.3048],
  28736. [ 0.1723, -0.6870, 0.9510, -1.2858, -0.4434, -1.2211, 0.2540, 0.2577]],
  28737. device='cuda:0', grad_fn=<AddmmBackward>)
  28738. landmarks are: tensor([[[ 6.1241e-01, -4.0100e-01, 1.4381e+00, -1.3544e+00, -5.7275e-02,
  28739. -1.5546e+00, 5.5732e-01, -3.6943e-02],
  28740. [ 6.0785e-01, -3.9761e-01, 1.4208e+00, -1.4314e+00, -3.6328e-01,
  28741. -1.1312e+00, 6.1950e-01, -9.2270e-04],
  28742. [ 6.1207e-01, -3.1378e-01, 1.8423e+00, 8.1601e-03, -6.4619e-01,
  28743. -3.0747e-01, 3.4688e-01, 3.6228e-01],
  28744. [ 5.7460e-01, -4.0208e-01, 1.0801e+00, -1.1312e+00, -3.2286e-01,
  28745. -1.1081e+00, 4.8034e-01, 6.0842e-01],
  28746. [ 6.2895e-01, -4.3934e-01, 1.3977e+00, 3.7768e-01, -5.1339e-01,
  28747. -4.5727e-02, 1.0984e+00, 1.8214e-01],
  28748. [ 5.7829e-01, -4.2163e-01, 1.6847e+00, -5.0778e-01, -6.7321e-01,
  28749. -5.3774e-01, 4.7523e-01, 8.3916e-02],
  28750. [ 5.4590e-01, -4.2148e-01, 9.0432e-01, -9.8382e-01, -5.8268e-01,
  28751. -1.0388e+00, 1.2363e-01, 3.3782e-01],
  28752. [ 5.2315e-01, -4.1886e-01, 9.3580e-01, -1.4006e+00, -3.9215e-01,
  28753. -1.3698e+00, 2.5553e-01, 2.9064e-01]]], device='cuda:0')
  28754. loss_train_step before backward: tensor(0.0213, device='cuda:0', grad_fn=<MseLossBackward>)
  28755. loss_train_step after backward: tensor(0.0213, device='cuda:0', grad_fn=<MseLossBackward>)
  28756. loss_train: 0.9329804559238255
  28757. step: 46
  28758. running loss: 0.020282183824430988
  28759. Train Steps: 46/90 Loss: 0.0203 torch.Size([8, 600, 800])
  28760. torch.Size([8, 8])
  28761. tensor([[0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758],
  28762. [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609],
  28763. [0.6277, 0.4118, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938],
  28764. [0.6246, 0.4028, 0.8738, 0.4867, 0.4088, 0.5667, 0.6362, 0.5200],
  28765. [0.6228, 0.4119, 0.7938, 0.2233, 0.4674, 0.1773, 0.6188, 0.5433],
  28766. [0.6170, 0.4102, 0.7468, 0.3695, 0.3463, 0.3767, 0.5238, 0.5823],
  28767. [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377],
  28768. [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609]],
  28769. device='cuda:0', dtype=torch.float64)
  28770. predictions are: tensor([[ 3.4157e-01, -5.4580e-01, 1.7160e+00, -9.9776e-02, -5.3122e-01,
  28771. 1.0598e-01, 4.4080e-01, 7.4866e-02],
  28772. [ 4.5865e-01, -4.6616e-01, 1.6395e+00, -8.6564e-01, -4.8457e-01,
  28773. -8.4187e-01, 6.6704e-01, 2.8578e-01],
  28774. [ 9.5002e-02, -7.0816e-01, 1.8023e+00, -5.4315e-01, -6.0169e-01,
  28775. -1.0344e+00, 3.4626e-01, 5.6686e-02],
  28776. [ 4.0714e-01, -5.0531e-01, 1.6999e+00, -8.3630e-02, -4.9261e-01,
  28777. 3.0629e-01, 5.2600e-01, 1.3596e-01],
  28778. [ 7.4241e-01, -3.0988e-01, 1.2474e+00, -1.2521e+00, -2.4741e-01,
  28779. -1.4120e+00, 4.0297e-01, 2.0329e-01],
  28780. [ 4.5643e-01, -4.0569e-01, 1.1043e+00, -6.4953e-01, -7.8430e-01,
  28781. -5.2145e-01, -1.3662e-03, 4.6987e-01],
  28782. [ 4.2579e-01, -5.1258e-01, 1.6009e+00, 2.2429e-01, -5.5763e-01,
  28783. -8.9889e-02, 2.6553e-01, 1.4516e-01],
  28784. [ 1.9168e-01, -6.7445e-01, 1.6228e+00, -1.1956e+00, 1.8815e-01,
  28785. -1.3783e+00, 1.0478e+00, 3.1273e-01]], device='cuda:0',
  28786. grad_fn=<AddmmBackward>)
  28787. landmarks are: tensor([[[ 0.6072, -0.4250, 1.8249, -0.0072, -0.4037, 0.1082, 0.6730,
  28788. -0.0881],
  28789. [ 0.6026, -0.4032, 1.7326, -0.7771, -0.3691, -0.8617, 0.9704,
  28790. 0.3050],
  28791. [ 0.6135, -0.3841, 1.8654, -0.5153, -0.4614, -1.0619, 0.6195,
  28792. -0.0049],
  28793. [ 0.5990, -0.4256, 1.7499, -0.0380, -0.3979, 0.3315, 0.6529,
  28794. 0.1159],
  28795. [ 0.5908, -0.3832, 1.3804, -1.2543, -0.1270, -1.4671, 0.5721,
  28796. 0.2237],
  28797. [ 0.5639, -0.3911, 1.1634, -0.5794, -0.6866, -0.5461, 0.1334,
  28798. 0.4036],
  28799. [ 0.5775, -0.4607, 1.6741, 0.1962, -0.4036, -0.1212, 0.4588,
  28800. 0.1979],
  28801. [ 0.6372, -0.3575, 1.6633, -1.2082, 0.2199, -1.2467, 1.1313,
  28802. 0.3050]]], device='cuda:0')
  28803. loss_train_step before backward: tensor(0.0230, device='cuda:0', grad_fn=<MseLossBackward>)
  28804. loss_train_step after backward: tensor(0.0230, device='cuda:0', grad_fn=<MseLossBackward>)
  28805. loss_train: 0.9559376635588706
  28806. step: 47
  28807. running loss: 0.02033909922465682
  28808. Train Steps: 47/90 Loss: 0.0203 torch.Size([8, 600, 800])
  28809. torch.Size([8, 8])
  28810. tensor([[0.6143, 0.4040, 0.8237, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973],
  28811. [0.6198, 0.4101, 0.8838, 0.5283, 0.3763, 0.5267, 0.5913, 0.5567],
  28812. [0.6185, 0.4079, 0.8838, 0.4617, 0.4838, 0.5650, 0.6175, 0.5850],
  28813. [0.6213, 0.4001, 0.7712, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183],
  28814. [0.6339, 0.4102, 0.8588, 0.3133, 0.4425, 0.2117, 0.6417, 0.5089],
  28815. [ nan, nan, 0.6512, 0.1717, 0.4100, 0.1983, 0.5253, 0.5240],
  28816. [0.6124, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485],
  28817. [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389]],
  28818. device='cuda:0', dtype=torch.float64)
  28819. predictions are: tensor([[ 0.0754, -0.6728, 1.4005, -0.7200, -0.4994, -1.1253, 0.1119, 0.0342],
  28820. [ 0.7075, -0.3032, 1.7789, 0.2578, -0.6072, 0.0754, 0.5373, 0.2307],
  28821. [ 0.4886, -0.4103, 1.7513, -0.1235, -0.1657, 0.1917, 0.4897, 0.3107],
  28822. [ 0.6213, -0.3726, 1.1898, -1.2564, -0.3196, -1.3608, 0.4322, 0.1022],
  28823. [ 0.7552, -0.2970, 1.6800, -0.7415, -0.2835, -1.2006, 0.7921, 0.0721],
  28824. [-2.7242, -2.5366, 0.9469, -1.2379, -0.4670, -1.2380, 0.2041, 0.1697],
  28825. [ 0.4366, -0.4563, 0.9809, -0.8987, -0.6418, -0.8829, 0.1738, 0.2443],
  28826. [ 0.6094, -0.3594, 1.1859, -1.1509, -0.5651, -0.9896, 0.5455, 0.2325]],
  28827. device='cuda:0', grad_fn=<AddmmBackward>)
  28828. landmarks are: tensor([[[ 0.5515, -0.4201, 1.5189, -0.7463, -0.4037, -1.3082, 0.0869,
  28829. 0.0111],
  28830. [ 0.5770, -0.3918, 1.7961, 0.1544, -0.5480, 0.1467, 0.4450,
  28831. 0.2853],
  28832. [ 0.5707, -0.4017, 1.7961, -0.1535, -0.0515, 0.3238, 0.5663,
  28833. 0.4162],
  28834. [ 0.5838, -0.4379, 1.2764, -1.3082, -0.2824, -1.4545, 0.4162,
  28835. 0.1082],
  28836. [ 0.6421, -0.3912, 1.6806, -0.8386, -0.2420, -1.3082, 0.6780,
  28837. 0.0646],
  28838. [-2.2859, -2.2859, 0.7222, -1.4930, -0.3921, -1.3698, 0.1404,
  28839. 0.1343],
  28840. [ 0.5425, -0.3998, 0.9263, -0.8683, -0.6000, -1.0157, 0.0990,
  28841. 0.2476],
  28842. [ 0.5697, -0.4466, 1.1973, -1.1871, -0.4571, -0.9965, 0.5219,
  28843. 0.2032]]], device='cuda:0')
  28844. loss_train_step before backward: tensor(0.0158, device='cuda:0', grad_fn=<MseLossBackward>)
  28845. loss_train_step after backward: tensor(0.0158, device='cuda:0', grad_fn=<MseLossBackward>)
  28846. loss_train: 0.9717486617155373
  28847. step: 48
  28848. running loss: 0.02024476378574036
  28849.  
  28850. Train Steps: 48/90 Loss: 0.0202 torch.Size([8, 600, 800])
  28851. torch.Size([8, 8])
  28852. tensor([[0.6204, 0.4013, 0.8075, 0.2400, 0.4313, 0.2050, 0.5800, 0.5150],
  28853. [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400],
  28854. [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100],
  28855. [0.6296, 0.4060, 0.9100, 0.3267, 0.4726, 0.3367, 0.7446, 0.5355],
  28856. [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320],
  28857. [0.6203, 0.4073, 0.8189, 0.2398, 0.4400, 0.2054, 0.5929, 0.5501],
  28858. [0.6221, 0.4107, 0.7788, 0.3033, 0.3950, 0.2817, 0.6075, 0.5517],
  28859. [0.6167, 0.4048, 0.6831, 0.3639, 0.3763, 0.3017, 0.5700, 0.5883]],
  28860. device='cuda:0', dtype=torch.float64)
  28861. predictions are: tensor([[ 0.5644, -0.4089, 1.3489, -1.0632, -0.3554, -1.2604, 0.3068, 0.0804],
  28862. [ 0.4687, -0.4830, 1.2589, -1.1416, -0.2557, -1.2853, 0.4649, 0.1626],
  28863. [ 0.4234, -0.5423, 1.6914, 0.3361, -0.5816, -0.0057, 0.5142, -0.0243],
  28864. [ 0.4425, -0.5201, 1.8059, -0.6999, -0.2962, -0.6888, 1.0130, 0.1350],
  28865. [ 0.2810, -0.5907, 1.3972, -1.1654, -0.4216, -0.9874, 0.6505, 0.1130],
  28866. [ 0.6097, -0.3620, 1.3889, -1.0538, -0.3057, -1.2302, 0.3695, 0.1679],
  28867. [ 0.5667, -0.4004, 1.3944, -0.9263, -0.5386, -0.9340, 0.3893, 0.2306],
  28868. [-0.0884, -0.7821, 1.0243, -0.7315, -0.6894, -0.8069, 0.1109, 0.4189]],
  28869. device='cuda:0', grad_fn=<AddmmBackward>)
  28870. landmarks are: tensor([[[ 0.5796, -0.4326, 1.4439, -1.1774, -0.2940, -1.3390, 0.3931,
  28871. 0.0928],
  28872. [ 0.5842, -0.3848, 1.2938, -1.2159, -0.2132, -1.4160, 0.5778,
  28873. 0.2083],
  28874. [ 0.6125, -0.4273, 1.6402, 0.2006, -0.4788, -0.0303, 0.6182,
  28875. 0.0697],
  28876. [ 0.6224, -0.4105, 1.9173, -0.7771, -0.1030, -0.7308, 1.1532,
  28877. 0.1875],
  28878. [ 0.6141, -0.4153, 1.4208, -1.2697, -0.2940, -1.0234, 0.8644,
  28879. 0.1715],
  28880. [ 0.5792, -0.4048, 1.4965, -1.1781, -0.2534, -1.3371, 0.4528,
  28881. 0.2549],
  28882. [ 0.5875, -0.3888, 1.3111, -0.8848, -0.4614, -0.9849, 0.5201,
  28883. 0.2622],
  28884. [ 0.5626, -0.4162, 0.8692, -0.6051, -0.5480, -0.8925, 0.3469,
  28885. 0.4316]]], device='cuda:0')
  28886. loss_train_step before backward: tensor(0.0207, device='cuda:0', grad_fn=<MseLossBackward>)
  28887. loss_train_step after backward: tensor(0.0207, device='cuda:0', grad_fn=<MseLossBackward>)
  28888. loss_train: 0.9924390469677746
  28889. step: 49
  28890. running loss: 0.020253858101383155
  28891. Train Steps: 49/90 Loss: 0.0203 torch.Size([8, 600, 800])
  28892. torch.Size([8, 8])
  28893. tensor([[0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672],
  28894. [0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400],
  28895. [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333],
  28896. [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261],
  28897. [0.6040, 0.4002, 0.7338, 0.2267, 0.3975, 0.2100, 0.5231, 0.4778],
  28898. [0.6200, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391],
  28899. [0.6265, 0.4071, 0.8875, 0.3367, 0.3975, 0.3350, 0.6312, 0.5250],
  28900. [0.6264, 0.4049, 0.8988, 0.4633, 0.3813, 0.4983, 0.6326, 0.4843]],
  28901. device='cuda:0', dtype=torch.float64)
  28902. predictions are: tensor([[ 0.3356, -0.6246, 1.6195, 0.2512, -0.6513, -0.3454, 0.5973, 0.0688],
  28903. [ 0.6370, -0.3359, 1.3365, -1.0085, -0.1253, -1.4134, 0.3820, 0.2478],
  28904. [ 0.4861, -0.4302, 1.6799, -0.2698, -0.5321, -0.0616, 0.5071, 0.2307],
  28905. [ 0.3765, -0.4768, 1.6450, -0.1350, -0.4039, -0.1223, 0.2556, 0.1823],
  28906. [ 0.5536, -0.4313, 0.9053, -1.4331, -0.4854, -1.2592, 0.2617, 0.0754],
  28907. [ 0.6161, -0.3597, 1.4798, -1.0293, -0.3282, -1.1504, 0.4840, 0.1977],
  28908. [ 0.4582, -0.4561, 1.7081, -0.8268, -0.5026, -0.7563, 0.6979, 0.2014],
  28909. [ 0.6078, -0.3919, 1.8090, -0.2198, -0.5793, -0.0794, 0.6614, 0.0188]],
  28910. device='cuda:0', grad_fn=<AddmmBackward>)
  28911. landmarks are: tensor([[[ 0.6128, -0.4375, 1.7788, 0.2699, -0.6346, -0.2536, 0.5463,
  28912. -0.1278],
  28913. [ 0.5911, -0.3888, 1.4727, -0.9541, -0.0919, -1.4930, 0.3988,
  28914. 0.2083],
  28915. [ 0.6003, -0.3747, 1.8885, -0.1997, -0.5018, -0.0149, 0.5490,
  28916. 0.1775],
  28917. [ 0.5746, -0.3623, 1.7961, -0.1150, -0.3691, -0.0380, 0.2208,
  28918. 0.1439],
  28919. [ 0.5037, -0.4375, 1.1032, -1.2390, -0.4499, -1.3159, 0.1301,
  28920. -0.0791],
  28921. [ 0.5776, -0.3987, 1.6005, -0.9121, -0.3844, -1.2358, 0.4247,
  28922. 0.2043],
  28923. [ 0.6078, -0.4057, 1.8134, -0.7309, -0.4499, -0.7386, 0.6298,
  28924. 0.1390],
  28925. [ 0.6075, -0.4159, 1.8654, -0.1458, -0.5249, 0.0159, 0.6359,
  28926. -0.0490]]], device='cuda:0')
  28927. loss_train_step before backward: tensor(0.0099, device='cuda:0', grad_fn=<MseLossBackward>)
  28928. loss_train_step after backward: tensor(0.0099, device='cuda:0', grad_fn=<MseLossBackward>)
  28929. loss_train: 1.0023726033978164
  28930. step: 50
  28931. running loss: 0.02004745206795633
  28932. Train Steps: 50/90 Loss: 0.0200 torch.Size([8, 600, 800])
  28933. torch.Size([8, 8])
  28934. tensor([[0.6205, 0.4004, 0.8938, 0.4883, 0.3663, 0.5000, 0.6357, 0.4947],
  28935. [0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6138, 0.5450],
  28936. [0.6278, 0.4253, 0.8875, 0.5017, 0.4113, 0.2750, 0.5413, 0.6196],
  28937. [0.6153, 0.4117, 0.8688, 0.5167, 0.4895, 0.5647, 0.5524, 0.5136],
  28938. [ nan, nan, 0.7192, 0.2346, 0.4037, 0.2050, 0.5138, 0.5650],
  28939. [0.6219, 0.3934, 0.8688, 0.5267, 0.4313, 0.4967, 0.5988, 0.4983],
  28940. [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051],
  28941. [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817]],
  28942. device='cuda:0', dtype=torch.float64)
  28943. predictions are: tensor([[ 0.7533, -0.3049, 1.7014, -0.1500, -0.6275, -0.0598, 0.6284, -0.0155],
  28944. [ 0.7431, -0.2644, 1.7102, -0.3314, -0.3599, 0.0643, 0.5581, 0.1520],
  28945. [ 0.8763, -0.1319, 1.6137, -0.1098, -0.3940, -1.0488, 0.3424, 0.4068],
  28946. [ 0.5484, -0.3760, 1.6466, -0.0710, -0.1373, 0.0032, 0.3874, 0.1075],
  28947. [-2.5872, -2.4539, 0.9287, -1.3730, -0.5087, -1.2914, 0.1816, 0.1541],
  28948. [ 0.5824, -0.4260, 1.6279, 0.0228, -0.3571, -0.1051, 0.5001, -0.1236],
  28949. [ 0.7043, -0.3324, 1.7657, -0.3025, -0.6113, -0.6340, 0.7717, 0.0398],
  28950. [ 0.6638, -0.2967, 1.3242, -1.0323, -0.3195, -1.1665, 0.3938, 0.3216]],
  28951. device='cuda:0', grad_fn=<AddmmBackward>)
  28952. landmarks are: tensor([[[ 5.8031e-01, -4.3672e-01, 1.8423e+00, -3.0331e-02, -5.9423e-01,
  28953. 2.3557e-02, 6.5034e-01, -9.2270e-04],
  28954. [ 5.9602e-01, -4.1016e-01, 1.8018e+00, -1.6120e-01, -3.3441e-01,
  28955. 1.1594e-01, 5.4896e-01, 2.3141e-01],
  28956. [ 6.1386e-01, -3.2163e-01, 1.8134e+00, 3.1255e-02, -3.8637e-01,
  28957. -1.0157e+00, 2.1441e-01, 5.7619e-01],
  28958. [ 5.5635e-01, -3.8422e-01, 1.7268e+00, 1.0054e-01, -2.4997e-02,
  28959. 3.2255e-01, 2.6581e-01, 8.6245e-02],
  28960. [-2.2859e+00, -2.2859e+00, 1.0361e+00, -1.2021e+00, -4.2102e-01,
  28961. -1.3390e+00, 8.7067e-02, 3.2379e-01],
  28962. [ 5.8643e-01, -4.6898e-01, 1.7268e+00, 1.4673e-01, -2.9400e-01,
  28963. 8.1601e-03, 4.7968e-01, 1.5858e-02],
  28964. [ 6.2038e-01, -4.3356e-01, 1.8654e+00, -6.8822e-02, -6.0577e-01,
  28965. -5.2302e-01, 6.5034e-01, 4.7170e-02],
  28966. [ 5.9636e-01, -3.3795e-01, 1.4785e+00, -8.3865e-01, -2.4203e-01,
  28967. -1.0619e+00, 3.2379e-01, 4.0077e-01]]], device='cuda:0')
  28968. loss_train_step before backward: tensor(0.0161, device='cuda:0', grad_fn=<MseLossBackward>)
  28969. loss_train_step after backward: tensor(0.0161, device='cuda:0', grad_fn=<MseLossBackward>)
  28970. loss_train: 1.0185070815496147
  28971. step: 51
  28972. running loss: 0.01997072708920813
  28973. Train Steps: 51/90 Loss: 0.0200 torch.Size([8, 600, 800])
  28974. torch.Size([8, 8])
  28975. tensor([[0.6142, 0.4127, 0.7575, 0.3067, 0.3438, 0.4383, 0.5778, 0.5207],
  28976. [0.6078, 0.4033, 0.8019, 0.3055, 0.3450, 0.4200, 0.6025, 0.5550],
  28977. [0.6250, 0.4131, 0.8688, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367],
  28978. [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383],
  28979. [0.6188, 0.4099, 0.7400, 0.2433, 0.3962, 0.2750, 0.6162, 0.5467],
  28980. [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633],
  28981. [0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719],
  28982. [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426]],
  28983. device='cuda:0', dtype=torch.float64)
  28984. predictions are: tensor([[ 0.5031, -0.4425, 1.2040, -0.7772, -0.5495, -0.3063, 0.4339, 0.1906],
  28985. [ 0.3934, -0.5387, 1.4250, -0.7125, -0.5673, -0.3420, 0.6277, 0.2145],
  28986. [ 0.7193, -0.3086, 1.8446, -0.8237, -0.1610, -1.2996, 0.7239, 0.0600],
  28987. [ 0.5397, -0.4609, 1.6552, -0.6620, -0.5809, -0.4994, 0.7042, 0.1421],
  28988. [ 0.8671, -0.2594, 1.1405, -1.1521, -0.4197, -1.0268, 0.5970, 0.0799],
  28989. [ 0.6018, -0.3752, 1.6716, -0.6814, -0.4735, -0.9963, 0.3710, 0.0484],
  28990. [ 0.2039, -0.6625, 1.6431, -0.2761, -0.5604, -0.3393, 0.3029, 0.1686],
  28991. [ 0.6398, -0.3982, 1.2430, -1.0984, -0.1309, -1.4904, 0.4661, 0.0773]],
  28992. device='cuda:0', grad_fn=<AddmmBackward>)
  28993. landmarks are: tensor([[[ 0.5509, -0.3798, 1.2129, -0.8694, -0.6982, -0.2613, 0.3830,
  28994. 0.1193],
  28995. [ 0.5215, -0.4232, 1.4182, -0.8747, -0.6924, -0.3460, 0.4970,
  28996. 0.2776],
  28997. [ 0.6011, -0.3778, 1.7268, -0.9079, -0.3113, -1.1928, 0.5605,
  28998. 0.1929],
  28999. [ 0.5779, -0.4275, 1.5894, -0.8362, -0.6577, -0.5153, 0.5605,
  29000. 0.2006],
  29001. [ 0.5725, -0.3926, 1.1321, -1.1620, -0.4557, -1.0157, 0.5605,
  29002. 0.2391],
  29003. [ 0.5944, -0.4008, 1.6748, -0.9002, -0.5711, -0.8848, 0.2776,
  29004. 0.3161],
  29005. [ 0.5367, -0.4294, 1.5709, -0.4999, -0.6693, -0.3075, 0.2455,
  29006. 0.3559],
  29007. [ 0.5784, -0.4153, 1.2972, -1.2541, -0.2265, -1.4553, 0.4741,
  29008. 0.2203]]], device='cuda:0')
  29009. loss_train_step before backward: tensor(0.0147, device='cuda:0', grad_fn=<MseLossBackward>)
  29010. loss_train_step after backward: tensor(0.0147, device='cuda:0', grad_fn=<MseLossBackward>)
  29011. loss_train: 1.0332335312850773
  29012. step: 52
  29013. running loss: 0.019869875601636104
  29014.  
  29015. Train Steps: 52/90 Loss: 0.0199 torch.Size([8, 600, 800])
  29016. torch.Size([8, 8])
  29017. tensor([[0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904],
  29018. [0.6120, 0.4014, 0.6863, 0.2817, 0.3700, 0.2783, 0.5513, 0.5667],
  29019. [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879],
  29020. [0.6064, 0.4019, 0.8650, 0.4517, 0.4037, 0.5367, 0.5703, 0.5609],
  29021. [0.6201, 0.4017, 0.8871, 0.4621, 0.3517, 0.4675, 0.5999, 0.5106],
  29022. [0.6255, 0.4017, 0.8688, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901],
  29023. [0.6113, 0.4104, 0.8650, 0.4250, 0.3925, 0.5967, 0.5787, 0.5116],
  29024. [0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433]],
  29025. device='cuda:0', dtype=torch.float64)
  29026. predictions are: tensor([[ 7.2992e-01, -3.0336e-01, 1.6313e+00, -1.1895e+00, -1.6060e-01,
  29027. -1.5407e+00, 7.1356e-01, -1.1083e-04],
  29028. [ 8.3763e-01, -2.4081e-01, 9.4407e-01, -1.0714e+00, -6.1049e-01,
  29029. -1.1304e+00, 3.6978e-01, 2.8219e-01],
  29030. [ 5.3184e-01, -4.5070e-01, 1.7964e+00, -2.1034e-01, -4.5707e-01,
  29031. 1.3236e-01, 5.4005e-01, 1.0987e-02],
  29032. [ 4.1670e-01, -5.2252e-01, 1.7129e+00, -1.5120e-01, -3.7215e-01,
  29033. 7.9002e-02, 4.5440e-01, 2.4114e-01],
  29034. [ 6.4734e-01, -3.8203e-01, 1.7813e+00, -1.8053e-01, -6.0651e-01,
  29035. -2.4972e-01, 4.8176e-01, 1.4656e-01],
  29036. [ 8.4762e-01, -2.9559e-01, 1.6944e+00, -8.0337e-01, -6.5135e-01,
  29037. -8.4902e-01, 8.1033e-01, -2.9006e-02],
  29038. [ 5.1166e-01, -4.3294e-01, 1.7072e+00, -3.1367e-01, -4.1504e-01,
  29039. 2.0122e-01, 4.7451e-01, 1.6525e-01],
  29040. [ 5.3567e-01, -4.6310e-01, 1.7171e+00, 1.5583e-01, 6.1324e-03,
  29041. -2.4901e-01, 2.7454e-01, 2.4279e-01]], device='cuda:0',
  29042. grad_fn=<AddmmBackward>)
  29043. landmarks are: tensor([[[ 0.6126, -0.3871, 1.6229, -1.0773, -0.2132, -1.3698, 0.5829,
  29044. -0.0209],
  29045. [ 0.5410, -0.4321, 0.8838, -0.9849, -0.5769, -1.0003, 0.2603,
  29046. 0.3315],
  29047. [ 0.5676, -0.4327, 1.8252, -0.2095, -0.4886, 0.3084, 0.4727,
  29048. -0.0322],
  29049. [ 0.5151, -0.4296, 1.7095, -0.1997, -0.4210, 0.1929, 0.3484,
  29050. 0.3047],
  29051. [ 0.5783, -0.4306, 1.8114, -0.1515, -0.6617, -0.1268, 0.4851,
  29052. 0.0727],
  29053. [ 0.6031, -0.4307, 1.7268, -0.8002, -0.6058, -0.6462, 0.6442,
  29054. -0.0220],
  29055. [ 0.5376, -0.3903, 1.7095, -0.3229, -0.4730, 0.4701, 0.3871,
  29056. 0.0772],
  29057. [ 0.5295, -0.4373, 1.6553, 0.1011, 0.0380, 0.0671, 0.0813,
  29058. 0.2237]]], device='cuda:0')
  29059. loss_train_step before backward: tensor(0.0120, device='cuda:0', grad_fn=<MseLossBackward>)
  29060. loss_train_step after backward: tensor(0.0120, device='cuda:0', grad_fn=<MseLossBackward>)
  29061. loss_train: 1.0451941271312535
  29062. step: 53
  29063. running loss: 0.019720643908136857
  29064. Train Steps: 53/90 Loss: 0.0197 torch.Size([8, 600, 800])
  29065. torch.Size([8, 8])
  29066. tensor([[0.6157, 0.3956, 0.8323, 0.4138, 0.3479, 0.4431, 0.5914, 0.5131],
  29067. [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402],
  29068. [0.6183, 0.4076, 0.8838, 0.4517, 0.3813, 0.4483, 0.5775, 0.5633],
  29069. [0.6364, 0.4154, 0.8938, 0.3717, 0.4500, 0.2583, 0.6448, 0.5285],
  29070. [0.6091, 0.3997, 0.8314, 0.4334, 0.3788, 0.4550, 0.5213, 0.5656],
  29071. [0.6100, 0.4016, 0.8600, 0.5067, 0.4612, 0.5233, 0.5086, 0.5519],
  29072. [0.6124, 0.4069, 0.8314, 0.5001, 0.3738, 0.4650, 0.5167, 0.5402],
  29073. [0.6277, 0.4118, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938]],
  29074. device='cuda:0', dtype=torch.float64)
  29075. predictions are: tensor([[ 0.6368, -0.3853, 1.6120, -0.5413, -0.6199, -0.2305, 0.5398, 0.0600],
  29076. [ 0.5209, -0.4816, 1.6451, 0.1398, -0.3906, -0.1227, 0.5984, 0.0783],
  29077. [ 0.9131, -0.1735, 1.7377, -0.3481, -0.4845, -0.1768, 0.5186, 0.2881],
  29078. [ 0.6196, -0.3521, 1.8370, -0.8313, -0.1900, -1.2255, 0.7835, 0.2021],
  29079. [ 0.6076, -0.3732, 1.5815, -0.3898, -0.5213, -0.1574, 0.3139, 0.2048],
  29080. [ 0.6049, -0.3922, 1.6878, 0.0123, -0.1104, 0.0630, 0.2517, 0.1497],
  29081. [ 0.7485, -0.2795, 1.5738, -0.0625, -0.4706, -0.1227, 0.3277, 0.2029],
  29082. [ 0.6269, -0.3977, 1.9121, -0.6803, -0.4985, -1.1353, 0.7050, -0.0465]],
  29083. device='cuda:0', grad_fn=<AddmmBackward>)
  29084. landmarks are: tensor([[[ 0.5581, -0.4586, 1.5586, -0.3747, -0.6792, -0.2391, 0.4455,
  29085. 0.0840],
  29086. [ 0.5780, -0.4565, 1.6221, 0.2532, -0.3728, -0.1718, 0.4357,
  29087. 0.2091],
  29088. [ 0.5700, -0.4032, 1.7961, -0.1997, -0.5249, -0.2151, 0.3815,
  29089. 0.3161],
  29090. [ 0.6537, -0.3671, 1.8423, -0.5692, -0.2074, -1.0927, 0.6926,
  29091. 0.1554],
  29092. [ 0.5275, -0.4396, 1.5543, -0.2841, -0.5365, -0.1843, 0.1221,
  29093. 0.3265],
  29094. [ 0.5318, -0.4310, 1.6864, 0.0543, -0.1554, 0.1313, 0.0635,
  29095. 0.2634],
  29096. [ 0.5425, -0.4067, 1.5543, 0.0241, -0.5596, -0.1381, 0.1005,
  29097. 0.2093],
  29098. [ 0.6135, -0.3841, 1.8654, -0.5153, -0.4614, -1.0619, 0.6195,
  29099. -0.0049]]], device='cuda:0')
  29100. loss_train_step before backward: tensor(0.0117, device='cuda:0', grad_fn=<MseLossBackward>)
  29101. loss_train_step after backward: tensor(0.0117, device='cuda:0', grad_fn=<MseLossBackward>)
  29102. loss_train: 1.0568622411228716
  29103. step: 54
  29104. running loss: 0.019571522983756882
  29105. Train Steps: 54/90 Loss: 0.0196 torch.Size([8, 600, 800])
  29106. torch.Size([8, 8])
  29107. tensor([[0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320],
  29108. [0.6170, 0.4102, 0.7468, 0.3695, 0.3463, 0.3767, 0.5238, 0.5823],
  29109. [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426],
  29110. [0.6147, 0.4107, 0.8137, 0.3333, 0.3750, 0.2683, 0.5006, 0.5412],
  29111. [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250],
  29112. [0.6200, 0.3999, 0.8653, 0.5207, 0.4100, 0.5125, 0.5975, 0.5103],
  29113. [ nan, nan, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999],
  29114. [0.6203, 0.4076, 0.8611, 0.2878, 0.4050, 0.2554, 0.5907, 0.5496]],
  29115. device='cuda:0', dtype=torch.float64)
  29116. predictions are: tensor([[ 0.7829, -0.3401, 1.8872, -0.0882, -0.5316, -0.3178, 0.9825, 0.0237],
  29117. [ 0.7225, -0.2570, 1.2862, -0.5971, -0.6629, -0.5322, 0.2681, 0.3659],
  29118. [ 0.7286, -0.3055, 1.2713, -1.2098, -0.2073, -1.4156, 0.5210, 0.1292],
  29119. [ 0.6605, -0.3209, 1.4038, -0.7051, -0.5712, -0.9730, 0.0354, 0.1156],
  29120. [ 0.8193, -0.2439, 1.8907, 0.0371, -0.3015, 0.2836, 0.6148, 0.1104],
  29121. [ 0.8284, -0.2746, 1.8211, 0.1799, -0.3628, 0.1998, 0.5393, -0.0055],
  29122. [-1.1938, -1.5406, 1.1514, -1.2486, -0.3078, -1.3804, 0.1980, 0.1326],
  29123. [ 0.6440, -0.3563, 1.7050, -0.9408, -0.3089, -0.9373, 0.5627, 0.1509]],
  29124. device='cuda:0', grad_fn=<AddmmBackward>)
  29125. landmarks are: tensor([[[ 0.6174, -0.4490, 1.8885, -0.0996, -0.4845, -0.3691, 0.9814,
  29126. 0.1715],
  29127. [ 0.5639, -0.3911, 1.1634, -0.5794, -0.6866, -0.5461, 0.1334,
  29128. 0.4036],
  29129. [ 0.5784, -0.4153, 1.2972, -1.2541, -0.2265, -1.4553, 0.4741,
  29130. 0.2203],
  29131. [ 0.5532, -0.3888, 1.4727, -0.7463, -0.5538, -1.0465, 0.0265,
  29132. 0.2138],
  29133. [ 0.5991, -0.3803, 1.8018, -0.0534, -0.3460, 0.1852, 0.5374,
  29134. 0.1390],
  29135. [ 0.5778, -0.4389, 1.7107, 0.1192, -0.3921, 0.0815, 0.4741,
  29136. 0.0711],
  29137. [-2.2859, -2.2859, 1.3400, -1.0388, -0.3055, -1.4930, 0.1157,
  29138. 0.0231],
  29139. [ 0.5790, -0.4031, 1.6915, -0.9564, -0.4152, -1.1063, 0.4425,
  29140. 0.2528]]], device='cuda:0')
  29141. loss_train_step before backward: tensor(0.0374, device='cuda:0', grad_fn=<MseLossBackward>)
  29142. loss_train_step after backward: tensor(0.0374, device='cuda:0', grad_fn=<MseLossBackward>)
  29143. loss_train: 1.0942944386042655
  29144. step: 55
  29145. running loss: 0.019896262520077555
  29146. Train Steps: 55/90 Loss: 0.0199 torch.Size([8, 600, 800])
  29147. torch.Size([8, 8])
  29148. tensor([[0.6179, 0.4008, 0.7505, 0.2678, 0.4368, 0.1891, 0.5831, 0.5263],
  29149. [0.6219, 0.4097, 0.8738, 0.3400, 0.3563, 0.4117, 0.5975, 0.5683],
  29150. [0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183],
  29151. [0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019],
  29152. [0.6115, 0.4081, 0.6725, 0.2433, 0.4088, 0.1933, 0.5167, 0.5544],
  29153. [0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090],
  29154. [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5637, 0.5633],
  29155. [0.6246, 0.4008, 0.8757, 0.5088, 0.4101, 0.5392, 0.6644, 0.5133]],
  29156. device='cuda:0', dtype=torch.float64)
  29157. predictions are: tensor([[ 0.5863, -0.4174, 1.1850, -1.1955, -0.2477, -1.3069, 0.3485, 0.1124],
  29158. [ 0.7275, -0.3032, 1.7512, -0.7057, -0.5881, -0.3120, 0.4027, 0.2238],
  29159. [ 0.7115, -0.3709, 1.8298, -0.0473, -0.5467, -0.2742, 0.4645, 0.0810],
  29160. [ 0.7314, -0.3769, 1.9074, -0.1583, -0.5646, -0.3302, 0.7126, 0.0212],
  29161. [ 0.6423, -0.3575, 0.8832, -1.2260, -0.3278, -1.3196, 0.0393, 0.2563],
  29162. [ 0.6789, -0.3773, 1.8181, -0.6685, -0.3828, -0.8828, 0.6894, 0.0553],
  29163. [ 0.6784, -0.3066, 1.7560, -0.0381, -0.5392, -0.4158, 0.2583, 0.3503],
  29164. [ 0.5078, -0.5002, 1.7740, -0.0045, -0.3177, 0.4120, 0.7222, 0.1060]],
  29165. device='cuda:0', grad_fn=<AddmmBackward>)
  29166. landmarks are: tensor([[[ 0.5680, -0.4345, 1.1806, -1.0490, -0.2683, -1.4127, 0.4074,
  29167. 0.1449],
  29168. [ 0.5867, -0.3937, 1.7499, -0.7155, -0.6404, -0.3844, 0.4739,
  29169. 0.3392],
  29170. [ 0.5944, -0.4543, 1.8018, 0.0082, -0.6058, -0.4306, 0.4162,
  29171. 0.1082],
  29172. [ 0.6372, -0.4129, 1.8942, -0.0765, -0.6173, -0.4768, 0.6999,
  29173. 0.0325],
  29174. [ 0.5385, -0.4010, 0.8203, -1.1620, -0.3979, -1.3929, 0.1005,
  29175. 0.2747],
  29176. [ 0.6165, -0.4249, 1.7961, -0.5384, -0.4499, -0.9695, 0.6401,
  29177. 0.0652],
  29178. [ 0.5902, -0.3493, 1.7961, -0.0072, -0.5942, -0.5615, 0.3180,
  29179. 0.3161],
  29180. [ 0.5993, -0.4345, 1.7587, 0.0641, -0.3917, 0.2048, 0.7827,
  29181. 0.0852]]], device='cuda:0')
  29182. loss_train_step before backward: tensor(0.0058, device='cuda:0', grad_fn=<MseLossBackward>)
  29183. loss_train_step after backward: tensor(0.0058, device='cuda:0', grad_fn=<MseLossBackward>)
  29184. loss_train: 1.1001434274949133
  29185. step: 56
  29186. running loss: 0.01964541834812345
  29187.  
  29188. Train Steps: 56/90 Loss: 0.0196 torch.Size([8, 600, 800])
  29189. torch.Size([8, 8])
  29190. tensor([[ nan, nan, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999],
  29191. [0.6126, 0.4067, 0.8638, 0.5383, 0.4188, 0.4850, 0.5016, 0.5392],
  29192. [0.6245, 0.4100, 0.7762, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417],
  29193. [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398],
  29194. [0.6200, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183],
  29195. [0.6204, 0.4013, 0.8075, 0.2400, 0.4313, 0.2050, 0.5800, 0.5150],
  29196. [0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297],
  29197. [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852]],
  29198. device='cuda:0', dtype=torch.float64)
  29199. predictions are: tensor([[-1.6910, -1.8692, 1.1434, -1.2963, -0.3648, -1.2930, 0.1842, 0.1635],
  29200. [ 0.7451, -0.2723, 1.7204, 0.1904, -0.3997, 0.1187, 0.1519, 0.1289],
  29201. [ 0.6097, -0.3337, 1.2839, -1.0808, -0.0696, -1.4428, 0.3928, 0.2679],
  29202. [ 0.7237, -0.3450, 1.7572, 0.1803, -0.5939, -0.1386, 0.5070, 0.0829],
  29203. [ 0.8000, -0.2632, 1.5401, -0.9677, -0.5274, -0.9060, 0.6557, 0.0595],
  29204. [ 0.5654, -0.4025, 1.4035, -1.1105, -0.3297, -1.2302, 0.4009, 0.1177],
  29205. [ 0.8601, -0.2298, 1.8324, 0.2251, -0.6189, -0.1713, 0.7841, 0.0564],
  29206. [ 0.7536, -0.2819, 1.8000, -0.0206, -0.2430, 0.1234, 0.2688, 0.2988]],
  29207. device='cuda:0', grad_fn=<AddmmBackward>)
  29208. landmarks are: tensor([[[-2.2859, -2.2859, 1.3400, -1.0388, -0.3055, -1.4930, 0.1157,
  29209. 0.0231],
  29210. [ 0.5436, -0.4076, 1.7037, 0.2006, -0.3517, -0.0457, 0.0311,
  29211. 0.2048],
  29212. [ 0.5985, -0.3921, 1.2995, -1.0927, 0.0062, -1.5854, 0.4277,
  29213. 0.2160],
  29214. [ 0.5786, -0.4463, 1.6655, 0.2216, -0.5115, -0.2675, 0.4236,
  29215. 0.2075],
  29216. [ 0.5781, -0.4129, 1.4142, -0.9057, -0.5115, -0.9937, 0.4621,
  29217. 0.1080],
  29218. [ 0.5796, -0.4326, 1.4439, -1.1774, -0.2940, -1.3390, 0.3931,
  29219. 0.0928],
  29220. [ 0.6504, -0.3647, 1.7730, 0.2930, -0.6058, -0.2382, 0.7109,
  29221. 0.1608],
  29222. [ 0.5500, -0.4060, 1.7326, 0.0236, -0.1554, -0.0226, 0.1438,
  29223. 0.4171]]], device='cuda:0')
  29224. loss_train_step before backward: tensor(0.0204, device='cuda:0', grad_fn=<MseLossBackward>)
  29225. loss_train_step after backward: tensor(0.0204, device='cuda:0', grad_fn=<MseLossBackward>)
  29226. loss_train: 1.1205182750709355
  29227. step: 57
  29228. running loss: 0.019658215352121675
  29229. Train Steps: 57/90 Loss: 0.0197 torch.Size([8, 600, 800])
  29230. torch.Size([8, 8])
  29231. tensor([[0.6252, 0.4158, 0.8988, 0.4083, 0.3788, 0.4783, 0.6225, 0.5633],
  29232. [0.6043, 0.4022, 0.6887, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136],
  29233. [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153],
  29234. [0.6250, 0.3993, 0.9138, 0.4333, 0.3763, 0.5217, 0.6995, 0.5320],
  29235. [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413],
  29236. [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389],
  29237. [0.6261, 0.4029, 0.8720, 0.3364, 0.3665, 0.3753, 0.6531, 0.5183],
  29238. [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150]],
  29239. device='cuda:0', dtype=torch.float64)
  29240. predictions are: tensor([[ 0.7240, -0.3128, 1.8758, -0.1588, -0.5622, -0.1100, 0.3184, 0.3145],
  29241. [ 0.6164, -0.3884, 0.9456, -1.0949, -0.4927, -1.2268, 0.0999, 0.0704],
  29242. [ 0.5727, -0.4196, 1.4462, -0.6541, -0.6380, -0.7790, 0.2244, 0.1123],
  29243. [ 0.7500, -0.3441, 1.8785, -0.1390, -0.5367, 0.1114, 0.7222, 0.2040],
  29244. [ 0.3065, -0.6051, 1.9755, -0.6526, 0.0485, -1.1576, 0.8616, 0.2635],
  29245. [ 0.5309, -0.4451, 1.2675, -1.0254, -0.4772, -1.0516, 0.3530, 0.2737],
  29246. [ 0.7435, -0.3372, 1.7588, -0.5859, -0.5617, -0.5718, 0.6055, 0.1405],
  29247. [ 0.4607, -0.4934, 1.8087, 0.0506, -0.2326, 0.4321, 0.3402, 0.1789]],
  29248. device='cuda:0', grad_fn=<AddmmBackward>)
  29249. landmarks are: tensor([[[ 0.6017, -0.3654, 1.8654, -0.3998, -0.5365, -0.0765, 0.5894,
  29250. 0.3161],
  29251. [ 0.5053, -0.4281, 0.8954, -1.3698, -0.5423, -1.1389, 0.2453,
  29252. 0.0862],
  29253. [ 0.5614, -0.4514, 1.3252, -0.7240, -0.6237, -0.7215, 0.4299,
  29254. 0.0942],
  29255. [ 0.6010, -0.4417, 1.9346, -0.2844, -0.5480, 0.1236, 0.9448,
  29256. 0.1715],
  29257. [ 0.6108, -0.4201, 1.8711, -0.7848, -0.0053, -1.2236, 1.0362,
  29258. 0.2142],
  29259. [ 0.5697, -0.4466, 1.1973, -1.1871, -0.4571, -0.9965, 0.5219,
  29260. 0.2032],
  29261. [ 0.6059, -0.4249, 1.7420, -0.7321, -0.5931, -0.5523, 0.7307,
  29262. 0.1080],
  29263. [ 0.6014, -0.3883, 1.8192, -0.1150, -0.2882, 0.4008, 0.5201,
  29264. 0.0928]]], device='cuda:0')
  29265. loss_train_step before backward: tensor(0.0138, device='cuda:0', grad_fn=<MseLossBackward>)
  29266. loss_train_step after backward: tensor(0.0138, device='cuda:0', grad_fn=<MseLossBackward>)
  29267. loss_train: 1.134296880569309
  29268. step: 58
  29269. running loss: 0.01955684276843636
  29270. Train Steps: 58/90 Loss: 0.0196 torch.Size([8, 600, 800])
  29271. torch.Size([8, 8])
  29272. tensor([[0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767],
  29273. [0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400],
  29274. [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791],
  29275. [0.6078, 0.4033, 0.8019, 0.3055, 0.3450, 0.4200, 0.6025, 0.5550],
  29276. [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283],
  29277. [0.6231, 0.3973, 0.8650, 0.3950, 0.3625, 0.3183, 0.5837, 0.5167],
  29278. [0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748],
  29279. [0.6040, 0.4002, 0.7338, 0.2267, 0.3975, 0.2100, 0.5231, 0.4778]],
  29280. device='cuda:0', dtype=torch.float64)
  29281. predictions are: tensor([[ 0.6055, -0.4108, 1.7864, 0.0511, -0.3500, -0.0902, 0.1680, 0.2737],
  29282. [ 0.7388, -0.3232, 1.8136, -0.4222, -0.5589, -0.1588, 0.5349, 0.2467],
  29283. [ 0.7091, -0.3533, 1.8356, -0.0353, -0.3907, 0.1420, 0.4466, 0.0463],
  29284. [ 0.5975, -0.3956, 1.4275, -0.8469, -0.5958, -0.3813, 0.4986, 0.3251],
  29285. [ 0.5958, -0.3976, 1.8374, -0.2642, -0.5575, -0.5204, 0.3262, 0.2197],
  29286. [ 0.6485, -0.3895, 1.7882, -0.4670, -0.5247, -0.8635, 0.3813, 0.1579],
  29287. [ 0.4192, -0.5455, 1.6479, 0.2483, -0.4039, -0.3421, 0.7505, 0.3025],
  29288. [ 0.4611, -0.4840, 1.1231, -1.3008, -0.3819, -1.3606, 0.1314, 0.0664]],
  29289. device='cuda:0', grad_fn=<AddmmBackward>)
  29290. landmarks are: tensor([[[ 0.5610, -0.4381, 1.7730, 0.0390, -0.4326, -0.1458, 0.1794,
  29291. 0.3777],
  29292. [ 0.5778, -0.4393, 1.8018, -0.4614, -0.6693, -0.1381, 0.5490,
  29293. 0.2083],
  29294. [ 0.6074, -0.4223, 1.8192, -0.0303, -0.4152, 0.1236, 0.6524,
  29295. -0.0731],
  29296. [ 0.5215, -0.4232, 1.4182, -0.8747, -0.6924, -0.3460, 0.4970,
  29297. 0.2776],
  29298. [ 0.5880, -0.3887, 1.8423, -0.3306, -0.6231, -0.5230, 0.4046,
  29299. 0.1544],
  29300. [ 0.5924, -0.4507, 1.7095, -0.4614, -0.6115, -0.8156, 0.4104,
  29301. 0.1005],
  29302. [ 0.6421, -0.3647, 1.5940, 0.3084, -0.5192, -0.3691, 1.1057,
  29303. 0.3692],
  29304. [ 0.5037, -0.4375, 1.1032, -1.2390, -0.4499, -1.3159, 0.1301,
  29305. -0.0791]]], device='cuda:0')
  29306. loss_train_step before backward: tensor(0.0076, device='cuda:0', grad_fn=<MseLossBackward>)
  29307. loss_train_step after backward: tensor(0.0076, device='cuda:0', grad_fn=<MseLossBackward>)
  29308. loss_train: 1.1419263114221394
  29309. step: 59
  29310. running loss: 0.01935468324444304
  29311. Train Steps: 59/90 Loss: 0.0194 torch.Size([8, 600, 800])
  29312. torch.Size([8, 8])
  29313. tensor([[0.6126, 0.4067, 0.8638, 0.5383, 0.4188, 0.4850, 0.5016, 0.5392],
  29314. [0.6222, 0.3957, 0.8838, 0.5017, 0.3937, 0.4600, 0.5900, 0.5017],
  29315. [ nan, nan, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533],
  29316. [0.6168, 0.4029, 0.8523, 0.3417, 0.3588, 0.5000, 0.6125, 0.5400],
  29317. [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869],
  29318. [0.6053, 0.4035, 0.6897, 0.1916, 0.4000, 0.2117, 0.5440, 0.5168],
  29319. [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317],
  29320. [0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317]],
  29321. device='cuda:0', dtype=torch.float64)
  29322. predictions are: tensor([[ 0.7460, -0.2874, 1.7264, 0.2952, -0.3615, 0.0596, 0.1324, 0.1865],
  29323. [ 0.6967, -0.3833, 1.8552, 0.0737, -0.5081, -0.1024, 0.4197, 0.0161],
  29324. [-1.7460, -1.9138, 1.3348, -0.9356, -0.5931, -0.9858, 0.2740, 0.2554],
  29325. [ 0.8198, -0.2550, 1.6915, -0.6744, -0.5717, 0.0033, 0.7498, 0.2701],
  29326. [ 0.5167, -0.4165, 1.8108, -0.7367, -0.2093, -1.3022, 0.3991, 0.0950],
  29327. [ 0.5604, -0.4277, 0.9121, -1.2715, -0.3794, -1.3098, 0.3249, 0.1936],
  29328. [ 0.7529, -0.2951, 1.8136, 0.2724, -0.5670, -0.5399, 0.4457, 0.1687],
  29329. [ 0.4675, -0.4694, 1.3008, -1.0542, -0.3341, -1.1477, 0.4661, 0.2529]],
  29330. device='cuda:0', grad_fn=<AddmmBackward>)
  29331. landmarks are: tensor([[[ 0.5436, -0.4076, 1.7037, 0.2006, -0.3517, -0.0457, 0.0311,
  29332. 0.2048],
  29333. [ 0.5879, -0.4583, 1.7961, 0.0313, -0.4672, -0.1612, 0.4393,
  29334. 0.0313],
  29335. [-2.2859, -2.2859, 1.5074, -1.0388, -0.5423, -0.9849, 0.2199,
  29336. 0.2699],
  29337. [ 0.5631, -0.4249, 1.6509, -0.7078, -0.6289, 0.0236, 0.5432,
  29338. 0.2083],
  29339. [ 0.6119, -0.3927, 1.6979, -0.7925, -0.2536, -1.3698, 0.4282,
  29340. -0.0368],
  29341. [ 0.5102, -0.4223, 0.8999, -1.4011, -0.4383, -1.3082, 0.2267,
  29342. 0.1013],
  29343. [ 0.5901, -0.4157, 1.7557, 0.1929, -0.5423, -0.5923, 0.3584,
  29344. 0.1698],
  29345. [ 0.5849, -0.3836, 1.2649, -1.2236, -0.3460, -1.2313, 0.4508,
  29346. 0.1698]]], device='cuda:0')
  29347. loss_train_step before backward: tensor(0.0150, device='cuda:0', grad_fn=<MseLossBackward>)
  29348. loss_train_step after backward: tensor(0.0150, device='cuda:0', grad_fn=<MseLossBackward>)
  29349. loss_train: 1.156885642092675
  29350. step: 60
  29351. running loss: 0.01928142736821125
  29352.  
  29353. Train Steps: 60/90 Loss: 0.0193 torch.Size([8, 600, 800])
  29354. torch.Size([8, 8])
  29355. tensor([[0.6125, 0.3974, 0.7725, 0.2517, 0.3538, 0.3317, 0.5887, 0.5500],
  29356. [0.6147, 0.4112, 0.7988, 0.3200, 0.3775, 0.2767, 0.5150, 0.5550],
  29357. [0.6357, 0.4159, 0.8788, 0.5583, 0.3638, 0.4433, 0.6488, 0.5297],
  29358. [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209],
  29359. [0.6251, 0.4163, 0.8662, 0.4467, 0.3625, 0.3567, 0.6038, 0.5533],
  29360. [0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567],
  29361. [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367],
  29362. [0.6300, 0.4133, 0.8538, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413]],
  29363. device='cuda:0', dtype=torch.float64)
  29364. predictions are: tensor([[ 0.3795, -0.5493, 1.2779, -1.1011, -0.6506, -0.7793, 0.3970, 0.1421],
  29365. [ 0.4603, -0.4661, 1.3833, -0.7593, -0.6260, -0.9102, -0.0130, 0.1901],
  29366. [ 0.6507, -0.3994, 1.7494, 0.2472, -0.6366, -0.2860, 0.6668, 0.0906],
  29367. [ 0.6627, -0.3054, 1.7199, -0.0872, -0.2746, -0.9652, 0.2201, 0.4598],
  29368. [ 0.6057, -0.4007, 1.7335, -0.2720, -0.6664, -0.7356, 0.3688, 0.2675],
  29369. [ 0.3530, -0.5622, 1.7161, -0.2440, -0.1596, 0.1293, 0.3165, 0.1964],
  29370. [ 0.3366, -0.5876, 1.3808, -0.9403, -0.6362, -0.8249, 0.4522, 0.1936],
  29371. [ 0.4302, -0.5576, 1.7444, -1.1820, 0.1793, -1.0450, 1.0431, 0.2124]],
  29372. device='cuda:0', grad_fn=<AddmmBackward>)
  29373. landmarks are: tensor([[[ 0.5430, -0.4503, 1.2822, -1.1235, -0.6520, -0.7540, 0.4335,
  29374. 0.2545],
  29375. [ 0.5532, -0.3864, 1.4035, -0.8079, -0.5423, -1.0080, 0.0928,
  29376. 0.2776],
  29377. [ 0.6504, -0.3647, 1.7730, 0.2930, -0.6058, -0.2382, 0.7109,
  29378. 0.1608],
  29379. [ 0.6174, -0.3118, 1.6402, -0.2074, -0.1958, -1.0927, 0.2267,
  29380. 0.5822],
  29381. [ 0.6016, -0.3633, 1.7152, -0.2228, -0.6115, -0.6385, 0.5028,
  29382. 0.2699],
  29383. [ 0.5538, -0.4139, 1.7557, -0.1843, -0.0459, 0.1242, 0.4219,
  29384. 0.2853],
  29385. [ 0.5697, -0.4442, 1.3529, -0.9515, -0.5774, -0.7801, 0.5253,
  29386. 0.1931],
  29387. [ 0.6240, -0.3768, 1.6575, -1.2851, 0.2949, -1.2467, 1.1276,
  29388. 0.2142]]], device='cuda:0')
  29389. loss_train_step before backward: tensor(0.0086, device='cuda:0', grad_fn=<MseLossBackward>)
  29390. loss_train_step after backward: tensor(0.0086, device='cuda:0', grad_fn=<MseLossBackward>)
  29391. loss_train: 1.1654705270193517
  29392. step: 61
  29393. running loss: 0.019106074213431995
  29394. Train Steps: 61/90 Loss: 0.0191 torch.Size([8, 600, 800])
  29395. torch.Size([8, 8])
  29396. tensor([[ nan, nan, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563],
  29397. [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663],
  29398. [ nan, nan, 0.6793, 0.2110, 0.4012, 0.2167, 0.5112, 0.5583],
  29399. [0.6273, 0.4143, 0.8750, 0.5700, 0.3987, 0.4717, 0.6013, 0.5467],
  29400. [0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733],
  29401. [0.6300, 0.4133, 0.8538, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413],
  29402. [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100],
  29403. [0.6204, 0.4110, 0.7913, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367]],
  29404. device='cuda:0', dtype=torch.float64)
  29405. predictions are: tensor([[-2.1501, -2.1734, 1.3345, -0.8381, -0.5983, -0.9237, 0.1402, 0.2465],
  29406. [ 0.4663, -0.4732, 1.1043, -1.1138, -0.4635, -1.1531, 0.1945, 0.2410],
  29407. [-1.5817, -1.7912, 0.9501, -1.3078, -0.3975, -1.2940, 0.0970, 0.2340],
  29408. [ 0.9681, -0.1215, 1.7032, 0.3756, -0.5523, -0.0942, 0.4063, 0.1481],
  29409. [ 0.9483, -0.1109, 1.8560, 0.1420, -0.6529, -0.2599, 0.2922, 0.3571],
  29410. [ 0.8209, -0.2577, 1.7611, -1.0484, 0.1209, -0.9930, 1.0093, 0.2220],
  29411. [ 0.9207, -0.2333, 1.6936, 0.2770, -0.5535, -0.0371, 0.5384, 0.0420],
  29412. [ 0.7437, -0.2797, 1.4979, -0.9594, -0.3998, -1.1038, 0.4923, 0.1269]],
  29413. device='cuda:0', grad_fn=<AddmmBackward>)
  29414. landmarks are: tensor([[[-2.2859, -2.2859, 1.4006, -0.8105, -0.6115, -0.8232, 0.0419,
  29415. 0.2837],
  29416. [ 0.5269, -0.4176, 1.0628, -1.3159, -0.4037, -1.2236, 0.2432,
  29417. 0.3297],
  29418. [-2.2859, -2.2859, 0.8516, -1.3112, -0.4326, -1.2851, 0.0755,
  29419. 0.2930],
  29420. [ 0.6115, -0.3724, 1.7557, 0.3469, -0.4441, -0.1073, 0.4912,
  29421. 0.2391],
  29422. [ 0.6121, -0.3138, 1.8423, 0.0082, -0.6462, -0.3075, 0.3469,
  29423. 0.3623],
  29424. [ 0.6240, -0.3768, 1.6575, -1.2851, 0.2949, -1.2467, 1.1276,
  29425. 0.2142],
  29426. [ 0.6125, -0.4273, 1.6402, 0.2006, -0.4788, -0.0303, 0.6182,
  29427. 0.0697],
  29428. [ 0.5796, -0.3878, 1.3688, -1.0542, -0.4095, -1.1312, 0.5894,
  29429. 0.1929]]], device='cuda:0')
  29430. loss_train_step before backward: tensor(0.0272, device='cuda:0', grad_fn=<MseLossBackward>)
  29431. loss_train_step after backward: tensor(0.0272, device='cuda:0', grad_fn=<MseLossBackward>)
  29432. loss_train: 1.192680740263313
  29433. step: 62
  29434. running loss: 0.019236786133279244
  29435. Train Steps: 62/90 Loss: 0.0192 torch.Size([8, 600, 800])
  29436. torch.Size([8, 8])
  29437. tensor([[0.6289, 0.4024, 0.9088, 0.4567, 0.3937, 0.5633, 0.7058, 0.5609],
  29438. [0.6250, 0.4236, 0.8638, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799],
  29439. [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413],
  29440. [0.6261, 0.4045, 0.8865, 0.5369, 0.3895, 0.4859, 0.6683, 0.5249],
  29441. [0.6293, 0.4097, 0.8800, 0.2517, 0.5262, 0.2600, 0.7430, 0.5378],
  29442. [0.6339, 0.4123, 0.8638, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436],
  29443. [ nan, nan, 0.7648, 0.2722, 0.3962, 0.2183, 0.5060, 0.5422],
  29444. [0.6339, 0.4112, 0.8838, 0.5067, 0.4000, 0.5433, 0.7549, 0.5378]],
  29445. device='cuda:0', dtype=torch.float64)
  29446. predictions are: tensor([[ 0.6573, -0.3446, 1.6966, -0.2443, -0.5200, 0.1686, 0.5751, 0.2832],
  29447. [ 0.6415, -0.3198, 1.6393, -0.4346, -0.4965, -0.9133, 0.0875, 0.4190],
  29448. [ 0.4149, -0.5455, 1.8838, -0.2986, -0.3281, -0.9900, 0.6586, 0.1507],
  29449. [ 0.6257, -0.3637, 1.6404, 0.1075, -0.6028, -0.1368, 0.3381, 0.1363],
  29450. [ 0.5099, -0.4514, 1.6478, -1.0770, 0.0905, -1.2842, 0.7178, 0.1467],
  29451. [ 0.6812, -0.3321, 1.5995, 0.0378, -0.5077, 0.1190, 0.6594, 0.2596],
  29452. [-2.2269, -2.2209, 1.0822, -1.1340, -0.4991, -1.3651, -0.0889, 0.1406],
  29453. [ 0.5776, -0.4028, 1.6504, -0.0414, -0.5256, 0.1455, 0.5969, 0.2451]],
  29454. device='cuda:0', grad_fn=<AddmmBackward>)
  29455. landmarks are: tensor([[[ 0.6191, -0.4273, 1.9115, -0.1766, -0.4672, 0.3161, 0.9741,
  29456. 0.3050],
  29457. [ 0.6008, -0.3293, 1.7037, -0.5461, -0.4152, -0.8309, 0.3234,
  29458. 0.3928],
  29459. [ 0.6224, -0.4345, 1.9404, -0.2921, -0.3171, -0.8771, 1.0655,
  29460. 0.2142],
  29461. [ 0.6059, -0.4177, 1.8087, 0.1941, -0.4868, -0.0414, 0.8010,
  29462. 0.1385],
  29463. [ 0.6207, -0.3936, 1.7788, -1.1235, 0.1448, -1.0850, 1.1459,
  29464. 0.1982],
  29465. [ 0.6421, -0.3816, 1.7037, 0.1929, -0.4037, 0.2391, 1.1861,
  29466. 0.2249],
  29467. [-2.2859, -2.2859, 1.2469, -1.0288, -0.4557, -1.2774, 0.0511,
  29468. 0.2183],
  29469. [ 0.6421, -0.3864, 1.7961, 0.0543, -0.4383, 0.2237, 1.2007,
  29470. 0.1982]]], device='cuda:0')
  29471. loss_train_step before backward: tensor(0.0305, device='cuda:0', grad_fn=<MseLossBackward>)
  29472. loss_train_step after backward: tensor(0.0305, device='cuda:0', grad_fn=<MseLossBackward>)
  29473. loss_train: 1.2231627046130598
  29474. step: 63
  29475. running loss: 0.019415281025604123
  29476. Train Steps: 63/90 Loss: 0.0194 torch.Size([8, 600, 800])
  29477. torch.Size([8, 8])
  29478. tensor([[0.6126, 0.4039, 0.8237, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138],
  29479. [0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373],
  29480. [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575],
  29481. [ nan, nan, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563],
  29482. [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882],
  29483. [ nan, nan, 0.6793, 0.2110, 0.4012, 0.2167, 0.5112, 0.5583],
  29484. [0.6350, 0.4043, 0.8738, 0.5650, 0.3850, 0.4750, 0.6401, 0.4950],
  29485. [0.6241, 0.4143, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550]],
  29486. device='cuda:0', dtype=torch.float64)
  29487. predictions are: tensor([[ 0.6864, -0.2961, 1.5186, -0.5382, -0.5293, -0.7114, 0.4208, 0.5046],
  29488. [ 0.9863, -0.1553, 1.5130, 0.3829, -0.4618, -0.3696, 0.4101, 0.2275],
  29489. [ 0.8880, -0.2267, 1.6710, -0.8084, -0.3837, -0.4435, 1.0219, 0.2616],
  29490. [-2.2763, -2.2673, 1.3555, -0.8581, -0.5137, -0.9928, 0.1230, 0.2752],
  29491. [ 0.7329, -0.2929, 1.7610, -0.4323, -0.5493, -0.6349, 0.1989, 0.0026],
  29492. [-1.8182, -1.9563, 0.9659, -1.3406, -0.3299, -1.3894, 0.0819, 0.2440],
  29493. [ 0.8465, -0.2613, 1.7509, 0.2371, -0.4888, -0.1813, 0.6480, 0.0601],
  29494. [ 0.8349, -0.2106, 1.8088, -0.2125, -0.3730, 0.0472, 0.5546, 0.3162]],
  29495. device='cuda:0', grad_fn=<AddmmBackward>)
  29496. landmarks are: tensor([[[ 5.4376e-01, -4.2055e-01, 1.5189e+00, -4.5373e-01, -6.1155e-01,
  29497. -6.2309e-01, 4.3649e-01, 5.4914e-01],
  29498. [ 5.6801e-01, -4.5619e-01, 1.5697e+00, 4.9469e-01, -4.9038e-01,
  29499. -1.5026e-01, 3.5357e-01, 1.9563e-01],
  29500. [ 6.1577e-01, -4.2490e-01, 1.8654e+00, -9.0023e-01, -3.2286e-01,
  29501. -3.5366e-01, 9.6675e-01, 2.8902e-01],
  29502. [-2.2859e+00, -2.2859e+00, 1.4006e+00, -8.1049e-01, -6.1155e-01,
  29503. -8.2325e-01, 4.1889e-02, 2.8371e-01],
  29504. [ 5.5953e-01, -3.9877e-01, 1.7672e+00, -4.4604e-01, -5.5381e-01,
  29505. -5.3841e-01, 8.2802e-02, -3.0981e-02],
  29506. [-2.2859e+00, -2.2859e+00, 8.5162e-01, -1.3112e+00, -4.3256e-01,
  29507. -1.2851e+00, 7.5520e-02, 2.9299e-01],
  29508. [ 6.4706e-01, -4.1832e-01, 1.7499e+00, 3.2379e-01, -5.0762e-01,
  29509. -9.1917e-02, 6.7064e-01, 4.6189e-04],
  29510. [ 5.9677e-01, -3.7252e-01, 1.8423e+00, -1.3811e-01, -4.0370e-01,
  29511. 1.8522e-01, 6.0092e-01, 2.7760e-01]]], device='cuda:0')
  29512. loss_train_step before backward: tensor(0.0195, device='cuda:0', grad_fn=<MseLossBackward>)
  29513. loss_train_step after backward: tensor(0.0195, device='cuda:0', grad_fn=<MseLossBackward>)
  29514. loss_train: 1.2426480515860021
  29515. step: 64
  29516. running loss: 0.019416375806031283
  29517.  
  29518. Train Steps: 64/90 Loss: 0.0194 torch.Size([8, 600, 800])
  29519. torch.Size([8, 8])
  29520. tensor([[0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209],
  29521. [0.6357, 0.4118, 0.8400, 0.2500, 0.5413, 0.1633, 0.6725, 0.5586],
  29522. [0.6201, 0.3970, 0.8413, 0.4950, 0.4413, 0.5183, 0.6088, 0.5400],
  29523. [0.6197, 0.4091, 0.8800, 0.4783, 0.3538, 0.4767, 0.5950, 0.5550],
  29524. [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267],
  29525. [0.6339, 0.4102, 0.8588, 0.3133, 0.4425, 0.2117, 0.6417, 0.5089],
  29526. [0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400],
  29527. [0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350]],
  29528. device='cuda:0', dtype=torch.float64)
  29529. predictions are: tensor([[ 0.5682, -0.3837, 1.6764, -0.2138, -0.3883, -0.8902, 0.3701, 0.5147],
  29530. [ 0.2831, -0.6369, 1.5697, -1.2638, 0.0999, -1.3712, 0.9089, 0.2294],
  29531. [ 0.2783, -0.6375, 1.5955, -0.1216, -0.4374, 0.1521, 0.4283, 0.2186],
  29532. [ 0.4816, -0.4981, 1.7127, -0.2116, -0.7235, -0.0632, 0.5347, 0.3418],
  29533. [ 0.4290, -0.5700, 1.5273, 0.2132, -0.5799, -0.1178, 0.4299, 0.1792],
  29534. [ 0.5207, -0.4917, 1.6541, -0.9631, -0.3377, -1.1652, 0.7063, 0.0927],
  29535. [ 0.2865, -0.5891, 1.4471, -0.9866, -0.2524, -1.2922, 0.3699, 0.2180],
  29536. [ 0.1956, -0.6708, 1.3196, -1.2192, -0.3199, -1.2378, 0.4614, 0.2082]],
  29537. device='cuda:0', grad_fn=<AddmmBackward>)
  29538. landmarks are: tensor([[[ 6.1742e-01, -3.1175e-01, 1.6402e+00, -2.0739e-01, -1.9584e-01,
  29539. -1.0927e+00, 2.2674e-01, 5.8220e-01],
  29540. [ 6.5036e-01, -3.8397e-01, 1.5940e+00, -1.1312e+00, 2.1409e-01,
  29541. -1.5315e+00, 8.2052e-01, 2.9436e-01],
  29542. [ 5.7812e-01, -4.5219e-01, 1.5998e+00, 4.6189e-04, -2.4781e-01,
  29543. 1.0824e-01, 5.2587e-01, 2.0831e-01],
  29544. [ 5.7633e-01, -3.9630e-01, 1.7788e+00, -7.6520e-02, -6.5196e-01,
  29545. -8.4219e-02, 4.6236e-01, 2.7760e-01],
  29546. [ 5.8915e-01, -4.5504e-01, 1.5132e+00, 3.5458e-01, -3.6905e-01,
  29547. -1.5350e-01, 3.8152e-01, 1.4673e-01],
  29548. [ 6.4212e-01, -3.9120e-01, 1.6806e+00, -8.3865e-01, -2.4203e-01,
  29549. -1.3082e+00, 6.7795e-01, 6.4585e-02],
  29550. [ 5.9107e-01, -3.8879e-01, 1.4727e+00, -9.5412e-01, -9.1917e-02,
  29551. -1.4930e+00, 3.9885e-01, 2.0831e-01],
  29552. [ 5.7679e-01, -4.0308e-01, 1.3838e+00, -1.1527e+00, -2.1876e-01,
  29553. -1.4216e+00, 4.3790e-01, 1.8502e-01]]], device='cuda:0')
  29554. loss_train_step before backward: tensor(0.0199, device='cuda:0', grad_fn=<MseLossBackward>)
  29555. loss_train_step after backward: tensor(0.0199, device='cuda:0', grad_fn=<MseLossBackward>)
  29556. loss_train: 1.2625863901339471
  29557. step: 65
  29558. running loss: 0.019424406002060725
  29559. Train Steps: 65/90 Loss: 0.0194 torch.Size([8, 600, 800])
  29560. torch.Size([8, 8])
  29561. tensor([[0.6250, 0.3993, 0.9138, 0.4333, 0.3763, 0.5217, 0.6995, 0.5320],
  29562. [0.6267, 0.4080, 0.8438, 0.2633, 0.4763, 0.1800, 0.6259, 0.5240],
  29563. [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879],
  29564. [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150],
  29565. [0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494],
  29566. [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649],
  29567. [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366],
  29568. [ nan, nan, 0.6412, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700]],
  29569. device='cuda:0', dtype=torch.float64)
  29570. predictions are: tensor([[ 0.5621, -0.4321, 1.7724, -0.4772, -0.5423, -0.1147, 0.8050, 0.2694],
  29571. [ 0.5308, -0.4434, 1.6116, -1.1489, -0.1159, -1.5357, 0.5198, 0.1760],
  29572. [ 0.5708, -0.3927, 1.7551, -0.3971, -0.4772, 0.0762, 0.3674, 0.0970],
  29573. [ 0.3569, -0.5271, 1.7537, -0.2564, -0.2742, 0.1578, 0.5440, 0.2621],
  29574. [ 0.5001, -0.4712, 1.6459, 0.1250, -0.5059, -0.5100, 0.7173, 0.2668],
  29575. [ 0.5423, -0.4400, 1.6106, 0.2562, -0.5150, -0.2214, 0.4051, 0.1300],
  29576. [ 0.4982, -0.4699, 1.4092, 0.0876, -0.4595, -0.2477, 0.7810, 0.3380],
  29577. [-2.5799, -2.5101, 0.9946, -1.3201, -0.3054, -1.6528, 0.1075, 0.3432]],
  29578. device='cuda:0', grad_fn=<AddmmBackward>)
  29579. landmarks are: tensor([[[ 0.6010, -0.4417, 1.9346, -0.2844, -0.5480, 0.1236, 0.9448,
  29580. 0.1715],
  29581. [ 0.6088, -0.4015, 1.6113, -1.0696, -0.0861, -1.4545, 0.6051,
  29582. 0.1343],
  29583. [ 0.5676, -0.4327, 1.8252, -0.2095, -0.4886, 0.3084, 0.4727,
  29584. -0.0322],
  29585. [ 0.6014, -0.3883, 1.8192, -0.1150, -0.2882, 0.4008, 0.5201,
  29586. 0.0928],
  29587. [ 0.6224, -0.4032, 1.5940, 0.2930, -0.5769, -0.2690, 0.8900,
  29588. 0.2516],
  29589. [ 0.6105, -0.4360, 1.6171, 0.5162, -0.5076, 0.0159, 0.5171,
  29590. -0.1385],
  29591. [ 0.6289, -0.4345, 1.3794, 0.3679, -0.4845, 0.0390, 0.9265,
  29592. 0.1928],
  29593. [-2.2859, -2.2859, 0.6760, -1.4083, -0.3286, -1.4160, 0.2487,
  29594. 0.3469]]], device='cuda:0')
  29595. loss_train_step before backward: tensor(0.0219, device='cuda:0', grad_fn=<MseLossBackward>)
  29596. loss_train_step after backward: tensor(0.0219, device='cuda:0', grad_fn=<MseLossBackward>)
  29597. loss_train: 1.2844528066925704
  29598. step: 66
  29599. running loss: 0.019461406162008643
  29600. Train Steps: 66/90 Loss: 0.0195 torch.Size([8, 600, 800])
  29601. torch.Size([8, 8])
  29602. tensor([[0.6278, 0.4253, 0.8875, 0.5017, 0.4113, 0.2750, 0.5413, 0.6196],
  29603. [0.6179, 0.4040, 0.7412, 0.1850, 0.3825, 0.2783, 0.5837, 0.5600],
  29604. [0.6095, 0.3970, 0.8688, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940],
  29605. [0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240],
  29606. [0.6055, 0.4015, 0.7425, 0.2033, 0.4113, 0.1883, 0.5217, 0.4823],
  29607. [ nan, nan, 0.7225, 0.2167, 0.3987, 0.2283, 0.5427, 0.5181],
  29608. [0.6268, 0.4052, 0.8175, 0.2250, 0.4688, 0.1917, 0.6375, 0.5267],
  29609. [0.6212, 0.4033, 0.8938, 0.4167, 0.3813, 0.4267, 0.5613, 0.5583]],
  29610. device='cuda:0', dtype=torch.float64)
  29611. predictions are: tensor([[ 0.6355, -0.3365, 1.7837, 0.2189, -0.3321, -0.7855, 0.6329, 0.4986],
  29612. [ 0.6189, -0.3985, 1.1583, -1.1676, -0.5172, -0.9246, 0.6778, 0.2728],
  29613. [ 0.4296, -0.5356, 1.8003, -0.0223, -0.1546, 0.1135, 0.4732, 0.0954],
  29614. [ 0.2051, -0.6368, 1.1759, -1.1390, -0.3122, -1.3071, 0.3184, 0.2113],
  29615. [ 0.2515, -0.6269, 1.0781, -1.1108, -0.3976, -1.3259, 0.3161, 0.0738],
  29616. [-2.5323, -2.4686, 1.0653, -1.2045, -0.4495, -1.2806, 0.2355, 0.2090],
  29617. [ 0.6640, -0.3917, 1.4862, -1.1260, -0.1107, -1.3607, 0.8365, 0.1686],
  29618. [ 0.4217, -0.5376, 1.9231, -0.1483, -0.5415, -0.2629, 0.6584, 0.2839]],
  29619. device='cuda:0', grad_fn=<AddmmBackward>)
  29620. landmarks are: tensor([[[ 0.6139, -0.3216, 1.8134, 0.0313, -0.3864, -1.0157, 0.2144,
  29621. 0.5762],
  29622. [ 0.5680, -0.4201, 1.1379, -1.4314, -0.5192, -1.0003, 0.4104,
  29623. 0.3007],
  29624. [ 0.5292, -0.4520, 1.7268, -0.0842, -0.0413, -0.0324, 0.1116,
  29625. -0.0039],
  29626. [ 0.5676, -0.4112, 1.1898, -1.2467, -0.2940, -1.4622, 0.2103,
  29627. 0.1343],
  29628. [ 0.5109, -0.4316, 1.1436, -1.3467, -0.3864, -1.4160, 0.1239,
  29629. -0.0580],
  29630. [-2.2859, -2.2859, 1.0513, -1.2851, -0.4441, -1.2313, 0.2206,
  29631. 0.1073],
  29632. [ 0.6092, -0.4143, 1.4901, -1.2467, -0.1208, -1.4006, 0.6587,
  29633. 0.1467],
  29634. [ 0.5832, -0.4231, 1.8423, -0.3614, -0.5249, -0.3152, 0.3065,
  29635. 0.2930]]], device='cuda:0')
  29636. loss_train_step before backward: tensor(0.0230, device='cuda:0', grad_fn=<MseLossBackward>)
  29637. loss_train_step after backward: tensor(0.0230, device='cuda:0', grad_fn=<MseLossBackward>)
  29638. loss_train: 1.3074766951613128
  29639. step: 67
  29640. running loss: 0.019514577539721085
  29641. Train Steps: 67/90 Loss: 0.0195 torch.Size([8, 600, 800])
  29642. torch.Size([8, 8])
  29643. tensor([[0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683],
  29644. [ nan, nan, 0.8850, 0.3000, 0.5363, 0.2250, 0.7343, 0.5771],
  29645. [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5988, 0.5283],
  29646. [ nan, nan, 0.7268, 0.2333, 0.4125, 0.1933, 0.5112, 0.5383],
  29647. [0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933],
  29648. [0.6212, 0.4033, 0.8938, 0.4167, 0.3813, 0.4267, 0.5613, 0.5583],
  29649. [0.6083, 0.3957, 0.8638, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980],
  29650. [0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748]],
  29651. device='cuda:0', dtype=torch.float64)
  29652. predictions are: tensor([[ 0.7521, -0.2619, 1.6282, -0.0429, -0.5543, -0.5445, 0.4366, 0.3235],
  29653. [-2.1303, -2.1780, 1.7012, -1.1529, 0.1408, -1.2760, 0.8930, 0.4179],
  29654. [ 0.7358, -0.3000, 1.6929, 0.0419, -0.5112, -0.0552, 0.5920, 0.1726],
  29655. [-1.7605, -1.9247, 0.9663, -1.3456, -0.4302, -1.4783, 0.0266, 0.2209],
  29656. [ 0.7053, -0.2838, 1.5613, 0.1993, -0.0641, -0.2609, 0.4373, 0.3980],
  29657. [ 0.5957, -0.3948, 1.7999, -0.4406, -0.5199, -0.3977, 0.4966, 0.2283],
  29658. [ 0.6119, -0.3963, 1.6988, -0.1708, -0.3118, 0.0706, 0.3834, -0.0591],
  29659. [ 0.6916, -0.3560, 1.5536, 0.0555, -0.4908, -0.3907, 1.0764, 0.2408]],
  29660. device='cuda:0', grad_fn=<AddmmBackward>)
  29661. landmarks are: tensor([[[ 5.7771e-01, -3.9153e-01, 1.7961e+00, 1.6982e-01, -5.1917e-01,
  29662. -5.3072e-01, 2.1409e-01, 3.3918e-01],
  29663. [-2.2859e+00, -2.2859e+00, 1.8018e+00, -9.0023e-01, 1.9099e-01,
  29664. -1.2467e+00, 1.1057e+00, 3.7986e-01],
  29665. [ 5.6028e-01, -4.3195e-01, 1.7788e+00, 1.7752e-01, -5.5381e-01,
  29666. -6.1124e-02, 4.7968e-01, 1.5443e-01],
  29667. [-2.2859e+00, -2.2859e+00, 1.0712e+00, -1.2085e+00, -3.8060e-01,
  29668. -1.3929e+00, 7.5520e-02, 2.0062e-01],
  29669. [ 5.9590e-01, -3.5789e-01, 1.6055e+00, 3.6228e-01, -5.7275e-02,
  29670. -2.0739e-01, 3.1224e-01, 4.5466e-01],
  29671. [ 5.8320e-01, -4.2309e-01, 1.8423e+00, -3.6135e-01, -5.2494e-01,
  29672. -3.1517e-01, 3.0647e-01, 2.9299e-01],
  29673. [ 5.2379e-01, -4.5797e-01, 1.7037e+00, 4.6189e-04, -2.7090e-01,
  29674. 6.2048e-02, 1.8356e-01, 1.4106e-02],
  29675. [ 6.4212e-01, -3.6471e-01, 1.5940e+00, 3.0839e-01, -5.1917e-01,
  29676. -3.6905e-01, 1.1057e+00, 3.6917e-01]]], device='cuda:0')
  29677. loss_train_step before backward: tensor(0.0182, device='cuda:0', grad_fn=<MseLossBackward>)
  29678. loss_train_step after backward: tensor(0.0182, device='cuda:0', grad_fn=<MseLossBackward>)
  29679. loss_train: 1.325725748669356
  29680. step: 68
  29681. running loss: 0.019495966892196414
  29682.  
  29683. Train Steps: 68/90 Loss: 0.0195 torch.Size([8, 600, 800])
  29684. torch.Size([8, 8])
  29685. tensor([[0.6336, 0.4086, 0.8900, 0.3950, 0.3900, 0.2950, 0.6504, 0.5066],
  29686. [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350],
  29687. [0.6263, 0.4233, 0.7924, 0.4626, 0.3788, 0.2883, 0.5573, 0.6047],
  29688. [0.6229, 0.4198, 0.7662, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783],
  29689. [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083],
  29690. [0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683],
  29691. [0.6199, 0.4102, 0.8950, 0.4417, 0.4012, 0.5367, 0.6112, 0.5967],
  29692. [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833]],
  29693. device='cuda:0', dtype=torch.float64)
  29694. predictions are: tensor([[ 0.5283, -0.5369, 1.9181, -0.6369, -0.4015, -0.9028, 0.8409, -0.0470],
  29695. [ 0.2845, -0.6451, 1.9472, -0.3014, -0.4665, -0.3839, 0.4376, 0.1284],
  29696. [ 0.5186, -0.4488, 1.4861, -0.1716, -0.4262, -0.8543, 0.4471, 0.4673],
  29697. [ 0.3839, -0.5428, 1.3026, -1.1962, -0.0284, -1.3630, 0.4675, 0.3162],
  29698. [ 0.1748, -0.7349, 1.7817, -0.0084, -0.4737, -0.1699, 0.6443, 0.3722],
  29699. [ 0.4493, -0.5394, 1.7518, 0.1026, -0.4578, -0.4625, 0.4444, 0.3099],
  29700. [ 0.4751, -0.5104, 1.9062, -0.2140, -0.2256, 0.3614, 0.8055, 0.2947],
  29701. [ 0.3398, -0.6077, 0.9313, -1.0825, -0.4828, -0.9722, 0.4175, 0.3324]],
  29702. device='cuda:0', grad_fn=<AddmmBackward>)
  29703. landmarks are: tensor([[[ 0.6405, -0.3984, 1.8249, -0.4614, -0.4845, -0.9233, 0.7182,
  29704. 0.0539],
  29705. [ 0.6072, -0.3239, 1.8423, -0.3537, -0.4961, -0.3921, 0.2083,
  29706. 0.1852],
  29707. [ 0.6069, -0.3309, 1.3742, -0.1493, -0.5365, -0.9541, 0.2884,
  29708. 0.5071],
  29709. [ 0.5912, -0.3467, 1.2533, -1.0388, -0.1150, -1.3005, 0.3353,
  29710. 0.3854],
  29711. [ 0.5713, -0.4105, 1.7557, 0.0467, -0.6520, -0.2690, 0.3988,
  29712. 0.5239],
  29713. [ 0.5777, -0.3915, 1.7961, 0.1698, -0.5192, -0.5307, 0.2141,
  29714. 0.3392],
  29715. [ 0.5772, -0.3913, 1.8480, -0.2459, -0.4326, 0.1929, 0.5374,
  29716. 0.4701],
  29717. [ 0.5614, -0.4080, 0.7741, -0.8848, -0.5423, -0.9156, 0.3584,
  29718. 0.4085]]], device='cuda:0')
  29719. loss_train_step before backward: tensor(0.0231, device='cuda:0', grad_fn=<MseLossBackward>)
  29720. loss_train_step after backward: tensor(0.0231, device='cuda:0', grad_fn=<MseLossBackward>)
  29721. loss_train: 1.3488060417585075
  29722. step: 69
  29723. running loss: 0.01954791364867402
  29724. Train Steps: 69/90 Loss: 0.0195 torch.Size([8, 600, 800])
  29725. torch.Size([8, 8])
  29726. tensor([[0.6104, 0.4029, 0.8738, 0.4900, 0.4088, 0.4533, 0.5070, 0.5510],
  29727. [ nan, nan, 0.8488, 0.2300, 0.5563, 0.2100, 0.7390, 0.5679],
  29728. [0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742],
  29729. [0.6030, 0.3969, 0.7988, 0.3917, 0.3450, 0.3667, 0.5266, 0.4700],
  29730. [ nan, nan, 0.7625, 0.2433, 0.3713, 0.2867, 0.5235, 0.5220],
  29731. [0.6164, 0.4102, 0.8850, 0.4867, 0.4213, 0.5983, 0.5609, 0.5038],
  29732. [0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183],
  29733. [ nan, nan, 0.8463, 0.2550, 0.5850, 0.2133, 0.7129, 0.6072]],
  29734. device='cuda:0', dtype=torch.float64)
  29735. predictions are: tensor([[ 0.7772, -0.2527, 1.6278, 0.0698, -0.3793, -0.1506, 0.2176, 0.2853],
  29736. [-1.7345, -1.9180, 1.5621, -1.2114, 0.2351, -1.2765, 1.1004, 0.4288],
  29737. [ 0.8021, -0.2626, 1.6834, 0.2813, -0.5456, -0.5102, 0.6018, 0.0368],
  29738. [ 0.8277, -0.2241, 1.4681, -0.4764, -0.6223, -0.5843, 0.2276, 0.0560],
  29739. [-1.8877, -2.0528, 1.1652, -1.0918, -0.5150, -1.0408, 0.1515, 0.2189],
  29740. [ 0.8518, -0.1916, 1.7081, -0.0081, -0.2950, 0.3841, 0.5503, 0.2017],
  29741. [ 0.8578, -0.2372, 1.6706, 0.0083, -0.5515, -0.4213, 0.4641, 0.1731],
  29742. [-0.7118, -1.2511, 1.4998, -1.1039, 0.2474, -1.1877, 1.1366, 0.5582]],
  29743. device='cuda:0', grad_fn=<AddmmBackward>)
  29744. landmarks are: tensor([[[ 0.5334, -0.4252, 1.7499, -0.0226, -0.3979, -0.1920, 0.0558,
  29745. 0.2589],
  29746. [-2.2859, -2.2859, 1.6344, -1.2236, 0.2834, -1.3159, 1.1276,
  29747. 0.3371],
  29748. [ 0.6113, -0.4057, 1.7499, 0.3007, -0.5885, -0.5384, 0.4513,
  29749. -0.0957],
  29750. [ 0.4992, -0.4525, 1.4035, -0.4768, -0.6924, -0.5923, 0.1465,
  29751. -0.1151],
  29752. [-2.2859, -2.2859, 1.2360, -1.1620, -0.5711, -0.9618, 0.1322,
  29753. 0.1253],
  29754. [ 0.5614, -0.3912, 1.8018, -0.0380, -0.3402, 0.4778, 0.3049,
  29755. 0.0412],
  29756. [ 0.5944, -0.4543, 1.8018, 0.0082, -0.6058, -0.4306, 0.4162,
  29757. 0.1082],
  29758. [-2.2859, -2.2859, 1.6229, -1.1081, 0.4162, -1.3005, 1.0070,
  29759. 0.5188]]], device='cuda:0')
  29760. loss_train_step before backward: tensor(0.0804, device='cuda:0', grad_fn=<MseLossBackward>)
  29761. loss_train_step after backward: tensor(0.0804, device='cuda:0', grad_fn=<MseLossBackward>)
  29762. loss_train: 1.429237652104348
  29763. step: 70
  29764. running loss: 0.020417680744347827
  29765. Train Steps: 70/90 Loss: 0.0204 torch.Size([8, 600, 800])
  29766. torch.Size([8, 8])
  29767. tensor([[0.6197, 0.4050, 0.7527, 0.2000, 0.4042, 0.2249, 0.5895, 0.4995],
  29768. [0.6147, 0.4081, 0.8538, 0.3400, 0.3663, 0.3150, 0.5142, 0.4875],
  29769. [0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162],
  29770. [ nan, nan, 0.8850, 0.3000, 0.5363, 0.2250, 0.7343, 0.5771],
  29771. [0.6153, 0.4117, 0.8688, 0.5167, 0.4895, 0.5647, 0.5524, 0.5136],
  29772. [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050],
  29773. [0.6231, 0.3973, 0.8650, 0.3950, 0.3625, 0.3183, 0.5837, 0.5167],
  29774. [0.6296, 0.4045, 0.9138, 0.4100, 0.4232, 0.4242, 0.7422, 0.5297]],
  29775. device='cuda:0', dtype=torch.float64)
  29776. predictions are: tensor([[ 0.4453, -0.4955, 1.0256, -1.3659, -0.3497, -1.3072, 0.4431, 0.1319],
  29777. [ 0.5297, -0.4232, 1.5883, -0.5915, -0.5403, -0.8735, 0.1991, 0.1346],
  29778. [ 0.4946, -0.4857, 1.7351, -0.2698, -0.5069, -0.1672, 0.6268, 0.2527],
  29779. [-2.8201, -2.6383, 1.6262, -0.9551, 0.1732, -1.2972, 0.9966, 0.4400],
  29780. [ 0.3718, -0.5021, 1.6190, 0.1805, -0.0066, 0.1725, 0.3168, 0.2959],
  29781. [ 0.3578, -0.5493, 1.6804, -0.0961, -0.4836, -0.3701, 0.3286, 0.1494],
  29782. [ 0.4327, -0.5074, 1.5830, -0.3938, -0.5189, -0.8052, 0.3056, 0.2072],
  29783. [ 0.6953, -0.3807, 1.7009, -0.3156, -0.4463, -0.3204, 0.8906, 0.2495]],
  29784. device='cuda:0', grad_fn=<AddmmBackward>)
  29785. landmarks are: tensor([[[ 0.5762, -0.4153, 1.1908, -1.3622, -0.4190, -1.2471, 0.4368,
  29786. 0.0213],
  29787. [ 0.5532, -0.4008, 1.6575, -0.7155, -0.5942, -0.8309, 0.0890,
  29788. -0.0340],
  29789. [ 0.6059, -0.4442, 1.8921, -0.3422, -0.6230, -0.1461, 0.7489,
  29790. 0.0985],
  29791. [-2.2859, -2.2859, 1.8018, -0.9002, 0.1910, -1.2467, 1.1057,
  29792. 0.3799],
  29793. [ 0.5564, -0.3842, 1.7268, 0.1005, -0.0250, 0.3225, 0.2658,
  29794. 0.0862],
  29795. [ 0.5844, -0.4466, 1.8423, -0.1997, -0.5942, -0.3998, 0.4219,
  29796. 0.0467],
  29797. [ 0.5924, -0.4507, 1.7095, -0.4614, -0.6115, -0.8156, 0.4104,
  29798. 0.1005],
  29799. [ 0.6224, -0.4177, 1.9346, -0.3921, -0.3314, -0.3264, 1.1422,
  29800. 0.1608]]], device='cuda:0')
  29801. loss_train_step before backward: tensor(0.0187, device='cuda:0', grad_fn=<MseLossBackward>)
  29802. loss_train_step after backward: tensor(0.0187, device='cuda:0', grad_fn=<MseLossBackward>)
  29803. loss_train: 1.4479092578403652
  29804. step: 71
  29805. running loss: 0.020393088138596693
  29806. Train Steps: 71/90 Loss: 0.0204 torch.Size([8, 600, 800])
  29807. torch.Size([8, 8])
  29808. tensor([[0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461],
  29809. [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833],
  29810. [0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283],
  29811. [0.6160, 0.4093, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808],
  29812. [0.6182, 0.3930, 0.8841, 0.3892, 0.3556, 0.4967, 0.6222, 0.5279],
  29813. [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367],
  29814. [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960],
  29815. [0.6257, 0.4167, 0.8775, 0.3433, 0.3563, 0.4133, 0.6200, 0.5667]],
  29816. device='cuda:0', dtype=torch.float64)
  29817. predictions are: tensor([[ 0.4783, -0.4974, 1.6404, -0.4530, -0.5869, -0.5150, 0.2844, 0.1453],
  29818. [ 0.3782, -0.5538, 0.8784, -0.8781, -0.4522, -1.1484, 0.3703, 0.4597],
  29819. [ 0.4230, -0.5561, 2.0172, -0.3372, -0.3733, -0.6211, 0.8404, 0.1709],
  29820. [ 0.3973, -0.5338, 1.6760, -0.0551, -0.4016, -0.2625, 0.2399, 0.4082],
  29821. [ 0.4056, -0.5981, 1.8446, -0.3088, -0.4308, -0.1141, 0.7484, 0.1713],
  29822. [ 0.4678, -0.5262, 1.4584, -0.9175, -0.4248, -1.0008, 0.6792, 0.2188],
  29823. [ 0.0892, -0.7973, 1.9030, 0.0130, -0.0089, -0.0517, 0.2399, 0.0862],
  29824. [ 0.2341, -0.6479, 1.8795, -0.4693, -0.4716, -0.5444, 0.6646, 0.3986]],
  29825. device='cuda:0', grad_fn=<AddmmBackward>)
  29826. landmarks are: tensor([[[ 0.5124, -0.4483, 1.5420, -0.6231, -0.7155, -0.4152, 0.2679,
  29827. 0.2365],
  29828. [ 0.5614, -0.4080, 0.7741, -0.8848, -0.5423, -0.9156, 0.3584,
  29829. 0.4085],
  29830. [ 0.6073, -0.4058, 1.8885, -0.4999, -0.5942, -0.4768, 0.6413,
  29831. 0.1544],
  29832. [ 0.5592, -0.3956, 1.5543, -0.2456, -0.5885, -0.1689, 0.1392,
  29833. 0.3968],
  29834. [ 0.5697, -0.4706, 1.7976, -0.4884, -0.6433, 0.0081, 0.5878,
  29835. 0.1525],
  29836. [ 0.5697, -0.4442, 1.3529, -0.9515, -0.5774, -0.7801, 0.5253,
  29837. 0.1931],
  29838. [ 0.5205, -0.4577, 1.7326, -0.1304, -0.1862, 0.0447, 0.1301,
  29839. 0.0051],
  29840. [ 0.6040, -0.3614, 1.7672, -0.7001, -0.6404, -0.3768, 0.5778,
  29841. 0.3315]]], device='cuda:0')
  29842. loss_train_step before backward: tensor(0.0253, device='cuda:0', grad_fn=<MseLossBackward>)
  29843. loss_train_step after backward: tensor(0.0253, device='cuda:0', grad_fn=<MseLossBackward>)
  29844. loss_train: 1.4732359643094242
  29845. step: 72
  29846. running loss: 0.02046161061540867
  29847.  
  29848. Train Steps: 72/90 Loss: 0.0205 torch.Size([8, 600, 800])
  29849. torch.Size([8, 8])
  29850. tensor([[0.6200, 0.4118, 0.8287, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799],
  29851. [0.6091, 0.3997, 0.8314, 0.4334, 0.3788, 0.4550, 0.5213, 0.5656],
  29852. [0.6289, 0.4019, 0.8113, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332],
  29853. [0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869],
  29854. [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268],
  29855. [0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804],
  29856. [0.6114, 0.4018, 0.7213, 0.1967, 0.3763, 0.2700, 0.5875, 0.5533],
  29857. [0.6078, 0.4033, 0.8019, 0.3055, 0.3450, 0.4200, 0.6025, 0.5550]],
  29858. device='cuda:0', dtype=torch.float64)
  29859. predictions are: tensor([[ 0.2125, -0.6521, 1.6035, -0.3375, -0.4940, -1.0512, 0.1700, 0.4392],
  29860. [ 0.4755, -0.4858, 1.6933, -0.1396, -0.4665, -0.1380, 0.0999, 0.3121],
  29861. [ 0.2651, -0.6861, 1.6333, 0.4201, -0.4264, 0.0603, 0.8666, 0.2789],
  29862. [ 0.2776, -0.6162, 1.7169, -1.3630, 0.0578, -1.5248, 0.7481, 0.0741],
  29863. [ 0.3534, -0.6021, 1.8435, 0.1591, -0.1584, 0.1252, 0.5831, 0.2369],
  29864. [ 0.5017, -0.5179, 1.9070, -0.4184, -0.5838, -0.5517, 0.5219, -0.0068],
  29865. [ 0.4388, -0.5183, 1.0337, -1.3808, -0.4962, -1.1296, 0.3308, 0.2330],
  29866. [ 0.4061, -0.5370, 1.5350, -0.8161, -0.6444, -0.3295, 0.4475, 0.3008]],
  29867. device='cuda:0', grad_fn=<AddmmBackward>)
  29868. landmarks are: tensor([[[ 0.5779, -0.3840, 1.5420, -0.4306, -0.5423, -0.9772, 0.2041,
  29869. 0.3928],
  29870. [ 0.5275, -0.4396, 1.5543, -0.2841, -0.5365, -0.1843, 0.1221,
  29871. 0.3265],
  29872. [ 0.6191, -0.4297, 1.4612, 0.2391, -0.4961, 0.0313, 1.1166,
  29873. 0.1768],
  29874. [ 0.6127, -0.3944, 1.5189, -1.2467, -0.1323, -1.4622, 0.5646,
  29875. -0.0369],
  29876. [ 0.5828, -0.4682, 1.7031, -0.0497, -0.2458, 0.0818, 0.6381,
  29877. 0.1474],
  29878. [ 0.6071, -0.4119, 1.7788, -0.5153, -0.6000, -0.5692, 0.6586,
  29879. -0.0670],
  29880. [ 0.5380, -0.4302, 1.0455, -1.3775, -0.5480, -1.0388, 0.4277,
  29881. 0.2699],
  29882. [ 0.5215, -0.4232, 1.4182, -0.8747, -0.6924, -0.3460, 0.4970,
  29883. 0.2776]]], device='cuda:0')
  29884. loss_train_step before backward: tensor(0.0192, device='cuda:0', grad_fn=<MseLossBackward>)
  29885. loss_train_step after backward: tensor(0.0192, device='cuda:0', grad_fn=<MseLossBackward>)
  29886. loss_train: 1.4924231390468776
  29887. step: 73
  29888. running loss: 0.020444152589683254
  29889. Train Steps: 73/90 Loss: 0.0204 torch.Size([8, 600, 800])
  29890. torch.Size([8, 8])
  29891. tensor([[0.6171, 0.4127, 0.8900, 0.4800, 0.4325, 0.5783, 0.5769, 0.5090],
  29892. [ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600],
  29893. [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483],
  29894. [0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433],
  29895. [0.6108, 0.4011, 0.8037, 0.3400, 0.3700, 0.2933, 0.5658, 0.5617],
  29896. [0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400],
  29897. [0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6188, 0.5433],
  29898. [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817]],
  29899. device='cuda:0', dtype=torch.float64)
  29900. predictions are: tensor([[ 0.4990, -0.3994, 1.8878, 0.1543, -0.1967, 0.2972, 0.2611, 0.1317],
  29901. [-2.5539, -2.4791, 1.3508, -0.8933, -0.3324, -1.1283, 0.5128, 0.2438],
  29902. [ 0.5593, -0.4488, 1.0732, -1.2685, -0.3754, -1.2378, 0.5518, 0.1295],
  29903. [ 0.4449, -0.4705, 1.3463, -0.9438, -0.6151, -0.6301, 0.4881, 0.1498],
  29904. [ 0.4011, -0.5055, 1.4918, -0.5525, -0.5540, -0.9145, 0.3074, 0.2244],
  29905. [ 0.5259, -0.4571, 1.8694, -0.2454, -0.5995, -0.1364, 0.5221, 0.1358],
  29906. [ 0.4467, -0.4976, 1.9176, 0.1158, -0.4520, 0.0610, 0.3982, 0.1398],
  29907. [ 0.4622, -0.4492, 1.5167, -0.7024, -0.1878, -1.1792, 0.4206, 0.3888]],
  29908. device='cuda:0', grad_fn=<AddmmBackward>)
  29909. landmarks are: tensor([[[ 0.5645, -0.3797, 1.8249, -0.0688, -0.2882, 0.3854, 0.3789,
  29910. 0.0652],
  29911. [-2.2859, -2.2859, 1.4266, -1.1389, -0.4499, -1.1235, 0.2891,
  29912. 0.3007],
  29913. [ 0.5784, -0.4085, 1.0859, -1.3929, -0.4037, -1.1158, 0.5605,
  29914. 0.2468],
  29915. [ 0.5665, -0.3963, 1.3457, -0.9926, -0.6058, -0.6308, 0.5721,
  29916. 0.2237],
  29917. [ 0.5355, -0.4335, 1.4266, -0.7155, -0.5769, -0.9310, 0.3275,
  29918. 0.3087],
  29919. [ 0.5778, -0.4393, 1.8018, -0.4614, -0.6693, -0.1381, 0.5490,
  29920. 0.2083],
  29921. [ 0.5773, -0.4105, 1.8192, -0.1304, -0.5307, 0.0467, 0.5721,
  29922. 0.2237],
  29923. [ 0.5964, -0.3380, 1.4785, -0.8386, -0.2420, -1.0619, 0.3238,
  29924. 0.4008]]], device='cuda:0')
  29925. loss_train_step before backward: tensor(0.0119, device='cuda:0', grad_fn=<MseLossBackward>)
  29926. loss_train_step after backward: tensor(0.0119, device='cuda:0', grad_fn=<MseLossBackward>)
  29927. loss_train: 1.5043019359000027
  29928. step: 74
  29929. running loss: 0.020328404539189226
  29930. Train Steps: 74/90 Loss: 0.0203 torch.Size([8, 600, 800])
  29931. torch.Size([8, 8])
  29932. tensor([[0.6250, 0.4236, 0.8638, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799],
  29933. [0.6113, 0.4006, 0.8700, 0.5350, 0.3638, 0.3767, 0.5097, 0.4882],
  29934. [0.6200, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183],
  29935. [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374],
  29936. [0.6165, 0.4106, 0.7575, 0.1733, 0.3838, 0.2650, 0.5680, 0.5116],
  29937. [0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166],
  29938. [0.6273, 0.4100, 0.7137, 0.2133, 0.4000, 0.2650, 0.6075, 0.5633],
  29939. [0.6083, 0.3957, 0.8638, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980]],
  29940. device='cuda:0', dtype=torch.float64)
  29941. predictions are: tensor([[ 0.7385, -0.2766, 1.7929, -0.4707, -0.3836, -0.8503, 0.2929, 0.5421],
  29942. [ 0.4394, -0.5178, 1.7771, 0.1945, -0.5874, -0.4860, 0.1179, 0.1616],
  29943. [ 0.5185, -0.4564, 1.5172, -0.9465, -0.5366, -0.9257, 0.4488, 0.1289],
  29944. [ 0.4509, -0.5212, 1.7735, -0.5872, -0.6069, -0.7674, 0.4284, 0.1717],
  29945. [ 0.3056, -0.6098, 1.1797, -1.4743, -0.5555, -1.1407, 0.5208, 0.0257],
  29946. [ 0.5038, -0.4859, 1.7634, 0.1764, -0.2746, 0.2542, 0.7511, 0.2093],
  29947. [-0.1378, -0.8732, 0.9934, -1.3726, -0.4818, -1.0837, 0.5299, 0.3079],
  29948. [ 0.3289, -0.6140, 1.8338, 0.0409, -0.2576, 0.1750, 0.2668, 0.0221]],
  29949. device='cuda:0', grad_fn=<AddmmBackward>)
  29950. landmarks are: tensor([[[ 6.0075e-01, -3.2925e-01, 1.7037e+00, -5.4611e-01, -4.1524e-01,
  29951. -8.3095e-01, 3.2339e-01, 3.9283e-01],
  29952. [ 5.3788e-01, -4.3580e-01, 1.7326e+00, 1.8522e-01, -6.0577e-01,
  29953. -5.4611e-01, 6.8408e-02, -3.0981e-02],
  29954. [ 5.7806e-01, -4.1286e-01, 1.4142e+00, -9.0574e-01, -5.1146e-01,
  29955. -9.9373e-01, 4.6205e-01, 1.0799e-01],
  29956. [ 5.7771e-01, -4.4157e-01, 1.7044e+00, -5.8275e-01, -5.9618e-01,
  29957. -8.3610e-01, 4.8621e-01, 1.9626e-01],
  29958. [ 5.6184e-01, -3.8945e-01, 1.2129e+00, -1.4853e+00, -5.1339e-01,
  29959. -1.0619e+00, 3.3778e-01, 7.7228e-02],
  29960. [ 6.0425e-01, -4.2731e-01, 1.6920e+00, 1.8595e-01, -2.7171e-01,
  29961. 1.4059e-01, 7.9965e-01, 1.0043e-01],
  29962. [ 6.1155e-01, -3.9238e-01, 1.0109e+00, -1.3005e+00, -4.3834e-01,
  29963. -1.0619e+00, 5.2009e-01, 3.1609e-01],
  29964. [ 5.2379e-01, -4.5797e-01, 1.7037e+00, 4.6189e-04, -2.7090e-01,
  29965. 6.2048e-02, 1.8356e-01, 1.4106e-02]]], device='cuda:0')
  29966. loss_train_step before backward: tensor(0.0201, device='cuda:0', grad_fn=<MseLossBackward>)
  29967. loss_train_step after backward: tensor(0.0201, device='cuda:0', grad_fn=<MseLossBackward>)
  29968. loss_train: 1.5244338628835976
  29969. step: 75
  29970. running loss: 0.020325784838447967
  29971. Train Steps: 75/90 Loss: 0.0203 torch.Size([8, 600, 800])
  29972. torch.Size([8, 8])
  29973. tensor([[ nan, nan, 0.8525, 0.2217, 0.5413, 0.2367, 0.7367, 0.5482],
  29974. [0.6346, 0.4092, 0.7712, 0.5917, 0.4037, 0.4767, 0.7343, 0.5725],
  29975. [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5363, 0.5550],
  29976. [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879],
  29977. [0.6097, 0.3988, 0.8650, 0.5250, 0.4213, 0.5200, 0.5675, 0.5050],
  29978. [0.6132, 0.4118, 0.8200, 0.3633, 0.3563, 0.5400, 0.5787, 0.5136],
  29979. [ nan, nan, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633],
  29980. [0.6175, 0.3957, 0.8700, 0.4817, 0.4662, 0.5133, 0.5800, 0.5517]],
  29981. device='cuda:0', dtype=torch.float64)
  29982. predictions are: tensor([[-2.2324, -2.2225, 1.6292, -1.1904, 0.0917, -1.4040, 1.1599, 0.2829],
  29983. [ 0.7922, -0.2702, 1.3739, 0.1742, -0.5550, -0.2331, 0.7177, 0.3121],
  29984. [ 0.7121, -0.2939, 0.9677, -1.1365, -0.5669, -1.3134, 0.0549, 0.2373],
  29985. [ 0.8527, -0.2089, 1.8690, -0.2207, -0.6001, 0.2241, 0.3038, -0.0493],
  29986. [ 0.7100, -0.3023, 1.7268, 0.1583, -0.4271, 0.0640, 0.1201, 0.0194],
  29987. [ 0.7107, -0.2707, 1.5645, -0.6021, -0.6825, 0.0030, 0.3742, 0.1236],
  29988. [-2.0439, -2.0895, 1.1987, -1.0126, -0.5981, -1.1250, 0.1630, 0.2282],
  29989. [ 0.7303, -0.3029, 1.7606, -0.0840, -0.2595, -0.0316, 0.2301, 0.1367]],
  29990. device='cuda:0', grad_fn=<AddmmBackward>)
  29991. landmarks are: tensor([[[-2.2859, -2.2859, 1.6517, -1.2620, 0.2141, -1.1928, 1.1166,
  29992. 0.2463],
  29993. [ 0.6454, -0.3960, 1.2764, 0.4470, -0.4210, -0.0842, 1.1057,
  29994. 0.3585],
  29995. [ 0.5448, -0.3859, 0.9242, -1.1466, -0.4152, -1.3005, 0.1910,
  29996. 0.2776],
  29997. [ 0.5676, -0.4327, 1.8252, -0.2095, -0.4886, 0.3084, 0.4727,
  29998. -0.0322],
  29999. [ 0.5303, -0.4440, 1.7095, 0.1390, -0.3402, 0.1159, 0.3353,
  30000. 0.0467],
  30001. [ 0.5466, -0.3840, 1.5016, -0.6077, -0.6404, 0.2083, 0.3871,
  30002. 0.0862],
  30003. [-2.2859, -2.2859, 1.2030, -1.0288, -0.4961, -1.1081, 0.0813,
  30004. 0.3161],
  30005. [ 0.5662, -0.4581, 1.7326, -0.0611, -0.1323, 0.0851, 0.3931,
  30006. 0.2622]]], device='cuda:0')
  30007. loss_train_step before backward: tensor(0.0172, device='cuda:0', grad_fn=<MseLossBackward>)
  30008. loss_train_step after backward: tensor(0.0172, device='cuda:0', grad_fn=<MseLossBackward>)
  30009. loss_train: 1.5416146223433316
  30010. step: 76
  30011. running loss: 0.020284402925570152
  30012.  
  30013. Train Steps: 76/90 Loss: 0.0203 torch.Size([8, 600, 800])
  30014. torch.Size([8, 8])
  30015. tensor([[0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467],
  30016. [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390],
  30017. [0.6101, 0.3977, 0.8550, 0.5667, 0.3862, 0.4417, 0.5128, 0.4869],
  30018. [0.6286, 0.4055, 0.9000, 0.4717, 0.3763, 0.4683, 0.7018, 0.5494],
  30019. [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5463, 0.5800],
  30020. [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5837, 0.5500],
  30021. [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617],
  30022. [0.6206, 0.4001, 0.8900, 0.3933, 0.3588, 0.3567, 0.5837, 0.5083]],
  30023. device='cuda:0', dtype=torch.float64)
  30024. predictions are: tensor([[-0.0093, -0.8159, 1.5997, -0.9410, -0.6810, -0.3697, 0.5364, 0.1780],
  30025. [ 0.4972, -0.5001, 1.7852, -0.1670, -0.4175, 0.0122, 0.3248, 0.1241],
  30026. [ 0.5981, -0.3969, 1.6569, 0.3041, -0.4702, -0.2394, 0.2030, 0.0229],
  30027. [ 0.5747, -0.4492, 1.8774, -0.2422, -0.5260, -0.1429, 1.0017, 0.2354],
  30028. [ 0.7515, -0.2731, 1.0945, -0.9634, -0.6238, -1.0175, 0.1457, 0.3325],
  30029. [ 0.5947, -0.4267, 1.8414, -0.3665, -0.3415, 0.1742, 0.4117, 0.1350],
  30030. [ 0.6684, -0.3270, 1.7121, 0.1320, -0.5190, -0.3264, 0.3048, 0.4452],
  30031. [ 0.4449, -0.5292, 1.8183, -0.6192, -0.6611, -0.6690, 0.3456, 0.0433]],
  30032. device='cuda:0', grad_fn=<AddmmBackward>)
  30033. landmarks are: tensor([[[ 0.5484, -0.4591, 1.5940, -0.7771, -0.6693, -0.3460, 0.4219,
  30034. 0.2391],
  30035. [ 0.5250, -0.4661, 1.7383, -0.0765, -0.4268, -0.0226, 0.2535,
  30036. 0.2035],
  30037. [ 0.5320, -0.4488, 1.6633, 0.3315, -0.5018, -0.2459, 0.0828,
  30038. -0.0370],
  30039. [ 0.6174, -0.4129, 1.8711, -0.1073, -0.5480, -0.1227, 0.9558,
  30040. 0.2516],
  30041. [ 0.5548, -0.3936, 1.1634, -0.8105, -0.5192, -1.0696, 0.2372,
  30042. 0.3931],
  30043. [ 0.5604, -0.4620, 1.7961, -0.1997, -0.3229, 0.1082, 0.4104,
  30044. 0.2545],
  30045. [ 0.5897, -0.3527, 1.8018, 0.2545, -0.5307, -0.3229, 0.3122,
  30046. 0.3084],
  30047. [ 0.5807, -0.4378, 1.8249, -0.4691, -0.6289, -0.6385, 0.4104,
  30048. 0.0620]]], device='cuda:0')
  30049. loss_train_step before backward: tensor(0.0129, device='cuda:0', grad_fn=<MseLossBackward>)
  30050. loss_train_step after backward: tensor(0.0129, device='cuda:0', grad_fn=<MseLossBackward>)
  30051. loss_train: 1.5544648678041995
  30052. step: 77
  30053. running loss: 0.020187855426028564
  30054. Train Steps: 77/90 Loss: 0.0202 torch.Size([8, 600, 800])
  30055. torch.Size([8, 8])
  30056. tensor([[ nan, nan, 0.7268, 0.2333, 0.4125, 0.1933, 0.5112, 0.5383],
  30057. [0.6137, 0.4038, 0.8563, 0.4050, 0.3813, 0.2550, 0.5106, 0.4954],
  30058. [0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5463, 0.5800],
  30059. [ nan, nan, 0.6512, 0.1717, 0.4100, 0.1983, 0.5253, 0.5240],
  30060. [0.6109, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567],
  30061. [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683],
  30062. [0.6246, 0.4008, 0.8757, 0.5088, 0.4101, 0.5392, 0.6644, 0.5133],
  30063. [0.6265, 0.4091, 0.8950, 0.3533, 0.3600, 0.3967, 0.6295, 0.4901]],
  30064. device='cuda:0', dtype=torch.float64)
  30065. predictions are: tensor([[-1.9846, -2.0517, 1.0544, -1.1034, -0.4659, -1.2181, 0.2361, 0.2311],
  30066. [ 0.6813, -0.3112, 1.6636, -0.3705, -0.5481, -0.9216, 0.0807, 0.0184],
  30067. [ 0.8089, -0.2280, 1.1881, -0.7429, -0.6169, -0.8958, 0.1599, 0.2940],
  30068. [-1.8478, -1.9572, 0.9084, -1.2993, -0.4090, -1.2715, 0.3189, 0.2186],
  30069. [ 0.6891, -0.3126, 1.0502, -1.2537, -0.5350, -0.9494, 0.4252, 0.2086],
  30070. [ 0.7851, -0.2646, 1.7832, 0.0043, -0.2070, 0.1093, 0.2041, 0.1600],
  30071. [ 0.7233, -0.3370, 1.8567, 0.0620, -0.4575, 0.3413, 0.8358, 0.1053],
  30072. [ 0.7289, -0.2822, 1.9100, -0.5723, -0.6314, -0.3930, 0.5902, 0.0267]],
  30073. device='cuda:0', grad_fn=<AddmmBackward>)
  30074. landmarks are: tensor([[[-2.2859e+00, -2.2859e+00, 1.0712e+00, -1.2085e+00, -3.8060e-01,
  30075. -1.3929e+00, 7.5520e-02, 2.0062e-01],
  30076. [ 5.4850e-01, -4.2094e-01, 1.6691e+00, -4.1524e-01, -5.2494e-01,
  30077. -1.1081e+00, 7.2521e-02, 2.0831e-03],
  30078. [ 5.5484e-01, -3.9360e-01, 1.1634e+00, -8.1049e-01, -5.1917e-01,
  30079. -1.0696e+00, 2.3718e-01, 3.9307e-01],
  30080. [-2.2859e+00, -2.2859e+00, 7.2217e-01, -1.4930e+00, -3.9215e-01,
  30081. -1.3698e+00, 1.4038e-01, 1.3434e-01],
  30082. [ 5.3557e-01, -4.2171e-01, 1.0339e+00, -1.4776e+00, -5.0762e-01,
  30083. -1.1081e+00, 4.2194e-01, 2.8530e-01],
  30084. [ 5.3585e-01, -4.3703e-01, 1.7095e+00, -3.0331e-02, -8.0370e-02,
  30085. -3.8029e-02, 1.0439e-01, 3.3918e-01],
  30086. [ 5.9931e-01, -4.3453e-01, 1.7587e+00, 6.4079e-02, -3.9175e-01,
  30087. 2.0479e-01, 7.8274e-01, 8.5217e-02],
  30088. [ 6.0774e-01, -3.9646e-01, 1.8480e+00, -6.5389e-01, -6.2309e-01,
  30089. -4.5373e-01, 6.2155e-01, -2.1963e-02]]], device='cuda:0')
  30090. loss_train_step before backward: tensor(0.0198, device='cuda:0', grad_fn=<MseLossBackward>)
  30091. loss_train_step after backward: tensor(0.0198, device='cuda:0', grad_fn=<MseLossBackward>)
  30092. loss_train: 1.5742266257293522
  30093. step: 78
  30094. running loss: 0.020182392637555797
  30095. Train Steps: 78/90 Loss: 0.0202 torch.Size([8, 600, 800])
  30096. torch.Size([8, 8])
  30097. tensor([[0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374],
  30098. [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350],
  30099. [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646],
  30100. [0.6286, 0.4040, 0.8696, 0.3047, 0.3924, 0.2887, 0.6300, 0.5367],
  30101. [0.6160, 0.4093, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808],
  30102. [0.6304, 0.4029, 0.8413, 0.5217, 0.4125, 0.5617, 0.7089, 0.5679],
  30103. [0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166],
  30104. [0.6265, 0.4071, 0.8875, 0.3367, 0.3975, 0.3350, 0.6312, 0.5250]],
  30105. device='cuda:0', dtype=torch.float64)
  30106. predictions are: tensor([[ 0.3673, -0.5634, 1.6359, -0.6539, -0.6386, -0.7926, 0.4197, 0.1037],
  30107. [ 0.7273, -0.2814, 1.7136, -0.3755, -0.6066, -0.4177, 0.0784, 0.1788],
  30108. [ 0.4522, -0.4597, 1.3317, -0.7944, -0.7027, -0.8624, -0.1578, 0.1287],
  30109. [ 0.5168, -0.4615, 1.5470, -0.9493, -0.4533, -0.9150, 0.5691, 0.2001],
  30110. [ 0.4989, -0.4498, 1.5476, -0.3193, -0.5839, -0.1889, 0.1756, 0.3494],
  30111. [ 0.5740, -0.4453, 1.6434, 0.0708, -0.4044, 0.3415, 0.9822, 0.2546],
  30112. [ 0.4685, -0.5086, 1.6151, 0.0564, -0.3354, 0.2027, 0.7280, 0.0913],
  30113. [ 0.3415, -0.5573, 1.7585, -0.8527, -0.4933, -0.7010, 0.5609, 0.1285]],
  30114. device='cuda:0', grad_fn=<AddmmBackward>)
  30115. landmarks are: tensor([[[ 0.5777, -0.4416, 1.7044, -0.5827, -0.5962, -0.8361, 0.4862,
  30116. 0.1963],
  30117. [ 0.6072, -0.3239, 1.8423, -0.3537, -0.4961, -0.3921, 0.2083,
  30118. 0.1852],
  30119. [ 0.5483, -0.4105, 1.4208, -0.8002, -0.6000, -0.9002, 0.0511,
  30120. 0.3220],
  30121. [ 0.6174, -0.4201, 1.7309, -0.8784, -0.4735, -0.9524, 0.6242,
  30122. 0.1931],
  30123. [ 0.5592, -0.3956, 1.5543, -0.2456, -0.5885, -0.1689, 0.1392,
  30124. 0.3968],
  30125. [ 0.6257, -0.4249, 1.5998, 0.1236, -0.3806, 0.3084, 0.9887,
  30126. 0.3371],
  30127. [ 0.6042, -0.4273, 1.6920, 0.1860, -0.2717, 0.1406, 0.7997,
  30128. 0.1004],
  30129. [ 0.6078, -0.4057, 1.8134, -0.7309, -0.4499, -0.7386, 0.6298,
  30130. 0.1390]]], device='cuda:0')
  30131. loss_train_step before backward: tensor(0.0083, device='cuda:0', grad_fn=<MseLossBackward>)
  30132. loss_train_step after backward: tensor(0.0083, device='cuda:0', grad_fn=<MseLossBackward>)
  30133. loss_train: 1.5825061020441353
  30134. step: 79
  30135. running loss: 0.020031722810685257
  30136. Train Steps: 79/90 Loss: 0.0200 torch.Size([8, 600, 800])
  30137. torch.Size([8, 8])
  30138. tensor([[0.6201, 0.4102, 0.7288, 0.2417, 0.4150, 0.2383, 0.6100, 0.5500],
  30139. [0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058],
  30140. [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690],
  30141. [0.6111, 0.3995, 0.8788, 0.4567, 0.3813, 0.4833, 0.5450, 0.5700],
  30142. [ nan, nan, 0.7612, 0.3250, 0.4037, 0.2533, 0.5438, 0.5767],
  30143. [0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204],
  30144. [0.6197, 0.4118, 0.8688, 0.5517, 0.4037, 0.5233, 0.5875, 0.5600],
  30145. [0.6099, 0.4030, 0.8638, 0.5117, 0.4983, 0.4965, 0.5086, 0.5388]],
  30146. device='cuda:0', dtype=torch.float64)
  30147. predictions are: tensor([[ 0.6953, -0.3039, 1.0919, -1.2795, -0.5539, -1.2303, 0.4363, 0.1660],
  30148. [ 0.6522, -0.3437, 1.6727, -0.4270, -0.6701, -0.1970, 0.1418, -0.0034],
  30149. [ 0.4385, -0.5066, 1.6041, 0.0371, -0.4771, 0.1611, 1.0337, 0.2570],
  30150. [ 0.6591, -0.3570, 1.7664, -0.3257, -0.6140, -0.1533, 0.2319, 0.1825],
  30151. [-1.6271, -1.8176, 1.2880, -1.0241, -0.5292, -1.1064, 0.1217, 0.2967],
  30152. [ 0.5874, -0.4156, 1.6548, 0.0828, -0.4626, 0.0553, 0.7523, 0.0670],
  30153. [ 0.7439, -0.2793, 1.7456, 0.1186, -0.5632, 0.0915, 0.3984, 0.1757],
  30154. [ 0.5422, -0.4143, 1.7436, -0.1030, -0.1236, -0.1744, 0.0474, 0.1065]],
  30155. device='cuda:0', grad_fn=<AddmmBackward>)
  30156. landmarks are: tensor([[[ 0.5784, -0.3913, 1.0801, -1.1697, -0.3691, -1.1851, 0.5316,
  30157. 0.2545],
  30158. [ 0.5245, -0.4361, 1.5940, -0.2921, -0.5480, -0.0919, 0.2432,
  30159. 0.0502],
  30160. [ 0.6306, -0.4153, 1.5141, 0.2224, -0.3691, 0.2622, 1.0033,
  30161. 0.3425],
  30162. [ 0.5368, -0.4406, 1.7730, -0.1766, -0.5249, -0.0534, 0.2314,
  30163. 0.3469],
  30164. [-2.2859, -2.2859, 1.2303, -0.7848, -0.4210, -1.1158, 0.2256,
  30165. 0.3777],
  30166. [ 0.6042, -0.4273, 1.7198, 0.2184, -0.3478, 0.1149, 0.8062,
  30167. 0.1176],
  30168. [ 0.5762, -0.3840, 1.7268, 0.2622, -0.4210, 0.1313, 0.4277,
  30169. 0.3007],
  30170. [ 0.5309, -0.4246, 1.7037, 0.0774, 0.0158, 0.0075, 0.0635,
  30171. 0.2026]]], device='cuda:0')
  30172. loss_train_step before backward: tensor(0.0213, device='cuda:0', grad_fn=<MseLossBackward>)
  30173. loss_train_step after backward: tensor(0.0213, device='cuda:0', grad_fn=<MseLossBackward>)
  30174. loss_train: 1.603833228815347
  30175. step: 80
  30176. running loss: 0.020047915360191838
  30177.  
  30178. Train Steps: 80/90 Loss: 0.0200 torch.Size([8, 600, 800])
  30179. torch.Size([8, 8])
  30180. tensor([[0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167],
  30181. [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320],
  30182. [0.6201, 0.4036, 0.8596, 0.3850, 0.3492, 0.3785, 0.5978, 0.5131],
  30183. [0.6275, 0.4050, 0.9038, 0.3767, 0.3838, 0.3533, 0.7074, 0.5575],
  30184. [0.6204, 0.4049, 0.7975, 0.2700, 0.3937, 0.2567, 0.5700, 0.5183],
  30185. [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617],
  30186. [0.6201, 0.3970, 0.8413, 0.4950, 0.4413, 0.5183, 0.6088, 0.5400],
  30187. [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208]],
  30188. device='cuda:0', dtype=torch.float64)
  30189. predictions are: tensor([[ 0.3734, -0.5190, 1.1472, -0.9185, -0.5289, -0.9328, 0.3290, 0.5005],
  30190. [ 0.1556, -0.7042, 1.3313, -1.2112, -0.4058, -0.9651, 0.7235, 0.1515],
  30191. [ 0.7030, -0.3688, 1.7134, -0.4899, -0.7348, -0.4037, 0.3331, 0.0598],
  30192. [ 0.4604, -0.5118, 1.8374, -0.4674, -0.5123, -0.5352, 0.8484, 0.2562],
  30193. [ 0.7388, -0.3145, 1.3714, -1.0121, -0.5813, -1.0120, 0.1488, 0.0172],
  30194. [ 0.5941, -0.3950, 1.7309, -0.0935, -0.3420, 0.1457, 0.3144, 0.1970],
  30195. [ 0.5540, -0.4376, 1.6098, 0.0253, -0.2959, 0.2028, 0.3435, 0.1415],
  30196. [ 0.2762, -0.6084, 1.4040, -0.9958, -0.4741, -1.0922, 0.3900, 0.0834]],
  30197. device='cuda:0', grad_fn=<AddmmBackward>)
  30198. landmarks are: tensor([[[ 5.6307e-01, -4.1286e-01, 1.2129e+00, -9.2333e-01, -4.1524e-01,
  30199. -1.0311e+00, 4.5658e-01, 5.6243e-01],
  30200. [ 6.1413e-01, -4.1527e-01, 1.4208e+00, -1.2697e+00, -2.9400e-01,
  30201. -1.0234e+00, 8.6439e-01, 1.7146e-01],
  30202. [ 5.7829e-01, -4.2163e-01, 1.6847e+00, -5.0778e-01, -6.7321e-01,
  30203. -5.3774e-01, 4.7523e-01, 8.3916e-02],
  30204. [ 6.1248e-01, -4.1527e-01, 1.8885e+00, -5.4611e-01, -5.1339e-01,
  30205. -6.5389e-01, 9.8137e-01, 2.8902e-01],
  30206. [ 5.7985e-01, -4.1555e-01, 1.3977e+00, -1.0388e+00, -4.6721e-01,
  30207. -1.1004e+00, 3.4688e-01, 1.0824e-01],
  30208. [ 5.6374e-01, -4.1432e-01, 1.7519e+00, -7.8656e-02, -3.0554e-01,
  30209. -1.4935e-02, 3.7575e-01, 3.0839e-01],
  30210. [ 5.7812e-01, -4.5219e-01, 1.5998e+00, 4.6189e-04, -2.4781e-01,
  30211. 1.0824e-01, 5.2587e-01, 2.0831e-01],
  30212. [ 5.8851e-01, -4.4288e-01, 1.4266e+00, -9.9261e-01, -4.3834e-01,
  30213. -1.2313e+00, 4.2276e-01, 1.1948e-01]]], device='cuda:0')
  30214. loss_train_step before backward: tensor(0.0144, device='cuda:0', grad_fn=<MseLossBackward>)
  30215. loss_train_step after backward: tensor(0.0144, device='cuda:0', grad_fn=<MseLossBackward>)
  30216. loss_train: 1.6182256652973592
  30217. step: 81
  30218. running loss: 0.01997809463330073
  30219. Train Steps: 81/90 Loss: 0.0200 torch.Size([8, 600, 800])
  30220. torch.Size([8, 8])
  30221. tensor([[0.6275, 0.4081, 0.8063, 0.2017, 0.4825, 0.1583, 0.6156, 0.4869],
  30222. [0.6125, 0.3974, 0.7725, 0.2517, 0.3538, 0.3317, 0.5887, 0.5500],
  30223. [0.6183, 0.4076, 0.8838, 0.4517, 0.3813, 0.4483, 0.5775, 0.5633],
  30224. [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400],
  30225. [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301],
  30226. [0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835],
  30227. [0.6250, 0.3993, 0.9138, 0.4333, 0.3763, 0.5217, 0.6995, 0.5320],
  30228. [0.6293, 0.4097, 0.8800, 0.2517, 0.5262, 0.2600, 0.7430, 0.5378]],
  30229. device='cuda:0', dtype=torch.float64)
  30230. predictions are: tensor([[ 0.2903, -0.5904, 1.3713, -1.2827, -0.1585, -1.3888, 0.3519, 0.0815],
  30231. [ 0.3774, -0.5230, 1.1499, -0.9511, -0.7478, -0.5249, 0.2670, 0.2296],
  30232. [ 0.6424, -0.3445, 1.6242, -0.1800, -0.6430, 0.0166, 0.1866, 0.3549],
  30233. [ 0.3907, -0.5244, 1.1680, -1.1063, -0.3419, -1.1686, 0.3648, 0.2469],
  30234. [ 0.7197, -0.3552, 1.6134, 0.0642, -0.4562, 0.1645, 0.5280, 0.1403],
  30235. [ 0.5325, -0.4282, 1.7349, -0.4375, -0.4583, -1.1334, 0.2133, 0.0699],
  30236. [ 0.4556, -0.5406, 1.7986, -0.2391, -0.6904, 0.2913, 0.8318, 0.1297],
  30237. [ 0.4897, -0.4823, 1.6821, -1.0455, 0.0482, -1.0562, 0.9698, 0.2569]],
  30238. device='cuda:0', grad_fn=<AddmmBackward>)
  30239. landmarks are: tensor([[[ 0.6124, -0.4010, 1.4381, -1.3544, -0.0573, -1.5546, 0.5573,
  30240. -0.0369],
  30241. [ 0.5430, -0.4503, 1.2822, -1.1235, -0.6520, -0.7540, 0.4335,
  30242. 0.2545],
  30243. [ 0.5700, -0.4032, 1.7961, -0.1997, -0.5249, -0.2151, 0.3815,
  30244. 0.3161],
  30245. [ 0.5842, -0.3848, 1.2938, -1.2159, -0.2132, -1.4160, 0.5778,
  30246. 0.2083],
  30247. [ 0.6010, -0.4562, 1.7198, -0.0090, -0.3464, 0.0108, 0.6294,
  30248. 0.1627],
  30249. [ 0.6117, -0.3877, 1.8249, -0.5230, -0.3517, -1.2851, 0.4200,
  30250. -0.0529],
  30251. [ 0.6010, -0.4417, 1.9346, -0.2844, -0.5480, 0.1236, 0.9448,
  30252. 0.1715],
  30253. [ 0.6207, -0.3936, 1.7788, -1.1235, 0.1448, -1.0850, 1.1459,
  30254. 0.1982]]], device='cuda:0')
  30255. loss_train_step before backward: tensor(0.0183, device='cuda:0', grad_fn=<MseLossBackward>)
  30256. loss_train_step after backward: tensor(0.0183, device='cuda:0', grad_fn=<MseLossBackward>)
  30257. loss_train: 1.6365177757106721
  30258. step: 82
  30259. running loss: 0.019957533850130146
  30260. Train Steps: 82/90 Loss: 0.0200 torch.Size([8, 600, 800])
  30261. torch.Size([8, 8])
  30262. tensor([[0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204],
  30263. [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6038, 0.4833],
  30264. [0.6230, 0.4113, 0.7213, 0.1983, 0.4325, 0.2367, 0.6262, 0.5400],
  30265. [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617],
  30266. [ nan, nan, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552],
  30267. [0.6199, 0.4071, 0.7600, 0.2117, 0.4037, 0.2767, 0.6138, 0.5550],
  30268. [0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767],
  30269. [ nan, nan, 0.6688, 0.2513, 0.4113, 0.2117, 0.5193, 0.5933]],
  30270. device='cuda:0', dtype=torch.float64)
  30271. predictions are: tensor([[ 0.8619, -0.2506, 1.5938, 0.1987, -0.4249, 0.2109, 0.7913, 0.0299],
  30272. [ 0.8644, -0.2650, 1.7133, -0.0653, -0.5711, -0.0539, 0.4279, -0.0110],
  30273. [ 0.7838, -0.2591, 1.1059, -1.2509, -0.3135, -1.1064, 0.5907, 0.2126],
  30274. [ 0.8056, -0.2161, 1.6969, 0.2063, -0.5404, -0.2532, 0.2598, 0.3049],
  30275. [-2.0843, -2.1210, 1.6906, -1.1510, 0.0211, -1.1015, 0.7819, 0.1954],
  30276. [ 0.7208, -0.2959, 1.2570, -1.1981, -0.3980, -0.8894, 0.5388, 0.1905],
  30277. [ 0.8011, -0.2003, 1.7396, -0.2197, -0.6043, -0.5010, 0.2048, 0.3001],
  30278. [-1.8934, -1.9678, 0.8566, -1.1885, -0.4570, -1.1944, 0.0053, 0.2951]],
  30279. device='cuda:0', grad_fn=<AddmmBackward>)
  30280. landmarks are: tensor([[[ 0.6042, -0.4273, 1.7198, 0.2184, -0.3478, 0.1149, 0.8062,
  30281. 0.1176],
  30282. [ 0.5726, -0.4795, 1.7788, -0.0149, -0.5711, -0.0688, 0.5028,
  30283. -0.0534],
  30284. [ 0.5917, -0.3861, 1.0455, -1.3698, -0.2882, -1.1928, 0.6067,
  30285. 0.2083],
  30286. [ 0.5897, -0.3527, 1.8018, 0.2545, -0.5307, -0.3229, 0.3122,
  30287. 0.3084],
  30288. [-2.2859, -2.2859, 1.7557, -1.1466, 0.0871, -1.0773, 1.1239,
  30289. 0.2783],
  30290. [ 0.5775, -0.4054, 1.2245, -1.3082, -0.4210, -1.0080, 0.5490,
  30291. 0.2776],
  30292. [ 0.6058, -0.3216, 1.8423, -0.2536, -0.5885, -0.6000, 0.3353,
  30293. 0.3777],
  30294. [-2.2859, -2.2859, 0.8033, -1.1250, -0.3864, -1.3082, 0.1126,
  30295. 0.4543]]], device='cuda:0')
  30296. loss_train_step before backward: tensor(0.0177, device='cuda:0', grad_fn=<MseLossBackward>)
  30297. loss_train_step after backward: tensor(0.0177, device='cuda:0', grad_fn=<MseLossBackward>)
  30298. loss_train: 1.654198335018009
  30299. step: 83
  30300. running loss: 0.019930100421903723
  30301. Train Steps: 83/90 Loss: 0.0199 torch.Size([8, 600, 800])
  30302. torch.Size([8, 8])
  30303. tensor([[0.6256, 0.4199, 0.8638, 0.5800, 0.3987, 0.4383, 0.5600, 0.5950],
  30304. [0.6325, 0.4165, 0.9000, 0.4617, 0.3813, 0.4900, 0.7485, 0.5447],
  30305. [0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333],
  30306. [0.6197, 0.4091, 0.8800, 0.4783, 0.3538, 0.4767, 0.5950, 0.5550],
  30307. [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649],
  30308. [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6188, 0.5283],
  30309. [0.6147, 0.4081, 0.8538, 0.3400, 0.3663, 0.3150, 0.5142, 0.4875],
  30310. [0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266]],
  30311. device='cuda:0', dtype=torch.float64)
  30312. predictions are: tensor([[ 0.5667, -0.4106, 1.5669, 0.2542, -0.3283, -0.3381, 0.2532, 0.4962],
  30313. [ 0.4737, -0.5161, 1.7973, -0.2878, -0.4420, 0.0469, 1.0790, 0.2255],
  30314. [ 0.5542, -0.4182, 1.7720, -0.4122, -0.4295, 0.0192, 0.5208, 0.1936],
  30315. [ 0.6322, -0.4009, 1.6925, -0.2425, -0.4987, -0.1369, 0.4719, 0.3256],
  30316. [ 0.5740, -0.4672, 1.5395, 0.3276, -0.4665, 0.0108, 0.5626, -0.0387],
  30317. [ 0.6924, -0.3483, 1.8122, -0.8088, -0.3993, -1.0986, 0.6015, 0.1548],
  30318. [ 0.6318, -0.3645, 1.6522, -0.8129, -0.5442, -0.8673, 0.1077, -0.0085],
  30319. [ 0.2982, -0.5549, 1.0625, -1.3172, -0.2534, -1.1064, 0.5091, 0.5923]],
  30320. device='cuda:0', grad_fn=<AddmmBackward>)
  30321. landmarks are: tensor([[[ 0.6038, -0.3464, 1.7037, 0.3931, -0.4441, -0.2613, 0.3007,
  30322. 0.4624],
  30323. [ 0.6355, -0.3623, 1.8711, -0.1535, -0.5249, -0.0226, 1.1715,
  30324. 0.2302],
  30325. [ 0.6003, -0.3747, 1.8885, -0.1997, -0.5018, -0.0149, 0.5490,
  30326. 0.1775],
  30327. [ 0.5763, -0.3963, 1.7788, -0.0765, -0.6520, -0.0842, 0.4624,
  30328. 0.2776],
  30329. [ 0.6105, -0.4360, 1.6171, 0.5162, -0.5076, 0.0159, 0.5171,
  30330. -0.1385],
  30331. [ 0.6062, -0.3778, 1.8423, -0.6462, -0.4383, -1.0465, 0.5721,
  30332. 0.1544],
  30333. [ 0.5532, -0.4008, 1.6575, -0.7155, -0.5942, -0.8309, 0.0890,
  30334. -0.0340],
  30335. [ 0.5746, -0.4021, 1.0801, -1.1312, -0.3229, -1.1081, 0.4803,
  30336. 0.6084]]], device='cuda:0')
  30337. loss_train_step before backward: tensor(0.0087, device='cuda:0', grad_fn=<MseLossBackward>)
  30338. loss_train_step after backward: tensor(0.0087, device='cuda:0', grad_fn=<MseLossBackward>)
  30339. loss_train: 1.6629118216224015
  30340. step: 84
  30341. running loss: 0.01979656930502859
  30342.  
  30343. Train Steps: 84/90 Loss: 0.0198 torch.Size([8, 600, 800])
  30344. torch.Size([8, 8])
  30345. tensor([[0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268],
  30346. [0.6223, 0.4028, 0.8988, 0.4200, 0.3763, 0.5733, 0.6375, 0.5167],
  30347. [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617],
  30348. [0.6083, 0.3957, 0.8638, 0.4950, 0.4363, 0.5083, 0.5346, 0.4980],
  30349. [0.6212, 0.4159, 0.8675, 0.5783, 0.4088, 0.4317, 0.5613, 0.5917],
  30350. [0.6248, 0.4032, 0.7738, 0.1900, 0.4813, 0.1400, 0.5941, 0.4904],
  30351. [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852],
  30352. [0.6280, 0.4101, 0.9050, 0.4533, 0.3775, 0.3217, 0.6338, 0.4915]],
  30353. device='cuda:0', dtype=torch.float64)
  30354. predictions are: tensor([[ 0.6178, -0.4126, 1.6719, -0.2266, -0.2874, 0.0123, 0.7712, 0.1822],
  30355. [ 0.6737, -0.3681, 1.8302, -0.4825, -0.5582, 0.3299, 0.8426, 0.1744],
  30356. [ 0.4683, -0.4485, 1.6903, 0.0651, -0.5430, -0.3624, 0.4358, 0.4397],
  30357. [ 0.4677, -0.5152, 1.6872, -0.1683, -0.3159, 0.0394, 0.3376, 0.0098],
  30358. [ 0.4362, -0.4553, 1.5300, 0.2041, -0.3977, -0.2993, 0.3835, 0.5141],
  30359. [ 0.7591, -0.2939, 1.3000, -1.4181, -0.0185, -1.6864, 0.5949, 0.1079],
  30360. [ 0.4794, -0.4698, 1.6591, -0.1646, -0.1904, -0.1011, 0.2546, 0.4571],
  30361. [ 0.5885, -0.4134, 1.8501, -0.3274, -0.6333, -0.8266, 0.6597, 0.0587]],
  30362. device='cuda:0', grad_fn=<AddmmBackward>)
  30363. landmarks are: tensor([[[ 5.8284e-01, -4.6823e-01, 1.7031e+00, -4.9668e-02, -2.4581e-01,
  30364. 8.1770e-02, 6.3811e-01, 1.4745e-01],
  30365. [ 5.8857e-01, -4.2525e-01, 1.8654e+00, -3.4596e-01, -5.4804e-01,
  30366. 3.6228e-01, 6.5866e-01, 1.0054e-01],
  30367. [ 5.8972e-01, -3.5273e-01, 1.8018e+00, 2.5450e-01, -5.3072e-01,
  30368. -3.2286e-01, 3.1224e-01, 3.0839e-01],
  30369. [ 5.2379e-01, -4.5797e-01, 1.7037e+00, 4.6189e-04, -2.7090e-01,
  30370. 6.2048e-02, 1.8356e-01, 1.4106e-02],
  30371. [ 5.8360e-01, -3.6490e-01, 1.7210e+00, 3.8537e-01, -3.9792e-01,
  30372. -2.9207e-01, 3.0647e-01, 4.4696e-01],
  30373. [ 5.9994e-01, -4.2363e-01, 1.2880e+00, -1.4083e+00, -6.3048e-02,
  30374. -1.6393e+00, 4.5840e-01, -2.0790e-02],
  30375. [ 5.5000e-01, -4.0600e-01, 1.7326e+00, 2.3557e-02, -1.5543e-01,
  30376. -2.2633e-02, 1.4385e-01, 4.1710e-01],
  30377. [ 6.1484e-01, -3.9184e-01, 1.8942e+00, -1.9199e-01, -5.4226e-01,
  30378. -8.0015e-01, 6.4140e-01, -1.5569e-02]]], device='cuda:0')
  30379. loss_train_step before backward: tensor(0.0094, device='cuda:0', grad_fn=<MseLossBackward>)
  30380. loss_train_step after backward: tensor(0.0094, device='cuda:0', grad_fn=<MseLossBackward>)
  30381. loss_train: 1.6722815376706421
  30382. step: 85
  30383. running loss: 0.019673900443184025
  30384. Train Steps: 85/90 Loss: 0.0197 torch.Size([8, 600, 800])
  30385. torch.Size([8, 8])
  30386. tensor([[0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167],
  30387. [0.6200, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913],
  30388. [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633],
  30389. [0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6138, 0.5450],
  30390. [0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938],
  30391. [0.6321, 0.4048, 0.8738, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927],
  30392. [0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136],
  30393. [0.6142, 0.3982, 0.8650, 0.4883, 0.3912, 0.4317, 0.5315, 0.5350]],
  30394. device='cuda:0', dtype=torch.float64)
  30395. predictions are: tensor([[ 0.5126, -0.3980, 1.1951, -0.8997, 0.0626, -1.4429, 0.3967, 0.6447],
  30396. [ 0.2296, -0.6835, 1.7541, -0.0746, -0.3331, 0.0221, 0.4547, 0.1091],
  30397. [ 0.6355, -0.3560, 1.8110, -0.0143, -0.5147, -0.2400, 0.4765, 0.4517],
  30398. [ 0.7565, -0.2948, 1.8072, -0.3907, -0.2909, 0.1811, 0.6384, 0.2877],
  30399. [ 0.8373, -0.2884, 1.7138, 0.2233, -0.4457, -0.0877, 0.8785, 0.0349],
  30400. [ 0.7473, -0.3367, 1.7696, 0.1339, -0.5059, -0.2661, 0.7932, 0.0021],
  30401. [ 0.4635, -0.4734, 1.6943, -0.5987, -0.5666, -0.5225, 0.3460, 0.1546],
  30402. [ 0.3809, -0.5535, 1.7622, -0.1957, -0.4246, -0.2758, 0.3063, 0.2153]],
  30403. device='cuda:0', grad_fn=<AddmmBackward>)
  30404. landmarks are: tensor([[[ 0.6161, -0.3075, 1.1678, -0.6500, 0.0813, -1.4006, 0.2545,
  30405. 0.5624],
  30406. [ 0.5776, -0.4784, 1.7730, 0.1236, -0.4037, 0.0697, 0.5132,
  30407. -0.0168],
  30408. [ 0.5879, -0.3591, 1.8018, 0.1236, -0.5596, -0.1612, 0.3469,
  30409. 0.3161],
  30410. [ 0.5960, -0.4102, 1.8018, -0.1612, -0.3344, 0.1159, 0.5490,
  30411. 0.2314],
  30412. [ 0.6395, -0.4213, 1.7037, 0.3623, -0.4326, -0.1073, 0.6560,
  30413. -0.0049],
  30414. [ 0.6335, -0.4162, 1.7499, 0.3084, -0.4961, -0.2459, 0.6524,
  30415. -0.0102],
  30416. [ 0.5235, -0.4273, 1.7499, -0.4306, -0.5827, -0.4614, 0.1651,
  30417. 0.0862],
  30418. [ 0.5513, -0.4467, 1.7095, -0.0303, -0.4788, -0.2921, 0.1692,
  30419. 0.1854]]], device='cuda:0')
  30420. loss_train_step before backward: tensor(0.0138, device='cuda:0', grad_fn=<MseLossBackward>)
  30421. loss_train_step after backward: tensor(0.0138, device='cuda:0', grad_fn=<MseLossBackward>)
  30422. loss_train: 1.686103314626962
  30423. step: 86
  30424. running loss: 0.019605852495662348
  30425. Train Steps: 86/90 Loss: 0.0196 torch.Size([8, 600, 800])
  30426. torch.Size([8, 8])
  30427. tensor([[0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986],
  30428. [0.6131, 0.4064, 0.8638, 0.5200, 0.4788, 0.4783, 0.5258, 0.5867],
  30429. [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567],
  30430. [ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600],
  30431. [0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317],
  30432. [0.6128, 0.4084, 0.8738, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397],
  30433. [ nan, nan, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621],
  30434. [0.6147, 0.4112, 0.7988, 0.3200, 0.3775, 0.2767, 0.5150, 0.5550]],
  30435. device='cuda:0', dtype=torch.float64)
  30436. predictions are: tensor([[ 0.7564, -0.2589, 1.8650, -0.5441, -0.1607, -1.1702, 0.7058, 0.0304],
  30437. [ 0.8369, -0.2295, 1.6265, 0.1115, -0.1178, 0.0394, 0.4619, 0.3338],
  30438. [ 0.8468, -0.2518, 1.6333, -0.1932, -0.5100, -0.3991, 0.6418, 0.2949],
  30439. [-1.7919, -1.9164, 1.2644, -0.9980, -0.3826, -0.9792, 0.2770, 0.2354],
  30440. [ 0.7303, -0.2999, 1.2405, -1.0710, -0.2661, -1.0750, 0.7384, 0.1537],
  30441. [ 0.7327, -0.2754, 1.6966, -0.0502, -0.5937, -0.4542, 0.3500, 0.1744],
  30442. [-2.0738, -2.0970, 1.5829, -1.1729, 0.1751, -1.2038, 0.8197, 0.2315],
  30443. [ 0.7515, -0.2552, 1.3871, -0.7542, -0.4840, -0.8544, 0.2894, 0.1536]],
  30444. device='cuda:0', grad_fn=<AddmmBackward>)
  30445. landmarks are: tensor([[[ 0.6054, -0.3641, 1.8711, -0.5153, -0.2305, -1.3005, 0.5948,
  30446. 0.0171],
  30447. [ 0.5462, -0.4090, 1.7037, 0.1159, -0.0746, -0.0765, 0.1427,
  30448. 0.4239],
  30449. [ 0.5432, -0.4336, 1.7095, -0.1766, -0.5942, -0.4845, 0.3007,
  30450. 0.2853],
  30451. [-2.2859, -2.2859, 1.4266, -1.1389, -0.4499, -1.1235, 0.2891,
  30452. 0.3007],
  30453. [ 0.5849, -0.3836, 1.2649, -1.2236, -0.3460, -1.2313, 0.4508,
  30454. 0.1698],
  30455. [ 0.5443, -0.3994, 1.7499, -0.1227, -0.6173, -0.5769, 0.0049,
  30456. 0.2071],
  30457. [-2.2859, -2.2859, 1.7210, -0.9772, 0.1852, -1.3698, 0.7986,
  30458. 0.3104],
  30459. [ 0.5532, -0.3864, 1.4035, -0.8079, -0.5423, -1.0080, 0.0928,
  30460. 0.2776]]], device='cuda:0')
  30461. loss_train_step before backward: tensor(0.0267, device='cuda:0', grad_fn=<MseLossBackward>)
  30462. loss_train_step after backward: tensor(0.0267, device='cuda:0', grad_fn=<MseLossBackward>)
  30463. loss_train: 1.7128123142756522
  30464. step: 87
  30465. running loss: 0.01968749786523738
  30466. Train Steps: 87/90 Loss: 0.0197 torch.Size([8, 600, 800])
  30467. torch.Size([8, 8])
  30468. tensor([[0.6120, 0.4014, 0.6863, 0.2817, 0.3700, 0.2783, 0.5513, 0.5667],
  30469. [0.6262, 0.4163, 0.8850, 0.5183, 0.3763, 0.4150, 0.6025, 0.5500],
  30470. [0.6075, 0.4007, 0.8275, 0.4917, 0.4050, 0.5100, 0.5167, 0.5280],
  30471. [0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835],
  30472. [0.6126, 0.4067, 0.8638, 0.5383, 0.4188, 0.4850, 0.5016, 0.5392],
  30473. [0.6163, 0.4114, 0.7650, 0.2017, 0.3763, 0.2867, 0.5631, 0.5071],
  30474. [0.6170, 0.4102, 0.7468, 0.3695, 0.3463, 0.3767, 0.5238, 0.5823],
  30475. [0.6204, 0.4007, 0.7838, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167]],
  30476. device='cuda:0', dtype=torch.float64)
  30477. predictions are: tensor([[ 5.6286e-01, -4.0666e-01, 1.1249e+00, -9.1667e-01, -4.4478e-01,
  30478. -9.4364e-01, 4.9949e-01, 3.6532e-01],
  30479. [ 4.0523e-01, -5.1145e-01, 1.8993e+00, 9.8369e-02, -4.8307e-01,
  30480. -4.0015e-01, 6.8296e-01, 2.3447e-01],
  30481. [ 4.5955e-01, -4.7740e-01, 1.6666e+00, -1.6434e-02, -2.6750e-01,
  30482. 5.4508e-02, 3.1240e-01, 1.9984e-01],
  30483. [ 2.8139e-01, -6.2768e-01, 2.0158e+00, -5.1398e-01, -2.0669e-01,
  30484. -1.3062e+00, 6.4697e-01, 7.9610e-03],
  30485. [ 5.3721e-01, -4.3750e-01, 1.7832e+00, 2.9688e-01, -2.3398e-01,
  30486. -1.2247e-03, 2.8872e-01, 1.9452e-01],
  30487. [ 6.6184e-01, -4.0100e-01, 1.3221e+00, -1.2418e+00, -4.3192e-01,
  30488. -9.4730e-01, 6.9494e-01, 1.1958e-01],
  30489. [ 5.3418e-01, -4.0369e-01, 1.3463e+00, -5.7753e-01, -5.9096e-01,
  30490. -5.4382e-01, 4.0242e-01, 4.7178e-01],
  30491. [ 5.9500e-01, -4.2295e-01, 1.4109e+00, -1.1970e+00, -6.6504e-02,
  30492. -1.4813e+00, 6.5024e-01, 1.0265e-01]], device='cuda:0',
  30493. grad_fn=<AddmmBackward>)
  30494. landmarks are: tensor([[[ 0.5410, -0.4321, 0.8838, -0.9849, -0.5769, -1.0003, 0.2603,
  30495. 0.3315],
  30496. [ 0.6066, -0.3632, 1.8018, 0.1082, -0.5480, -0.3691, 0.4970,
  30497. 0.2545],
  30498. [ 0.5200, -0.4353, 1.5363, -0.0149, -0.4152, 0.0697, 0.1005,
  30499. 0.1530],
  30500. [ 0.6117, -0.3877, 1.8249, -0.5230, -0.3517, -1.2851, 0.4200,
  30501. -0.0529],
  30502. [ 0.5436, -0.4076, 1.7037, 0.2006, -0.3517, -0.0457, 0.0311,
  30503. 0.2048],
  30504. [ 0.5606, -0.3856, 1.2476, -1.3544, -0.5480, -0.9618, 0.3152,
  30505. 0.0562],
  30506. [ 0.5639, -0.3911, 1.1634, -0.5794, -0.6866, -0.5461, 0.1334,
  30507. 0.4036],
  30508. [ 0.5796, -0.4351, 1.3342, -1.3159, -0.2189, -1.4853, 0.4046,
  30509. 0.1005]]], device='cuda:0')
  30510. loss_train_step before backward: tensor(0.0178, device='cuda:0', grad_fn=<MseLossBackward>)
  30511. loss_train_step after backward: tensor(0.0178, device='cuda:0', grad_fn=<MseLossBackward>)
  30512. loss_train: 1.7306391117163002
  30513. step: 88
  30514. running loss: 0.019666353542230685
  30515.  
  30516. Train Steps: 88/90 Loss: 0.0197 torch.Size([8, 600, 800])
  30517. torch.Size([8, 8])
  30518. tensor([[0.6197, 0.4118, 0.8688, 0.5517, 0.4037, 0.5233, 0.5875, 0.5600],
  30519. [0.6031, 0.3986, 0.7462, 0.2833, 0.3638, 0.2717, 0.5253, 0.4726],
  30520. [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285],
  30521. [0.6346, 0.4144, 0.9088, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899],
  30522. [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600],
  30523. [ nan, nan, 0.8363, 0.3317, 0.3563, 0.3367, 0.5329, 0.5142],
  30524. [0.6304, 0.4024, 0.8925, 0.4800, 0.3937, 0.4817, 0.7485, 0.5297],
  30525. [0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038]],
  30526. device='cuda:0', dtype=torch.float64)
  30527. predictions are: tensor([[ 0.6136, -0.3292, 1.7236, 0.2335, -0.3311, 0.0264, 0.4263, 0.2954],
  30528. [ 0.6881, -0.2870, 1.2374, -1.0072, -0.4346, -1.1297, 0.2528, 0.0145],
  30529. [ 0.5846, -0.4008, 1.6797, 0.3026, -0.3951, -0.1516, 0.7463, 0.1915],
  30530. [ 0.4226, -0.4940, 1.8812, -0.1699, -0.2164, -0.4229, 0.9591, 0.3326],
  30531. [ 0.6220, -0.3374, 1.7867, 0.2923, -0.3024, -0.1980, 0.3742, 0.3229],
  30532. [-2.1704, -2.2239, 1.4709, -0.8261, -0.5161, -0.9168, 0.2164, 0.1590],
  30533. [ 0.6821, -0.3473, 1.7859, -0.0876, -0.3812, -0.1809, 0.9278, 0.1667],
  30534. [ 0.6349, -0.3336, 1.7291, -0.1699, -0.1123, -0.1446, 0.1518, 0.0453]],
  30535. device='cuda:0', grad_fn=<AddmmBackward>)
  30536. landmarks are: tensor([[[ 5.7625e-01, -3.8397e-01, 1.7268e+00, 2.6220e-01, -4.2102e-01,
  30537. 1.3133e-01, 4.2771e-01, 3.0069e-01],
  30538. [ 4.9971e-01, -4.4465e-01, 1.1610e+00, -9.7721e-01, -6.0577e-01,
  30539. -1.0311e+00, 1.4038e-01, -1.0312e-01],
  30540. [ 6.5365e-01, -3.9601e-01, 1.6517e+00, 3.1609e-01, -4.9607e-01,
  30541. 4.6189e-04, 7.6203e-01, 1.5543e-01],
  30542. [ 6.4542e-01, -3.7194e-01, 1.9115e+00, -1.3041e-01, -5.0762e-01,
  30543. -2.8437e-01, 1.0033e+00, 4.3864e-01],
  30544. [ 5.8834e-01, -3.5935e-01, 1.7557e+00, 2.5450e-01, -4.1524e-01,
  30545. -6.1124e-02, 3.3533e-01, 3.0069e-01],
  30546. [-2.2859e+00, -2.2859e+00, 1.5767e+00, -7.5396e-01, -6.4042e-01,
  30547. -7.3087e-01, 1.7534e-01, 8.9251e-02],
  30548. [ 6.2566e-01, -4.2731e-01, 1.8365e+00, -6.8822e-02, -4.6721e-01,
  30549. -6.1124e-02, 1.1715e+00, 1.6077e-01],
  30550. [ 5.2494e-01, -4.4734e-01, 1.7326e+00, -9.1917e-02, -2.0162e-01,
  30551. 1.5443e-01, 1.7328e-01, 4.1158e-02]]], device='cuda:0')
  30552. loss_train_step before backward: tensor(0.0114, device='cuda:0', grad_fn=<MseLossBackward>)
  30553. loss_train_step after backward: tensor(0.0114, device='cuda:0', grad_fn=<MseLossBackward>)
  30554. loss_train: 1.7420769152231514
  30555. step: 89
  30556. running loss: 0.019573897923855634
  30557. Train Steps: 89/90 Loss: 0.0196 torch.Size([8, 600, 800])
  30558. torch.Size([8, 8])
  30559. tensor([[0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960],
  30560. [0.6147, 0.4026, 0.6600, 0.2467, 0.4088, 0.2150, 0.5489, 0.5773],
  30561. [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100],
  30562. [0.6339, 0.4102, 0.9088, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390],
  30563. [0.6147, 0.4107, 0.8137, 0.3333, 0.3750, 0.2683, 0.5006, 0.5412],
  30564. [0.6262, 0.4052, 0.8888, 0.4700, 0.3675, 0.5117, 0.6350, 0.5233],
  30565. [0.6260, 0.4214, 0.8538, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983],
  30566. [0.6310, 0.4017, 0.8563, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006]],
  30567. device='cuda:0', dtype=torch.float64)
  30568. predictions are: tensor([[ 0.4671, -0.5049, 1.8774, -0.1466, -0.0864, -0.0608, 0.2093, 0.0941],
  30569. [ 0.5628, -0.4053, 0.9370, -1.2464, -0.2642, -1.3898, 0.2510, 0.3911],
  30570. [ 0.7089, -0.3867, 1.8857, -0.3630, -0.5566, -0.5843, 0.7402, 0.1295],
  30571. [ 0.5594, -0.4493, 1.9432, -0.0785, -0.3768, 0.2006, 1.0250, 0.2871],
  30572. [ 0.0606, -0.7394, 1.5011, -0.7235, -0.4577, -1.0457, 0.0626, 0.1358],
  30573. [ 0.7851, -0.2815, 1.9567, -0.1113, -0.4422, 0.0439, 0.6305, 0.0925],
  30574. [ 0.4946, -0.4309, 1.6859, 0.3787, -0.4591, -0.5892, 0.3183, 0.4996],
  30575. [ 0.5189, -0.5028, 1.7568, 0.4543, -0.5045, -0.1441, 0.6245, 0.0330]],
  30576. device='cuda:0', grad_fn=<AddmmBackward>)
  30577. landmarks are: tensor([[[ 0.5205, -0.4577, 1.7326, -0.1304, -0.1862, 0.0447, 0.1301,
  30578. 0.0051],
  30579. [ 0.5532, -0.4264, 0.7626, -1.1466, -0.3979, -1.2928, 0.2494,
  30580. 0.3808],
  30581. [ 0.6220, -0.4436, 1.8711, -0.3691, -0.6173, -0.4999, 0.6702,
  30582. 0.0697],
  30583. [ 0.6421, -0.3912, 1.9115, -0.0842, -0.4730, 0.1544, 1.1824,
  30584. 0.2035],
  30585. [ 0.5532, -0.3888, 1.4727, -0.7463, -0.5538, -1.0465, 0.0265,
  30586. 0.2138],
  30587. [ 0.6063, -0.4142, 1.8192, -0.1150, -0.5885, 0.0774, 0.6471,
  30588. 0.1313],
  30589. [ 0.6055, -0.3393, 1.6575, 0.2545, -0.5942, -0.5461, 0.2949,
  30590. 0.4778],
  30591. [ 0.6286, -0.4303, 1.6691, 0.3931, -0.5827, -0.0919, 0.6359,
  30592. 0.0261]]], device='cuda:0')
  30593. loss_train_step before backward: tensor(0.0123, device='cuda:0', grad_fn=<MseLossBackward>)
  30594. loss_train_step after backward: tensor(0.0123, device='cuda:0', grad_fn=<MseLossBackward>)
  30595. loss_train: 1.7543703555129468
  30596. step: 90
  30597. running loss: 0.019493003950143854
  30598. Valid Steps: 10/10 Loss: nan 3.8413
  30599. --------------------------------------------------
  30600. Epoch: 8 Train Loss: 0.0195 Valid Loss: nan
  30601. --------------------------------------------------
  30602. size of train loader is: 90
  30603. torch.Size([8, 600, 800])
  30604. torch.Size([8, 8])
  30605. tensor([[0.6205, 0.4062, 0.8337, 0.2683, 0.3675, 0.4283, 0.6338, 0.5250],
  30606. [0.6204, 0.4049, 0.7975, 0.2700, 0.3937, 0.2567, 0.5700, 0.5183],
  30607. [0.6135, 0.3994, 0.7913, 0.3050, 0.3625, 0.3050, 0.5837, 0.5050],
  30608. [0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967],
  30609. [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402],
  30610. [0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297],
  30611. [0.6353, 0.4128, 0.8488, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540],
  30612. [0.6127, 0.4115, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495]],
  30613. device='cuda:0', dtype=torch.float64)
  30614. predictions are: tensor([[ 0.6140, -0.3871, 1.6023, -0.7491, -0.5414, -0.3933, 0.5851, 0.2746],
  30615. [ 0.5138, -0.4631, 1.5303, -0.9427, -0.4560, -1.2151, 0.2397, 0.0554],
  30616. [ 0.4892, -0.4673, 1.4873, -0.7423, -0.5505, -0.9114, 0.4542, 0.1094],
  30617. [ 0.6205, -0.3850, 1.7838, -0.1074, -0.5529, -0.0047, 0.3428, 0.1468],
  30618. [ 0.6016, -0.4252, 1.7565, 0.4326, -0.3560, -0.1970, 0.4493, 0.1951],
  30619. [ 0.7335, -0.3303, 1.8009, 0.5891, -0.5046, -0.0271, 0.7923, 0.1182],
  30620. [-0.4382, -1.0592, 1.7135, -1.0684, 0.3674, -1.3955, 0.8355, 0.2889],
  30621. [ 0.4659, -0.4597, 1.1708, -0.8018, -0.5448, -0.9428, 0.1353, 0.3279]],
  30622. device='cuda:0', grad_fn=<AddmmBackward>)
  30623. landmarks are: tensor([[[ 5.7991e-01, -4.0985e-01, 1.5651e+00, -1.0465e+00, -5.8845e-01,
  30624. -3.0747e-01, 6.4134e-01, 1.3903e-01],
  30625. [ 5.7985e-01, -4.1555e-01, 1.3977e+00, -1.0388e+00, -4.6721e-01,
  30626. -1.1004e+00, 3.4688e-01, 1.0824e-01],
  30627. [ 5.4769e-01, -4.4126e-01, 1.3688e+00, -8.7714e-01, -6.1155e-01,
  30628. -8.7714e-01, 4.1039e-01, 4.6651e-02],
  30629. [ 4.9740e-01, -4.4819e-01, 1.6633e+00, -3.3056e-01, -6.1732e-01,
  30630. 1.3133e-01, 2.9255e-01, 8.0947e-03],
  30631. [ 5.7800e-01, -4.5651e-01, 1.6221e+00, 2.5323e-01, -3.7281e-01,
  30632. -1.7182e-01, 4.3570e-01, 2.0910e-01],
  30633. [ 6.5036e-01, -3.7434e-01, 1.6171e+00, 4.3156e-01, -5.4226e-01,
  30634. 4.6189e-04, 7.1085e-01, 1.6077e-01],
  30635. [ 6.4871e-01, -3.7916e-01, 1.6344e+00, -1.0850e+00, 2.6592e-01,
  30636. -1.5397e+00, 8.0590e-01, 2.7299e-01],
  30637. [ 5.4417e-01, -3.8545e-01, 1.0224e+00, -9.5412e-01, -6.1155e-01,
  30638. -9.2333e-01, 1.7452e-01, 2.5215e-01]]], device='cuda:0')
  30639. loss_train_step before backward: tensor(0.0350, device='cuda:0', grad_fn=<MseLossBackward>)
  30640. loss_train_step after backward: tensor(0.0350, device='cuda:0', grad_fn=<MseLossBackward>)
  30641. loss_train: 0.035023126751184464
  30642. step: 1
  30643. running loss: 0.035023126751184464
  30644. Train Steps: 1/90 Loss: 0.0350 torch.Size([8, 600, 800])
  30645. torch.Size([8, 8])
  30646. tensor([[ nan, nan, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621],
  30647. [0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222],
  30648. [0.6265, 0.4091, 0.8950, 0.3533, 0.3600, 0.3967, 0.6295, 0.4901],
  30649. [0.6201, 0.3970, 0.8413, 0.4950, 0.4413, 0.5183, 0.6088, 0.5400],
  30650. [ nan, nan, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733],
  30651. [0.6187, 0.4104, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683],
  30652. [0.6048, 0.3928, 0.8538, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719],
  30653. [0.6289, 0.4081, 0.8720, 0.3487, 0.3900, 0.3183, 0.6703, 0.5376]],
  30654. device='cuda:0', dtype=torch.float64)
  30655. predictions are: tensor([[-1.5893, -1.8049, 1.6660, -0.9662, 0.1232, -1.2444, 0.7696, 0.2287],
  30656. [ 0.6649, -0.2618, 1.4795, -0.3987, -0.1274, -1.2144, 0.1891, 0.4023],
  30657. [ 0.8279, -0.1916, 1.8551, -0.5088, -0.6687, -0.4753, 0.5482, -0.0837],
  30658. [ 0.6262, -0.3445, 1.6581, 0.1843, -0.3398, 0.0883, 0.3233, 0.1184],
  30659. [-2.2111, -2.2507, 0.9937, -1.0590, -0.2451, -1.3357, 0.2370, 0.3124],
  30660. [ 0.7010, -0.3028, 1.0150, -1.1334, -0.4862, -1.0880, 0.4025, 0.2193],
  30661. [ 0.6674, -0.3327, 1.6469, 0.4851, -0.5619, 0.0573, 0.2066, 0.0539],
  30662. [ 0.6414, -0.3664, 1.7301, -0.4614, -0.4897, -0.7804, 0.7746, 0.1199]],
  30663. device='cuda:0', grad_fn=<AddmmBackward>)
  30664. landmarks are: tensor([[[-2.2859e+00, -2.2859e+00, 1.7210e+00, -9.7721e-01, 1.8522e-01,
  30665. -1.3698e+00, 7.9859e-01, 3.1039e-01],
  30666. [ 6.2367e-01, -2.9831e-01, 1.3919e+00, -4.6913e-01, -4.5727e-02,
  30667. -1.2313e+00, 2.4525e-01, 5.8821e-01],
  30668. [ 6.0774e-01, -3.9646e-01, 1.8480e+00, -6.5389e-01, -6.2309e-01,
  30669. -4.5373e-01, 6.2155e-01, -2.1963e-02],
  30670. [ 5.7812e-01, -4.5219e-01, 1.5998e+00, 4.6189e-04, -2.4781e-01,
  30671. 1.0824e-01, 5.2587e-01, 2.0831e-01],
  30672. [-2.2859e+00, -2.2859e+00, 7.1062e-01, -1.4468e+00, -2.8822e-01,
  30673. -1.4237e+00, 2.4296e-01, 3.6228e-01],
  30674. [ 5.7182e-01, -3.9053e-01, 1.0053e+00, -1.3305e+00, -4.6143e-01,
  30675. -1.1235e+00, 4.4503e-01, 3.3918e-01],
  30676. [ 5.0785e-01, -4.7144e-01, 1.6575e+00, 2.2371e-01, -4.9607e-01,
  30677. 7.7444e-02, 1.4655e-01, -1.0613e-01],
  30678. [ 6.1907e-01, -4.0082e-01, 1.7420e+00, -6.7528e-01, -4.8453e-01,
  30679. -8.1555e-01, 8.1006e-01, 1.9744e-01]]], device='cuda:0')
  30680. loss_train_step before backward: tensor(0.0246, device='cuda:0', grad_fn=<MseLossBackward>)
  30681. loss_train_step after backward: tensor(0.0246, device='cuda:0', grad_fn=<MseLossBackward>)
  30682. loss_train: 0.05961088836193085
  30683. step: 2
  30684. running loss: 0.029805444180965424
  30685.  
  30686. Train Steps: 2/90 Loss: 0.0298 torch.Size([8, 600, 800])
  30687. torch.Size([8, 8])
  30688. tensor([[0.6136, 0.4117, 0.8700, 0.5167, 0.4188, 0.5083, 0.5147, 0.5495],
  30689. [0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938],
  30690. [0.6259, 0.4156, 0.8812, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960],
  30691. [0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400],
  30692. [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467],
  30693. [0.6141, 0.4038, 0.8650, 0.4833, 0.4839, 0.5176, 0.5787, 0.5600],
  30694. [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413],
  30695. [0.6113, 0.4006, 0.8700, 0.5350, 0.3638, 0.3767, 0.5097, 0.4882]],
  30696. device='cuda:0', dtype=torch.float64)
  30697. predictions are: tensor([[ 0.3977, -0.5190, 1.6397, 0.2275, -0.4261, 0.0427, 0.1332, 0.2363],
  30698. [ 0.5997, -0.4350, 1.6313, 0.4328, -0.5913, -0.0857, 0.5645, -0.0069],
  30699. [ 0.7061, -0.3038, 1.7128, -0.8875, -0.1318, -1.3337, 0.5031, 0.0823],
  30700. [ 0.4596, -0.4776, 1.4336, -1.1474, -0.1885, -1.2794, 0.5792, 0.2321],
  30701. [ 0.5913, -0.3999, 1.4971, -0.7091, -0.7462, -0.3909, 0.3259, 0.3171],
  30702. [ 0.2882, -0.5980, 1.6730, -0.0155, -0.1775, 0.0260, 0.2027, 0.3243],
  30703. [ 0.4600, -0.5590, 1.9039, -0.2497, -0.3810, -0.9030, 1.0014, 0.1673],
  30704. [ 0.2536, -0.6194, 1.6788, 0.1859, -0.7088, -0.5439, 0.0605, 0.0532]],
  30705. device='cuda:0', grad_fn=<AddmmBackward>)
  30706. landmarks are: tensor([[[ 0.5482, -0.3841, 1.7326, 0.1005, -0.3517, 0.0620, 0.0912,
  30707. 0.2522],
  30708. [ 0.6395, -0.4213, 1.7037, 0.3623, -0.4326, -0.1073, 0.6560,
  30709. -0.0049],
  30710. [ 0.6052, -0.3663, 1.7845, -0.8156, -0.0804, -1.4237, 0.5866,
  30711. 0.0051],
  30712. [ 0.5958, -0.3818, 1.5536, -1.1081, -0.2074, -1.3390, 0.5663,
  30713. 0.2083],
  30714. [ 0.5484, -0.4591, 1.5940, -0.7771, -0.6693, -0.3460, 0.4219,
  30715. 0.2391],
  30716. [ 0.5505, -0.4207, 1.7095, -0.0534, -0.0509, 0.1050, 0.3873,
  30717. 0.3007],
  30718. [ 0.6224, -0.4345, 1.9404, -0.2921, -0.3171, -0.8771, 1.0655,
  30719. 0.2142],
  30720. [ 0.5379, -0.4358, 1.7326, 0.1852, -0.6058, -0.5461, 0.0684,
  30721. -0.0310]]], device='cuda:0')
  30722. loss_train_step before backward: tensor(0.0095, device='cuda:0', grad_fn=<MseLossBackward>)
  30723. loss_train_step after backward: tensor(0.0095, device='cuda:0', grad_fn=<MseLossBackward>)
  30724. loss_train: 0.06913623213768005
  30725. step: 3
  30726. running loss: 0.023045410712560017
  30727. Train Steps: 3/90 Loss: 0.0230 torch.Size([8, 600, 800])
  30728. torch.Size([8, 8])
  30729. tensor([[0.6273, 0.4143, 0.8750, 0.5700, 0.3987, 0.4717, 0.6013, 0.5467],
  30730. [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383],
  30731. [0.6128, 0.4115, 0.7934, 0.3778, 0.3450, 0.4033, 0.5337, 0.5456],
  30732. [ nan, nan, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621],
  30733. [ nan, nan, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563],
  30734. [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153],
  30735. [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267],
  30736. [0.6198, 0.4130, 0.8762, 0.4117, 0.3650, 0.4900, 0.5707, 0.5103]],
  30737. device='cuda:0', dtype=torch.float64)
  30738. predictions are: tensor([[ 0.7372, -0.2586, 1.7167, 0.3747, -0.4748, -0.1449, 0.4796, 0.0674],
  30739. [ 0.8608, -0.2171, 1.2050, -1.0956, -0.1502, -1.5177, 0.4047, 0.2153],
  30740. [ 0.5684, -0.3655, 1.4035, -0.4764, -0.7140, -0.4578, 0.2291, 0.2330],
  30741. [-1.4736, -1.7437, 1.6607, -1.0372, 0.2038, -1.3720, 0.9044, 0.2716],
  30742. [-2.0843, -2.1717, 1.3421, -0.6876, -0.6274, -0.8590, 0.1328, 0.2106],
  30743. [ 0.6394, -0.3625, 1.4053, -0.7157, -0.6669, -0.7526, 0.3336, 0.1017],
  30744. [ 0.7157, -0.3242, 1.5919, 0.4262, -0.4433, -0.1878, 0.3655, 0.0658],
  30745. [ 0.7816, -0.2302, 1.8276, -0.3053, -0.6136, -0.1140, 0.3521, 0.0361]],
  30746. device='cuda:0', grad_fn=<AddmmBackward>)
  30747. landmarks are: tensor([[[ 0.6115, -0.3724, 1.7557, 0.3469, -0.4441, -0.1073, 0.4912,
  30748. 0.2391],
  30749. [ 0.5911, -0.3984, 1.1956, -1.0850, -0.0804, -1.5392, 0.4393,
  30750. 0.2006],
  30751. [ 0.5444, -0.3852, 1.3786, -0.5409, -0.6924, -0.4229, 0.1791,
  30752. 0.2341],
  30753. [-2.2859, -2.2859, 1.7210, -0.9772, 0.1852, -1.3698, 0.7986,
  30754. 0.3104],
  30755. [-2.2859, -2.2859, 1.4006, -0.8105, -0.6115, -0.8232, 0.0419,
  30756. 0.2837],
  30757. [ 0.5614, -0.4514, 1.3252, -0.7240, -0.6237, -0.7215, 0.4299,
  30758. 0.0942],
  30759. [ 0.5891, -0.4550, 1.5132, 0.3546, -0.3691, -0.1535, 0.3815,
  30760. 0.1467],
  30761. [ 0.5771, -0.3784, 1.7614, -0.3844, -0.6000, -0.0226, 0.3501,
  30762. 0.0712]]], device='cuda:0')
  30763. loss_train_step before backward: tensor(0.0222, device='cuda:0', grad_fn=<MseLossBackward>)
  30764. loss_train_step after backward: tensor(0.0222, device='cuda:0', grad_fn=<MseLossBackward>)
  30765. loss_train: 0.0913512222468853
  30766. step: 4
  30767. running loss: 0.022837805561721325
  30768. Train Steps: 4/90 Loss: 0.0228 torch.Size([8, 600, 800])
  30769. torch.Size([8, 8])
  30770. tensor([[0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407],
  30771. [0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133],
  30772. [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377],
  30773. [0.6079, 0.3964, 0.7420, 0.2958, 0.3563, 0.2917, 0.5351, 0.4980],
  30774. [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834],
  30775. [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350],
  30776. [0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494],
  30777. [0.6182, 0.3998, 0.8793, 0.4191, 0.3552, 0.4285, 0.6038, 0.5312]],
  30778. device='cuda:0', dtype=torch.float64)
  30779. predictions are: tensor([[ 0.3963, -0.5111, 1.5654, -0.2544, -0.6555, -0.8049, -0.0029, 0.2036],
  30780. [ 0.6896, -0.3829, 1.4768, -0.9736, -0.0825, -1.4334, 0.5826, 0.1867],
  30781. [ 0.5362, -0.5103, 1.6700, 0.3681, -0.4911, -0.0497, 0.4941, 0.1507],
  30782. [ 0.2169, -0.6776, 1.1213, -0.8193, -0.6600, -0.8720, 0.0790, 0.0277],
  30783. [ 0.5976, -0.4572, 1.3894, -1.2061, -0.0525, -1.4916, 0.4835, 0.0843],
  30784. [ 0.4290, -0.4816, 1.7752, -0.1858, -0.6380, -0.3565, 0.2010, 0.2168],
  30785. [ 0.3762, -0.6093, 1.8550, -0.4568, -0.4097, -0.4039, 1.0037, 0.2467],
  30786. [ 0.5470, -0.4521, 1.7105, -0.2674, -0.6784, -0.2623, 0.4499, 0.1900]],
  30787. device='cuda:0', grad_fn=<AddmmBackward>)
  30788. landmarks are: tensor([[[ 0.5470, -0.3952, 1.6377, -0.4253, -0.6289, -0.8079, 0.0249,
  30789. 0.2116],
  30790. [ 0.6085, -0.4084, 1.5536, -1.1466, -0.0746, -1.4853, 0.6298,
  30791. 0.0851],
  30792. [ 0.5775, -0.4607, 1.6741, 0.1962, -0.4036, -0.1212, 0.4588,
  30793. 0.1979],
  30794. [ 0.5219, -0.4550, 1.1415, -0.9196, -0.6404, -0.9387, 0.1856,
  30795. 0.0141],
  30796. [ 0.6126, -0.4057, 1.4439, -1.3159, -0.1150, -1.5777, 0.5537,
  30797. -0.0530],
  30798. [ 0.6072, -0.3239, 1.8423, -0.3537, -0.4961, -0.3921, 0.2083,
  30799. 0.1852],
  30800. [ 0.6388, -0.3792, 1.9635, -0.6616, -0.2536, -0.5153, 1.1605,
  30801. 0.2516],
  30802. [ 0.5697, -0.4393, 1.7754, -0.3503, -0.6453, -0.3067, 0.5028,
  30803. 0.1677]]], device='cuda:0')
  30804. loss_train_step before backward: tensor(0.0120, device='cuda:0', grad_fn=<MseLossBackward>)
  30805. loss_train_step after backward: tensor(0.0120, device='cuda:0', grad_fn=<MseLossBackward>)
  30806. loss_train: 0.10337931476533413
  30807. step: 5
  30808. running loss: 0.020675862953066827
  30809. Train Steps: 5/90 Loss: 0.0207 torch.Size([8, 600, 800])
  30810. torch.Size([8, 8])
  30811. tensor([[0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329],
  30812. [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650],
  30813. [0.6149, 0.4054, 0.6713, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695],
  30814. [0.6199, 0.4102, 0.8950, 0.4417, 0.4012, 0.5367, 0.6112, 0.5967],
  30815. [0.6266, 0.4067, 0.8588, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267],
  30816. [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098],
  30817. [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459],
  30818. [0.6209, 0.3920, 0.8650, 0.5367, 0.4400, 0.5067, 0.6025, 0.4950]],
  30819. device='cuda:0', dtype=torch.float64)
  30820. predictions are: tensor([[ 0.5466, -0.4835, 1.6164, -0.6194, -0.7410, -0.6682, 0.3838, 0.1722],
  30821. [ 0.4471, -0.5391, 1.2867, -0.9394, -0.5697, -1.0218, 0.1520, 0.2237],
  30822. [ 0.1865, -0.6866, 0.8140, -1.0698, -0.4618, -1.3994, -0.0308, 0.2866],
  30823. [ 0.4823, -0.4738, 1.8417, -0.0641, -0.4494, 0.1759, 0.4307, 0.3456],
  30824. [ 0.5742, -0.4413, 1.7019, -0.8798, -0.3884, -1.1352, 0.6841, 0.1134],
  30825. [ 0.5016, -0.4895, 1.7512, 0.0568, -0.5928, 0.0645, 0.3931, 0.1011],
  30826. [ 0.5847, -0.4574, 1.7197, -0.7646, -0.0677, -1.3968, 0.8645, 0.1314],
  30827. [ 0.3921, -0.5961, 1.7098, 0.3133, -0.3360, 0.0101, 0.3127, -0.0480]],
  30828. device='cuda:0', grad_fn=<AddmmBackward>)
  30829. landmarks are: tensor([[[ 5.6801e-01, -4.3934e-01, 1.5920e+00, -6.6715e-01, -6.4527e-01,
  30830. -5.4566e-01, 5.1492e-01, 1.7534e-01],
  30831. [ 5.3204e-01, -4.1886e-01, 1.3053e+00, -1.0773e+00, -5.7113e-01,
  30832. -9.8491e-01, 2.2674e-01, 3.2370e-01],
  30833. [ 5.5445e-01, -4.1332e-01, 8.1455e-01, -1.2082e+00, -4.2679e-01,
  30834. -1.3544e+00, 1.2208e-01, 3.4458e-01],
  30835. [ 5.7719e-01, -3.9130e-01, 1.8480e+00, -2.4588e-01, -4.3256e-01,
  30836. 1.9292e-01, 5.3741e-01, 4.7005e-01],
  30837. [ 6.0837e-01, -4.0762e-01, 1.6806e+00, -9.6182e-01, -2.9977e-01,
  30838. -9.6952e-01, 6.3557e-01, 1.4673e-01],
  30839. [ 5.4660e-01, -4.7064e-01, 1.7198e+00, -9.0292e-02, -5.7125e-01,
  30840. 1.2613e-01, 4.7328e-01, 6.8827e-02],
  30841. [ 6.5201e-01, -3.9120e-01, 1.7095e+00, -9.0793e-01, -2.8406e-02,
  30842. -1.3621e+00, 8.0956e-01, 2.3558e-01],
  30843. [ 5.8199e-01, -4.7544e-01, 1.7095e+00, 1.9292e-01, -2.5358e-01,
  30844. 5.4350e-02, 4.9700e-01, 4.6189e-04]]], device='cuda:0')
  30845. loss_train_step before backward: tensor(0.0106, device='cuda:0', grad_fn=<MseLossBackward>)
  30846. loss_train_step after backward: tensor(0.0106, device='cuda:0', grad_fn=<MseLossBackward>)
  30847. loss_train: 0.11402785126119852
  30848. step: 6
  30849. running loss: 0.01900464187686642
  30850.  
  30851. Train Steps: 6/90 Loss: 0.0190 torch.Size([8, 600, 800])
  30852. torch.Size([8, 8])
  30853. tensor([[0.6276, 0.4235, 0.8888, 0.5333, 0.3800, 0.3117, 0.5427, 0.6164],
  30854. [0.6138, 0.4054, 0.8750, 0.4750, 0.4363, 0.5017, 0.5086, 0.5822],
  30855. [0.6055, 0.4015, 0.7425, 0.2033, 0.4113, 0.1883, 0.5217, 0.4823],
  30856. [0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461],
  30857. [0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320],
  30858. [0.6213, 0.4131, 0.8438, 0.3550, 0.3513, 0.4400, 0.5716, 0.5123],
  30859. [ nan, nan, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533],
  30860. [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268]],
  30861. device='cuda:0', dtype=torch.float64)
  30862. predictions are: tensor([[ 0.5891, -0.3890, 1.6729, 0.2073, -0.4570, -0.8302, 0.3594, 0.4329],
  30863. [ 0.4704, -0.4707, 1.7011, -0.0808, -0.2426, -0.0565, 0.0563, 0.1839],
  30864. [ 0.4041, -0.5540, 1.0145, -1.2279, -0.3703, -1.4565, 0.2284, -0.0221],
  30865. [ 0.5034, -0.4659, 1.5339, -0.5810, -0.6889, -0.4715, 0.2772, 0.1573],
  30866. [ 0.5466, -0.4884, 1.7959, -0.1409, -0.5534, -0.4282, 0.8416, 0.0456],
  30867. [ 0.6808, -0.3508, 1.6525, -0.6522, -0.6313, -0.4027, 0.3507, 0.0792],
  30868. [-2.1821, -2.2742, 1.3833, -0.8228, -0.5880, -1.0219, 0.2936, 0.2045],
  30869. [ 0.6380, -0.4148, 1.7247, -0.0332, -0.2073, 0.0171, 0.6617, 0.0978]],
  30870. device='cuda:0', grad_fn=<AddmmBackward>)
  30871. landmarks are: tensor([[[ 0.6131, -0.3299, 1.8192, 0.1775, -0.5307, -0.8463, 0.2206,
  30872. 0.5612],
  30873. [ 0.5491, -0.4132, 1.7557, -0.0919, -0.2709, 0.0313, 0.0635,
  30874. 0.4032],
  30875. [ 0.5109, -0.4316, 1.1436, -1.3467, -0.3864, -1.4160, 0.1239,
  30876. -0.0580],
  30877. [ 0.5124, -0.4483, 1.5420, -0.6231, -0.7155, -0.4152, 0.2679,
  30878. 0.2365],
  30879. [ 0.6174, -0.4490, 1.8885, -0.0996, -0.4845, -0.3691, 0.9814,
  30880. 0.1715],
  30881. [ 0.5841, -0.3778, 1.6113, -0.6462, -0.6635, -0.2536, 0.3542,
  30882. 0.0802],
  30883. [-2.2859, -2.2859, 1.5074, -1.0388, -0.5423, -0.9849, 0.2199,
  30884. 0.2699],
  30885. [ 0.5828, -0.4682, 1.7031, -0.0497, -0.2458, 0.0818, 0.6381,
  30886. 0.1474]]], device='cuda:0')
  30887. loss_train_step before backward: tensor(0.0065, device='cuda:0', grad_fn=<MseLossBackward>)
  30888. loss_train_step after backward: tensor(0.0065, device='cuda:0', grad_fn=<MseLossBackward>)
  30889. loss_train: 0.12055541574954987
  30890. step: 7
  30891. running loss: 0.017222202249935696
  30892. Train Steps: 7/90 Loss: 0.0172 torch.Size([8, 600, 800])
  30893. torch.Size([8, 8])
  30894. tensor([[0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973],
  30895. [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483],
  30896. [0.6091, 0.3997, 0.8314, 0.4334, 0.3788, 0.4550, 0.5213, 0.5656],
  30897. [0.6135, 0.3994, 0.7913, 0.3050, 0.3625, 0.3050, 0.5837, 0.5050],
  30898. [0.6182, 0.3972, 0.8552, 0.5914, 0.3683, 0.4181, 0.5688, 0.5378],
  30899. [0.6222, 0.3937, 0.8350, 0.5617, 0.4138, 0.4600, 0.5800, 0.5233],
  30900. [0.6145, 0.4007, 0.8775, 0.4533, 0.4562, 0.5533, 0.6088, 0.5533],
  30901. [0.6198, 0.4130, 0.8762, 0.4117, 0.3650, 0.4900, 0.5707, 0.5103]],
  30902. device='cuda:0', dtype=torch.float64)
  30903. predictions are: tensor([[ 0.6121, -0.4770, 1.8885, 0.0605, -0.6429, -0.2336, 0.6899, 0.0278],
  30904. [ 0.7454, -0.3639, 1.0604, -1.3949, -0.4047, -1.2772, 0.4862, 0.2483],
  30905. [ 0.2872, -0.6198, 1.6279, -0.3692, -0.5507, -0.2969, 0.1379, 0.3256],
  30906. [ 0.3327, -0.6100, 1.4320, -0.9929, -0.6452, -0.9560, 0.4174, 0.0762],
  30907. [ 0.7374, -0.3535, 1.6367, 0.2674, -0.6135, -0.4852, 0.2430, 0.1632],
  30908. [ 0.5788, -0.5004, 1.6847, 0.2857, -0.3745, -0.2482, 0.4787, 0.2059],
  30909. [ 0.5147, -0.4656, 1.8758, -0.2675, -0.1389, 0.0108, 0.3955, 0.3097],
  30910. [ 0.5872, -0.4207, 1.8811, -0.4503, -0.5698, -0.2127, 0.3556, 0.0867]],
  30911. device='cuda:0', grad_fn=<AddmmBackward>)
  30912. landmarks are: tensor([[[ 0.6240, -0.4321, 1.8423, 0.1852, -0.5885, -0.1612, 0.6962,
  30913. 0.0111],
  30914. [ 0.5784, -0.4085, 1.0859, -1.3929, -0.4037, -1.1158, 0.5605,
  30915. 0.2468],
  30916. [ 0.5275, -0.4396, 1.5543, -0.2841, -0.5365, -0.1843, 0.1221,
  30917. 0.3265],
  30918. [ 0.5477, -0.4413, 1.3688, -0.8771, -0.6115, -0.8771, 0.4104,
  30919. 0.0467],
  30920. [ 0.5697, -0.4514, 1.6642, 0.4459, -0.5850, -0.3546, 0.3414,
  30921. 0.1982],
  30922. [ 0.5880, -0.4676, 1.5709, 0.3084, -0.3748, -0.1612, 0.3931,
  30923. 0.1313],
  30924. [ 0.5526, -0.4351, 1.7672, -0.1920, -0.1785, 0.2699, 0.5259,
  30925. 0.2699],
  30926. [ 0.5771, -0.3784, 1.7614, -0.3844, -0.6000, -0.0226, 0.3501,
  30927. 0.0712]]], device='cuda:0')
  30928. loss_train_step before backward: tensor(0.0095, device='cuda:0', grad_fn=<MseLossBackward>)
  30929. loss_train_step after backward: tensor(0.0095, device='cuda:0', grad_fn=<MseLossBackward>)
  30930. loss_train: 0.1300251130014658
  30931. step: 8
  30932. running loss: 0.016253139125183225
  30933. Train Steps: 8/90 Loss: 0.0163 torch.Size([8, 600, 800])
  30934. torch.Size([8, 8])
  30935. tensor([[0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600],
  30936. [0.6200, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183],
  30937. [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901],
  30938. [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683],
  30939. [0.6182, 0.3972, 0.8552, 0.5914, 0.3683, 0.4181, 0.5688, 0.5378],
  30940. [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841],
  30941. [0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033],
  30942. [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320]],
  30943. device='cuda:0', dtype=torch.float64)
  30944. predictions are: tensor([[ 0.5256, -0.4524, 1.7984, 0.1450, -0.4289, -0.0788, 0.2589, 0.3393],
  30945. [ 0.4977, -0.4804, 1.3995, -1.0820, -0.5870, -1.0135, 0.3227, 0.1081],
  30946. [ 0.6630, -0.3975, 1.3779, -1.1074, -0.6264, -1.1621, 0.1380, -0.0220],
  30947. [ 0.5724, -0.4312, 1.7310, -0.1963, -0.0846, -0.1130, 0.1525, 0.2617],
  30948. [ 0.7592, -0.3167, 1.5519, 0.1764, -0.6423, -0.4017, 0.1510, 0.1811],
  30949. [ 0.4704, -0.5583, 1.8519, -0.4481, -0.4448, -0.6366, 0.9631, 0.3323],
  30950. [ 0.4849, -0.5122, 1.7827, -0.0995, -0.3785, 0.1170, 0.3145, 0.1204],
  30951. [ 0.3807, -0.6192, 1.7006, 0.0377, -0.5237, 0.0271, 0.9367, 0.1956]],
  30952. device='cuda:0', grad_fn=<AddmmBackward>)
  30953. landmarks are: tensor([[[ 0.5883, -0.3594, 1.7557, 0.2545, -0.4152, -0.0611, 0.3353,
  30954. 0.3007],
  30955. [ 0.5781, -0.4129, 1.4142, -0.9057, -0.5115, -0.9937, 0.4621,
  30956. 0.1080],
  30957. [ 0.5515, -0.4129, 1.4785, -1.0080, -0.5192, -1.1004, 0.1034,
  30958. -0.0220],
  30959. [ 0.5359, -0.4370, 1.7095, -0.0303, -0.0804, -0.0380, 0.1044,
  30960. 0.3392],
  30961. [ 0.5697, -0.4514, 1.6642, 0.4459, -0.5850, -0.3546, 0.3414,
  30962. 0.1982],
  30963. [ 0.6438, -0.3936, 2.0039, -0.2690, -0.2940, -0.6231, 1.0618,
  30964. 0.4119],
  30965. [ 0.5668, -0.4306, 1.7730, 0.0697, -0.4037, 0.1390, 0.4450,
  30966. 0.0390],
  30967. [ 0.6273, -0.4393, 1.6402, 0.1313, -0.5076, 0.0467, 1.1532,
  30968. 0.1715]]], device='cuda:0')
  30969. loss_train_step before backward: tensor(0.0117, device='cuda:0', grad_fn=<MseLossBackward>)
  30970. loss_train_step after backward: tensor(0.0117, device='cuda:0', grad_fn=<MseLossBackward>)
  30971. loss_train: 0.14171226508915424
  30972. step: 9
  30973. running loss: 0.015745807232128248
  30974. Train Steps: 9/90 Loss: 0.0157 torch.Size([8, 600, 800])
  30975. torch.Size([8, 8])
  30976. tensor([[0.6213, 0.4131, 0.8438, 0.3550, 0.3513, 0.4400, 0.5716, 0.5123],
  30977. [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5513, 0.5750],
  30978. [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402],
  30979. [0.6100, 0.4071, 0.7601, 0.3444, 0.3400, 0.4117, 0.5625, 0.5617],
  30980. [0.6218, 0.4137, 0.7263, 0.2233, 0.4075, 0.2650, 0.6212, 0.5783],
  30981. [0.6321, 0.4048, 0.8738, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927],
  30982. [0.6118, 0.4052, 0.8463, 0.3917, 0.3538, 0.3450, 0.5053, 0.5593],
  30983. [0.6317, 0.4038, 0.8287, 0.5900, 0.3800, 0.4717, 0.6295, 0.4986]],
  30984. device='cuda:0', dtype=torch.float64)
  30985. predictions are: tensor([[ 0.5621, -0.4350, 1.8083, -0.7072, -0.6060, -0.3735, 0.3713, 0.1477],
  30986. [-0.2548, -0.9666, 0.9694, -1.2713, -0.2514, -1.5026, 0.2285, 0.4129],
  30987. [ 0.5501, -0.4866, 1.7867, 0.1286, -0.3355, -0.1685, 0.5161, 0.1942],
  30988. [ 0.5518, -0.4300, 1.3952, -0.7849, -0.7246, -0.3439, 0.2739, 0.2838],
  30989. [ 0.6204, -0.4182, 1.1728, -1.2457, -0.4112, -1.0753, 0.5919, 0.3851],
  30990. [ 0.7153, -0.3850, 1.8950, 0.1854, -0.4805, -0.2747, 0.6471, -0.0912],
  30991. [ 0.5189, -0.4356, 1.7161, -0.5475, -0.5983, -0.7178, -0.0889, 0.2347],
  30992. [ 0.6597, -0.4147, 1.7465, 0.3388, -0.5111, -0.1052, 0.6238, 0.0331]],
  30993. device='cuda:0', grad_fn=<AddmmBackward>)
  30994. landmarks are: tensor([[[ 0.5841, -0.3778, 1.6113, -0.6462, -0.6635, -0.2536, 0.3542,
  30995. 0.0802],
  30996. [ 0.5663, -0.4396, 0.8261, -1.1312, -0.2940, -1.3929, 0.2603,
  30997. 0.3700],
  30998. [ 0.5780, -0.4565, 1.6221, 0.2532, -0.3728, -0.1718, 0.4357,
  30999. 0.2091],
  31000. [ 0.5318, -0.4056, 1.2249, -0.6949, -0.7155, -0.3844, 0.3122,
  31001. 0.3084],
  31002. [ 0.5863, -0.3749, 1.0686, -1.2543, -0.4037, -1.0619, 0.5836,
  31003. 0.3854],
  31004. [ 0.6335, -0.4162, 1.7499, 0.3084, -0.4961, -0.2459, 0.6524,
  31005. -0.0102],
  31006. [ 0.5399, -0.4142, 1.6229, -0.4768, -0.6520, -0.6924, 0.0481,
  31007. 0.2972],
  31008. [ 0.6320, -0.4206, 1.5420, 0.4393, -0.5307, -0.1073, 0.6216,
  31009. 0.0171]]], device='cuda:0')
  31010. loss_train_step before backward: tensor(0.0208, device='cuda:0', grad_fn=<MseLossBackward>)
  31011. loss_train_step after backward: tensor(0.0208, device='cuda:0', grad_fn=<MseLossBackward>)
  31012. loss_train: 0.16252397932112217
  31013. step: 10
  31014. running loss: 0.016252397932112216
  31015.  
  31016. Train Steps: 10/90 Loss: 0.0163 torch.Size([8, 600, 800])
  31017. torch.Size([8, 8])
  31018. tensor([[0.6257, 0.4060, 0.8300, 0.2333, 0.4688, 0.2583, 0.7050, 0.5633],
  31019. [0.6022, 0.3994, 0.8025, 0.3350, 0.3350, 0.4400, 0.5565, 0.5025],
  31020. [0.6289, 0.4081, 0.8720, 0.3487, 0.3900, 0.3183, 0.6703, 0.5376],
  31021. [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350],
  31022. [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400],
  31023. [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123],
  31024. [0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350],
  31025. [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717]],
  31026. device='cuda:0', dtype=torch.float64)
  31027. predictions are: tensor([[ 0.4952, -0.5065, 1.6989, -1.0779, -0.1564, -1.1010, 0.9690, 0.3485],
  31028. [ 0.2338, -0.6378, 1.5139, -0.6089, -0.7161, -0.2372, 0.2462, 0.1252],
  31029. [ 0.4851, -0.5297, 1.8211, -0.5048, -0.4673, -0.7272, 0.9266, 0.2146],
  31030. [ 0.4475, -0.4619, 1.8858, -0.2298, -0.6026, -0.3046, 0.2469, 0.2047],
  31031. [ 0.6366, -0.3988, 1.3170, -1.0641, -0.1947, -1.3048, 0.4453, 0.1899],
  31032. [ 0.4931, -0.4868, 1.0749, -1.0042, -0.5348, -0.9385, 0.0530, 0.0562],
  31033. [ 0.4666, -0.4832, 1.4160, -1.0601, -0.1922, -1.2653, 0.4001, 0.2308],
  31034. [ 0.6729, -0.3513, 0.8831, -1.0201, -0.5309, -0.9789, 0.1976, 0.2888]],
  31035. device='cuda:0', grad_fn=<AddmmBackward>)
  31036. landmarks are: tensor([[[ 0.6042, -0.4105, 1.5478, -1.2082, -0.1208, -1.0927, 0.9704,
  31037. 0.3157],
  31038. [ 0.4956, -0.4413, 1.4208, -0.7386, -0.7386, -0.2536, 0.2843,
  31039. 0.0351],
  31040. [ 0.6191, -0.4008, 1.7420, -0.6753, -0.4845, -0.8156, 0.8101,
  31041. 0.1974],
  31042. [ 0.6072, -0.3239, 1.8423, -0.3537, -0.4961, -0.3921, 0.2083,
  31043. 0.1852],
  31044. [ 0.5842, -0.3848, 1.2938, -1.2159, -0.2132, -1.4160, 0.5778,
  31045. 0.2083],
  31046. [ 0.5342, -0.4280, 1.0051, -1.0619, -0.5711, -1.0388, 0.2226,
  31047. 0.0802],
  31048. [ 0.5768, -0.4031, 1.3838, -1.1527, -0.2188, -1.4216, 0.4379,
  31049. 0.1850],
  31050. [ 0.5633, -0.3883, 0.8648, -1.0288, -0.5192, -1.1081, 0.2776,
  31051. 0.3546]]], device='cuda:0')
  31052. loss_train_step before backward: tensor(0.0092, device='cuda:0', grad_fn=<MseLossBackward>)
  31053. loss_train_step after backward: tensor(0.0092, device='cuda:0', grad_fn=<MseLossBackward>)
  31054. loss_train: 0.1717460323125124
  31055. step: 11
  31056. running loss: 0.015613275664773855
  31057. Train Steps: 11/90 Loss: 0.0156 torch.Size([8, 600, 800])
  31058. torch.Size([8, 8])
  31059. tensor([[ nan, nan, 0.6688, 0.2513, 0.4113, 0.2117, 0.5193, 0.5933],
  31060. [0.6193, 0.3930, 0.8949, 0.4437, 0.3852, 0.5435, 0.6263, 0.5263],
  31061. [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927],
  31062. [0.6109, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567],
  31063. [0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343],
  31064. [0.6275, 0.4081, 0.8063, 0.2017, 0.4825, 0.1583, 0.6156, 0.4869],
  31065. [0.6104, 0.4029, 0.8738, 0.4900, 0.4088, 0.4533, 0.5070, 0.5510],
  31066. [0.6059, 0.4002, 0.7562, 0.2767, 0.3538, 0.3033, 0.5529, 0.5455]],
  31067. device='cuda:0', dtype=torch.float64)
  31068. predictions are: tensor([[-2.1810, -2.2031, 0.8995, -1.0844, -0.4663, -1.2910, 0.0934, 0.3814],
  31069. [ 0.4687, -0.4924, 1.8094, -0.1270, -0.4819, 0.2563, 0.5449, 0.1873],
  31070. [ 0.4851, -0.4925, 1.7876, -0.3265, -0.6856, -0.1683, 0.4772, 0.1175],
  31071. [ 0.7273, -0.2978, 0.9829, -1.2828, -0.4788, -1.0173, 0.3975, 0.2628],
  31072. [ 0.4733, -0.5219, 1.8115, -0.5131, -0.3404, -0.8316, 0.9798, 0.1113],
  31073. [ 0.9110, -0.2001, 1.4314, -1.2082, -0.0286, -1.5178, 0.5554, 0.0304],
  31074. [ 0.4506, -0.4711, 1.7314, 0.1099, -0.3668, -0.0410, 0.1666, 0.2598],
  31075. [ 0.4830, -0.4528, 1.1680, -0.8904, -0.5668, -0.7985, 0.3129, 0.1844]],
  31076. device='cuda:0', grad_fn=<AddmmBackward>)
  31077. landmarks are: tensor([[[-2.2859, -2.2859, 0.8033, -1.1250, -0.3864, -1.3082, 0.1126,
  31078. 0.4543],
  31079. [ 0.5746, -0.4706, 1.8476, -0.2365, -0.5068, 0.2245, 0.6069,
  31080. 0.1449],
  31081. [ 0.6010, -0.4345, 1.8480, -0.3537, -0.6462, -0.2613, 0.6524,
  31082. -0.0099],
  31083. [ 0.5356, -0.4217, 1.0339, -1.4776, -0.5076, -1.1081, 0.4219,
  31084. 0.2853],
  31085. [ 0.6125, -0.4369, 1.9173, -0.5384, -0.2594, -0.8386, 0.9741,
  31086. 0.1821],
  31087. [ 0.6124, -0.4010, 1.4381, -1.3544, -0.0573, -1.5546, 0.5573,
  31088. -0.0369],
  31089. [ 0.5334, -0.4252, 1.7499, -0.0226, -0.3979, -0.1920, 0.0558,
  31090. 0.2589],
  31091. [ 0.5129, -0.4374, 1.2072, -1.0080, -0.6520, -0.8848, 0.2679,
  31092. 0.2335]]], device='cuda:0')
  31093. loss_train_step before backward: tensor(0.0088, device='cuda:0', grad_fn=<MseLossBackward>)
  31094. loss_train_step after backward: tensor(0.0088, device='cuda:0', grad_fn=<MseLossBackward>)
  31095. loss_train: 0.18051145039498806
  31096. step: 12
  31097. running loss: 0.015042620866249004
  31098. Train Steps: 12/90 Loss: 0.0150 torch.Size([8, 600, 800])
  31099. torch.Size([8, 8])
  31100. tensor([[0.6193, 0.4034, 0.7757, 0.2347, 0.3733, 0.2919, 0.5930, 0.4926],
  31101. [0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100],
  31102. [0.6296, 0.4045, 0.9138, 0.4100, 0.4232, 0.4242, 0.7422, 0.5297],
  31103. [0.6261, 0.4045, 0.8865, 0.5369, 0.3895, 0.4859, 0.6683, 0.5249],
  31104. [0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019],
  31105. [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398],
  31106. [0.6286, 0.4060, 0.9188, 0.4333, 0.3675, 0.4167, 0.7034, 0.5528],
  31107. [0.6166, 0.4008, 0.8563, 0.5667, 0.4388, 0.4933, 0.5575, 0.5567]],
  31108. device='cuda:0', dtype=torch.float64)
  31109. predictions are: tensor([[ 0.5929, -0.3831, 1.1603, -1.3763, -0.5561, -0.9755, 0.1903, 0.0160],
  31110. [ 0.5210, -0.4703, 1.7357, -0.5514, -0.6131, -0.5008, 0.5070, 0.1872],
  31111. [ 0.7239, -0.3509, 1.8028, -0.5375, -0.4470, -0.3079, 0.8961, 0.2300],
  31112. [ 0.6579, -0.3626, 1.6936, 0.0632, -0.4777, 0.0678, 0.5938, 0.2176],
  31113. [ 0.5153, -0.4594, 1.7896, -0.2423, -0.6064, -0.4672, 0.4889, 0.1511],
  31114. [ 0.6139, -0.3908, 1.5405, 0.1031, -0.4729, -0.2208, 0.2265, 0.2722],
  31115. [ 0.6588, -0.3622, 1.8513, -0.3756, -0.4461, -0.3094, 0.7993, 0.3618],
  31116. [ 0.6246, -0.3740, 1.5466, 0.2366, -0.1276, 0.0130, 0.1270, 0.3720]],
  31117. device='cuda:0', grad_fn=<AddmmBackward>)
  31118. landmarks are: tensor([[[ 0.5746, -0.4225, 1.2968, -1.2019, -0.5615, -0.9374, 0.4533,
  31119. -0.0108],
  31120. [ 0.6220, -0.4436, 1.8711, -0.3691, -0.6173, -0.4999, 0.6702,
  31121. 0.0697],
  31122. [ 0.6224, -0.4177, 1.9346, -0.3921, -0.3314, -0.3264, 1.1422,
  31123. 0.1608],
  31124. [ 0.6059, -0.4177, 1.8087, 0.1941, -0.4868, -0.0414, 0.8010,
  31125. 0.1385],
  31126. [ 0.6372, -0.4129, 1.8942, -0.0765, -0.6173, -0.4768, 0.6999,
  31127. 0.0325],
  31128. [ 0.5786, -0.4463, 1.6655, 0.2216, -0.5115, -0.2675, 0.4236,
  31129. 0.2075],
  31130. [ 0.6174, -0.4105, 1.9577, -0.2844, -0.5885, -0.3614, 0.9631,
  31131. 0.2676],
  31132. [ 0.5620, -0.4346, 1.6691, 0.3315, -0.2594, -0.0072, 0.2891,
  31133. 0.2853]]], device='cuda:0')
  31134. loss_train_step before backward: tensor(0.0127, device='cuda:0', grad_fn=<MseLossBackward>)
  31135. loss_train_step after backward: tensor(0.0127, device='cuda:0', grad_fn=<MseLossBackward>)
  31136. loss_train: 0.19325176812708378
  31137. step: 13
  31138. running loss: 0.01486552062516029
  31139. Train Steps: 13/90 Loss: 0.0149 torch.Size([8, 600, 800])
  31140. torch.Size([8, 8])
  31141. tensor([[0.6249, 0.4138, 0.9038, 0.4517, 0.3862, 0.4917, 0.6138, 0.5333],
  31142. [0.6042, 0.3990, 0.6831, 0.2875, 0.3500, 0.3133, 0.5143, 0.5510],
  31143. [0.6307, 0.4045, 0.8025, 0.5833, 0.3775, 0.4867, 0.6892, 0.5459],
  31144. [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150],
  31145. [0.6350, 0.4144, 0.9250, 0.4100, 0.4125, 0.3750, 0.7129, 0.5945],
  31146. [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973],
  31147. [0.6371, 0.4092, 0.8337, 0.5850, 0.3950, 0.5117, 0.6559, 0.5262],
  31148. [ nan, nan, 0.7240, 0.2722, 0.3900, 0.2567, 0.5168, 0.5933]],
  31149. device='cuda:0', dtype=torch.float64)
  31150. predictions are: tensor([[ 0.6185, -0.3156, 1.8265, -0.3126, -0.4726, 0.0292, 0.3994, 0.2266],
  31151. [ 0.5336, -0.3887, 0.9285, -1.1986, -0.5366, -0.9300, 0.0476, 0.2809],
  31152. [ 0.6409, -0.3765, 1.4790, 0.1583, -0.4156, -0.0975, 0.8417, 0.2170],
  31153. [ 0.4891, -0.4365, 1.8433, -0.1204, -0.5106, -0.2830, 0.4545, 0.1341],
  31154. [ 0.7050, -0.3099, 1.9224, -0.4509, -0.2300, -0.5580, 0.9066, 0.3912],
  31155. [ 0.6713, -0.3561, 1.7889, 0.0174, -0.5189, -0.1541, 0.5659, 0.0518],
  31156. [ 0.6875, -0.3242, 1.5716, 0.2427, -0.3745, 0.0493, 0.6705, 0.1036],
  31157. [-1.9622, -2.0539, 1.0842, -1.0704, -0.5110, -1.1750, -0.0437, 0.3613]],
  31158. device='cuda:0', grad_fn=<AddmmBackward>)
  31159. landmarks are: tensor([[[ 0.6003, -0.3747, 1.8885, -0.1997, -0.5018, -0.0149, 0.5490,
  31160. 0.1775],
  31161. [ 0.5049, -0.4428, 0.8692, -0.9581, -0.6693, -0.8386, 0.0897,
  31162. 0.2589],
  31163. [ 0.6273, -0.4177, 1.4208, 0.4085, -0.5423, -0.0380, 0.8973,
  31164. 0.2356],
  31165. [ 0.6096, -0.4046, 1.8249, -0.0072, -0.6115, -0.3537, 0.6182,
  31166. 0.0928],
  31167. [ 0.6471, -0.3719, 1.9866, -0.3921, -0.3806, -0.5538, 1.0070,
  31168. 0.4600],
  31169. [ 0.6240, -0.4321, 1.8423, 0.1852, -0.5885, -0.1612, 0.6962,
  31170. 0.0111],
  31171. [ 0.6569, -0.3960, 1.5651, 0.4162, -0.4614, 0.0774, 0.7438,
  31172. 0.1447],
  31173. [-2.2859, -2.2859, 1.0580, -1.0288, -0.4845, -1.1004, 0.1011,
  31174. 0.4543]]], device='cuda:0')
  31175. loss_train_step before backward: tensor(0.0104, device='cuda:0', grad_fn=<MseLossBackward>)
  31176. loss_train_step after backward: tensor(0.0104, device='cuda:0', grad_fn=<MseLossBackward>)
  31177. loss_train: 0.20368763711303473
  31178. step: 14
  31179. running loss: 0.014549116936645337
  31180.  
  31181. Train Steps: 14/90 Loss: 0.0145 torch.Size([8, 600, 800])
  31182. torch.Size([8, 8])
  31183. tensor([[0.6109, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117],
  31184. [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364],
  31185. [0.6250, 0.3993, 0.9138, 0.4333, 0.3763, 0.5217, 0.6995, 0.5320],
  31186. [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483],
  31187. [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150],
  31188. [0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350],
  31189. [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367],
  31190. [0.6102, 0.4005, 0.8688, 0.5100, 0.4813, 0.5400, 0.5404, 0.5064]],
  31191. device='cuda:0', dtype=torch.float64)
  31192. predictions are: tensor([[ 0.5511, -0.4068, 1.7669, -0.1467, -0.6737, -0.4046, 0.2705, 0.1262],
  31193. [ 0.6452, -0.3602, 1.3157, -1.0661, -0.3160, -1.2657, 0.7119, 0.1955],
  31194. [ 0.4112, -0.5618, 1.8742, -0.2422, -0.5652, 0.0540, 1.0783, 0.2086],
  31195. [ 0.6926, -0.3282, 1.0094, -1.2379, -0.4402, -1.1454, 0.4859, 0.2555],
  31196. [ 0.5699, -0.3966, 1.7770, -0.0380, -0.3195, 0.3987, 0.6858, 0.2068],
  31197. [ 0.6017, -0.3589, 1.3299, -1.0885, -0.2723, -1.3584, 0.3951, 0.2467],
  31198. [ 0.4981, -0.4629, 1.3449, -0.8999, -0.6521, -0.7340, 0.4934, 0.2876],
  31199. [ 0.5636, -0.4064, 1.6784, 0.0822, -0.1191, 0.0233, 0.1590, 0.1523]],
  31200. device='cuda:0', grad_fn=<AddmmBackward>)
  31201. landmarks are: tensor([[[ 0.5356, -0.4343, 1.7441, -0.1150, -0.6462, -0.3614, 0.3238,
  31202. 0.0774],
  31203. [ 0.6158, -0.3960, 1.4092, -1.2774, -0.2074, -1.1851, 0.8491,
  31204. 0.1917],
  31205. [ 0.6010, -0.4417, 1.9346, -0.2844, -0.5480, 0.1236, 0.9448,
  31206. 0.1715],
  31207. [ 0.5784, -0.4085, 1.0859, -1.3929, -0.4037, -1.1158, 0.5605,
  31208. 0.2468],
  31209. [ 0.6014, -0.3883, 1.8192, -0.1150, -0.2882, 0.4008, 0.5201,
  31210. 0.0928],
  31211. [ 0.5768, -0.4031, 1.3838, -1.1527, -0.2188, -1.4216, 0.4379,
  31212. 0.1850],
  31213. [ 0.5697, -0.4442, 1.3529, -0.9515, -0.5774, -0.7801, 0.5253,
  31214. 0.1931],
  31215. [ 0.5328, -0.4361, 1.7268, 0.0697, -0.0630, 0.2083, 0.2103,
  31216. 0.0532]]], device='cuda:0')
  31217. loss_train_step before backward: tensor(0.0061, device='cuda:0', grad_fn=<MseLossBackward>)
  31218. loss_train_step after backward: tensor(0.0061, device='cuda:0', grad_fn=<MseLossBackward>)
  31219. loss_train: 0.20982471108436584
  31220. step: 15
  31221. running loss: 0.013988314072291057
  31222. Train Steps: 15/90 Loss: 0.0140 torch.Size([8, 600, 800])
  31223. torch.Size([8, 8])
  31224. tensor([[0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517],
  31225. [0.6207, 0.4110, 0.8738, 0.5000, 0.4800, 0.5633, 0.6300, 0.5433],
  31226. [0.6261, 0.4131, 0.8938, 0.3550, 0.4000, 0.2683, 0.6188, 0.5283],
  31227. [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100],
  31228. [0.6314, 0.4050, 0.8227, 0.5431, 0.4150, 0.5517, 0.7121, 0.5690],
  31229. [0.6092, 0.4001, 0.8638, 0.4867, 0.4288, 0.5367, 0.5484, 0.5064],
  31230. [0.6151, 0.4085, 0.8750, 0.4367, 0.3887, 0.4367, 0.5066, 0.5846],
  31231. [0.6161, 0.4055, 0.8675, 0.3867, 0.3713, 0.4033, 0.5195, 0.5162]],
  31232. device='cuda:0', dtype=torch.float64)
  31233. predictions are: tensor([[ 0.6932, -0.3002, 0.9223, -1.2076, -0.4446, -1.2456, 0.4461, 0.3540],
  31234. [ 0.7345, -0.2983, 1.7527, -0.1153, -0.1353, 0.1659, 0.6014, 0.3645],
  31235. [ 0.7010, -0.3145, 1.8317, -0.7354, -0.5536, -1.0791, 0.7113, 0.1254],
  31236. [ 0.5710, -0.4561, 1.6580, 0.2129, -0.4870, -0.0747, 0.8136, 0.0383],
  31237. [ 0.5815, -0.4396, 1.5620, 0.1961, -0.4041, 0.2234, 1.1959, 0.3494],
  31238. [ 0.4783, -0.4605, 1.6568, -0.0696, -0.3142, 0.1620, 0.2832, 0.1092],
  31239. [ 0.4819, -0.4050, 1.6976, -0.2870, -0.5468, -0.2685, 0.1350, 0.3684],
  31240. [ 0.4149, -0.4674, 1.7142, -0.4944, -0.6656, -0.4889, 0.2146, 0.0351]],
  31241. device='cuda:0', grad_fn=<AddmmBackward>)
  31242. landmarks are: tensor([[[ 0.5713, -0.3671, 0.8665, -1.0696, -0.3691, -1.2236, 0.3527,
  31243. 0.2622],
  31244. [ 0.5811, -0.3878, 1.7499, 0.0236, -0.0688, 0.3161, 0.6240,
  31245. 0.2237],
  31246. [ 0.6062, -0.3778, 1.8423, -0.6462, -0.4383, -1.0465, 0.5721,
  31247. 0.1544],
  31248. [ 0.6125, -0.4273, 1.6402, 0.2006, -0.4788, -0.0303, 0.6182,
  31249. 0.0697],
  31250. [ 0.6306, -0.4153, 1.5141, 0.2224, -0.3691, 0.2622, 1.0033,
  31251. 0.3425],
  31252. [ 0.5278, -0.4377, 1.7037, -0.0380, -0.3055, 0.1929, 0.2473,
  31253. 0.0532],
  31254. [ 0.5553, -0.3992, 1.7557, -0.2690, -0.4903, -0.2690, 0.0542,
  31255. 0.4145],
  31256. [ 0.5598, -0.4129, 1.7210, -0.4999, -0.5711, -0.4229, 0.1136,
  31257. 0.0983]]], device='cuda:0')
  31258. loss_train_step before backward: tensor(0.0061, device='cuda:0', grad_fn=<MseLossBackward>)
  31259. loss_train_step after backward: tensor(0.0061, device='cuda:0', grad_fn=<MseLossBackward>)
  31260. loss_train: 0.21596113871783018
  31261. step: 16
  31262. running loss: 0.013497571169864386
  31263. Train Steps: 16/90 Loss: 0.0135 torch.Size([8, 600, 800])
  31264. torch.Size([8, 8])
  31265. tensor([[ nan, nan, 0.8888, 0.3100, 0.5262, 0.2817, 0.7145, 0.6003],
  31266. [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6063, 0.5617],
  31267. [0.6245, 0.4115, 0.8700, 0.4883, 0.4625, 0.5517, 0.6100, 0.5217],
  31268. [0.6115, 0.4005, 0.8838, 0.3867, 0.3763, 0.4700, 0.5800, 0.5550],
  31269. [0.6336, 0.4191, 0.8938, 0.5167, 0.3937, 0.3517, 0.7343, 0.5748],
  31270. [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301],
  31271. [0.6274, 0.4003, 0.8638, 0.5967, 0.3688, 0.4900, 0.6108, 0.4661],
  31272. [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114]],
  31273. device='cuda:0', dtype=torch.float64)
  31274. predictions are: tensor([[-1.4391e+00, -1.7195e+00, 1.6965e+00, -8.3241e-01, -7.4165e-03,
  31275. -1.2388e+00, 9.2158e-01, 4.8889e-01],
  31276. [ 6.7004e-01, -2.8881e-01, 1.7075e+00, -1.5896e-01, -6.4548e-01,
  31277. -2.0172e-01, 4.7335e-01, 2.4208e-01],
  31278. [ 5.9918e-01, -3.3162e-01, 1.6278e+00, -2.1823e-01, -2.4393e-01,
  31279. 1.9002e-01, 4.4332e-01, 2.2739e-01],
  31280. [ 6.2093e-01, -3.1903e-01, 1.6007e+00, -6.1591e-01, -6.5189e-01,
  31281. -1.0654e-01, 3.2741e-01, 1.7108e-01],
  31282. [ 6.0972e-01, -3.6531e-01, 1.7034e+00, 2.6776e-03, -4.8217e-01,
  31283. -6.0080e-01, 8.8999e-01, 2.6616e-01],
  31284. [ 5.8475e-01, -4.0440e-01, 1.6300e+00, -1.0673e-01, -3.9924e-01,
  31285. -6.8137e-02, 7.1257e-01, 1.2286e-01],
  31286. [ 7.6503e-01, -2.8282e-01, 1.5378e+00, 3.3768e-01, -5.7958e-01,
  31287. -9.7698e-02, 6.3861e-01, 9.1715e-04],
  31288. [ 5.7975e-01, -3.8051e-01, 1.5980e+00, 9.9984e-03, -2.3316e-01,
  31289. 1.7962e-01, 4.1097e-01, 9.1926e-02]], device='cuda:0',
  31290. grad_fn=<AddmmBackward>)
  31291. landmarks are: tensor([[[-2.2859, -2.2859, 1.8192, -0.8540, 0.1448, -0.9849, 1.0143,
  31292. 0.4867],
  31293. [ 0.5797, -0.3965, 1.8480, -0.0765, -0.6173, -0.1535, 0.5143,
  31294. 0.3084],
  31295. [ 0.5987, -0.3852, 1.7326, -0.0303, -0.1497, 0.2622, 0.5316,
  31296. 0.1236],
  31297. [ 0.5384, -0.4361, 1.7961, -0.4999, -0.5480, -0.1150, 0.3931,
  31298. 0.2776],
  31299. [ 0.6405, -0.3503, 1.8423, 0.1005, -0.4672, -0.6616, 1.1057,
  31300. 0.3692],
  31301. [ 0.6010, -0.4562, 1.7198, -0.0090, -0.3464, 0.0108, 0.6294,
  31302. 0.1627],
  31303. [ 0.6120, -0.4371, 1.7037, 0.4701, -0.5827, -0.0226, 0.5354,
  31304. -0.1331],
  31305. [ 0.5450, -0.4730, 1.7420, 0.1372, -0.1919, 0.2614, 0.4976,
  31306. 0.0764]]], device='cuda:0')
  31307. loss_train_step before backward: tensor(0.0266, device='cuda:0', grad_fn=<MseLossBackward>)
  31308. loss_train_step after backward: tensor(0.0266, device='cuda:0', grad_fn=<MseLossBackward>)
  31309. loss_train: 0.24258825462311506
  31310. step: 17
  31311. running loss: 0.014269897330771475
  31312. Train Steps: 17/90 Loss: 0.0143 torch.Size([8, 600, 800])
  31313. torch.Size([8, 8])
  31314. tensor([[0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6063, 0.5617],
  31315. [ nan, nan, 0.9088, 0.3783, 0.4562, 0.2617, 0.6741, 0.5575],
  31316. [0.6329, 0.4196, 0.9238, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748],
  31317. [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466],
  31318. [0.6246, 0.4008, 0.8757, 0.5088, 0.4101, 0.5392, 0.6644, 0.5133],
  31319. [0.6198, 0.3997, 0.8582, 0.5361, 0.4117, 0.5016, 0.5942, 0.5134],
  31320. [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663],
  31321. [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633]],
  31322. device='cuda:0', dtype=torch.float64)
  31323. predictions are: tensor([[ 0.7508, -0.2604, 1.7657, -0.0488, -0.5387, -0.0755, 0.5828, 0.2377],
  31324. [-1.1455, -1.5244, 1.9091, -0.5219, -0.2446, -1.0815, 0.7982, 0.3013],
  31325. [ 0.7796, -0.2696, 1.9089, -0.1851, -0.3437, -0.8657, 1.0811, 0.2547],
  31326. [ 0.5238, -0.3978, 1.2088, -0.5805, -0.6452, -0.2787, 0.2464, 0.2178],
  31327. [ 0.7354, -0.3332, 1.7281, 0.0586, -0.3375, 0.3449, 0.9514, 0.0944],
  31328. [ 0.6988, -0.3625, 1.6217, 0.1694, -0.3268, 0.1545, 0.4640, 0.0219],
  31329. [ 0.4548, -0.4465, 0.9776, -1.3111, -0.4202, -1.1218, 0.3172, 0.2431],
  31330. [ 0.5979, -0.3529, 1.5608, -0.8263, -0.5420, -0.7747, 0.3924, 0.1048]],
  31331. device='cuda:0', grad_fn=<AddmmBackward>)
  31332. landmarks are: tensor([[[ 0.5797, -0.3965, 1.8480, -0.0765, -0.6173, -0.1535, 0.5143,
  31333. 0.3084],
  31334. [-2.2859, -2.2859, 1.9115, -0.5384, -0.1785, -1.0773, 0.8278,
  31335. 0.2890],
  31336. [ 0.6372, -0.3479, 1.9808, -0.1689, -0.3171, -0.9310, 1.1057,
  31337. 0.3692],
  31338. [ 0.5076, -0.4443, 1.2337, -0.5023, -0.6808, -0.3614, 0.0866,
  31339. 0.2386],
  31340. [ 0.5993, -0.4345, 1.7587, 0.0641, -0.3917, 0.2048, 0.7827,
  31341. 0.0852],
  31342. [ 0.5767, -0.4396, 1.6782, 0.1905, -0.3844, 0.0308, 0.4588,
  31343. 0.0855],
  31344. [ 0.5269, -0.4176, 1.0628, -1.3159, -0.4037, -1.2236, 0.2432,
  31345. 0.3297],
  31346. [ 0.5944, -0.4008, 1.6748, -0.9002, -0.5711, -0.8848, 0.2776,
  31347. 0.3161]]], device='cuda:0')
  31348. loss_train_step before backward: tensor(0.0362, device='cuda:0', grad_fn=<MseLossBackward>)
  31349. loss_train_step after backward: tensor(0.0362, device='cuda:0', grad_fn=<MseLossBackward>)
  31350. loss_train: 0.2787413029000163
  31351. step: 18
  31352. running loss: 0.015485627938889794
  31353.  
  31354. Train Steps: 18/90 Loss: 0.0155 torch.Size([8, 600, 800])
  31355. torch.Size([8, 8])
  31356. tensor([[0.6087, 0.3976, 0.8337, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300],
  31357. [0.6266, 0.4067, 0.8588, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267],
  31358. [0.6308, 0.3990, 0.8688, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133],
  31359. [0.6125, 0.4076, 0.8488, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505],
  31360. [0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528],
  31361. [0.6250, 0.4110, 0.7238, 0.2067, 0.4263, 0.1883, 0.5625, 0.5633],
  31362. [0.6126, 0.3954, 0.8538, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350],
  31363. [0.6189, 0.4033, 0.8650, 0.5267, 0.4487, 0.5150, 0.5925, 0.5050]],
  31364. device='cuda:0', dtype=torch.float64)
  31365. predictions are: tensor([[ 0.5391, -0.4250, 1.5793, -0.5507, -0.5772, -0.7345, 0.5866, 0.2346],
  31366. [ 0.6453, -0.3269, 1.6569, -0.9600, -0.3731, -0.8979, 0.7733, 0.1688],
  31367. [ 0.6124, -0.4489, 1.7579, 0.1917, -0.4467, 0.1514, 0.9555, 0.0758],
  31368. [ 0.4832, -0.4268, 1.5981, -0.4180, -0.6525, -0.4576, 0.0383, 0.2399],
  31369. [ 0.6428, -0.4099, 1.9707, -0.5950, -0.3578, -0.4111, 1.2218, 0.2484],
  31370. [ 0.4976, -0.4160, 0.9908, -1.2584, -0.3256, -1.3835, 0.4925, 0.2604],
  31371. [ 0.6152, -0.4093, 1.6455, 0.1106, -0.3446, 0.0551, 0.3082, 0.1433],
  31372. [ 0.5605, -0.4519, 1.6999, 0.1790, -0.2418, 0.2339, 0.5199, 0.1525]],
  31373. device='cuda:0', grad_fn=<AddmmBackward>)
  31374. landmarks are: tensor([[[ 0.5255, -0.4495, 1.5651, -0.4999, -0.5711, -0.8463, 0.4566,
  31375. 0.1621],
  31376. [ 0.6084, -0.4076, 1.6806, -0.9618, -0.2998, -0.9695, 0.6356,
  31377. 0.1467],
  31378. [ 0.6275, -0.4430, 1.7268, 0.1082, -0.4614, 0.0159, 0.6644,
  31379. 0.0851],
  31380. [ 0.5433, -0.4032, 1.6344, -0.4922, -0.5769, -0.5846, 0.0357,
  31381. 0.2567],
  31382. [ 0.6388, -0.3623, 1.9173, -0.7386, -0.3517, -0.5846, 1.1495,
  31383. 0.2676],
  31384. [ 0.6010, -0.3875, 1.0570, -1.3313, -0.3171, -1.4160, 0.3122,
  31385. 0.3161],
  31386. [ 0.5435, -0.4597, 1.6575, 0.0159, -0.3229, -0.1150, 0.1877,
  31387. 0.1854],
  31388. [ 0.5726, -0.4230, 1.7095, 0.1467, -0.2132, 0.0928, 0.4508,
  31389. 0.0467]]], device='cuda:0')
  31390. loss_train_step before backward: tensor(0.0069, device='cuda:0', grad_fn=<MseLossBackward>)
  31391. loss_train_step after backward: tensor(0.0069, device='cuda:0', grad_fn=<MseLossBackward>)
  31392. loss_train: 0.2856500316411257
  31393. step: 19
  31394. running loss: 0.015034212191638193
  31395. Train Steps: 19/90 Loss: 0.0150 torch.Size([8, 600, 800])
  31396. torch.Size([8, 8])
  31397. tensor([[0.6200, 0.4118, 0.8287, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799],
  31398. [0.6278, 0.4253, 0.8875, 0.5017, 0.4113, 0.2750, 0.5413, 0.6196],
  31399. [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351],
  31400. [0.6100, 0.4016, 0.8600, 0.5067, 0.4612, 0.5233, 0.5086, 0.5519],
  31401. [0.6040, 0.4002, 0.7338, 0.2267, 0.3975, 0.2100, 0.5231, 0.4778],
  31402. [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611],
  31403. [0.6118, 0.4052, 0.8463, 0.3917, 0.3538, 0.3450, 0.5053, 0.5593],
  31404. [0.6215, 0.4119, 0.7688, 0.2300, 0.4200, 0.2283, 0.5925, 0.5317]],
  31405. device='cuda:0', dtype=torch.float64)
  31406. predictions are: tensor([[ 0.6642, -0.3373, 1.6242, -0.4397, -0.5543, -0.7999, 0.5567, 0.2802],
  31407. [ 0.3577, -0.5237, 1.7865, 0.0304, -0.3552, -0.8476, 0.5982, 0.5128],
  31408. [ 0.7060, -0.3369, 1.3578, -1.1410, -0.2806, -1.2913, 0.6984, 0.0911],
  31409. [ 0.4874, -0.5204, 1.7267, 0.1068, -0.1824, 0.2917, 0.4077, 0.1674],
  31410. [ 0.6098, -0.4015, 1.1021, -1.3660, -0.4870, -1.1326, 0.5583, -0.0315],
  31411. [ 0.5276, -0.4831, 1.8539, -0.0930, -0.3060, 0.1485, 0.5581, 0.2284],
  31412. [ 0.2133, -0.6280, 1.6494, -0.4383, -0.6503, -0.4903, 0.2719, 0.2101],
  31413. [ 0.6117, -0.4047, 1.2955, -1.1935, -0.3856, -0.9915, 0.8060, 0.1934]],
  31414. device='cuda:0', grad_fn=<AddmmBackward>)
  31415. landmarks are: tensor([[[ 0.5779, -0.3840, 1.5420, -0.4306, -0.5423, -0.9772, 0.2041,
  31416. 0.3928],
  31417. [ 0.6139, -0.3216, 1.8134, 0.0313, -0.3864, -1.0157, 0.2144,
  31418. 0.5762],
  31419. [ 0.5664, -0.4321, 1.2862, -1.0003, -0.2189, -1.4608, 0.3883,
  31420. 0.1855],
  31421. [ 0.5318, -0.4310, 1.6864, 0.0543, -0.1554, 0.1313, 0.0635,
  31422. 0.2634],
  31423. [ 0.5037, -0.4375, 1.1032, -1.2390, -0.4499, -1.3159, 0.1301,
  31424. -0.0791],
  31425. [ 0.5491, -0.4290, 1.7788, -0.1073, -0.2651, -0.1073, 0.2555,
  31426. 0.3057],
  31427. [ 0.5399, -0.4142, 1.6229, -0.4768, -0.6520, -0.6924, 0.0481,
  31428. 0.2972],
  31429. [ 0.5849, -0.3836, 1.2649, -1.2236, -0.3460, -1.2313, 0.4508,
  31430. 0.1698]]], device='cuda:0')
  31431. loss_train_step before backward: tensor(0.0265, device='cuda:0', grad_fn=<MseLossBackward>)
  31432. loss_train_step after backward: tensor(0.0265, device='cuda:0', grad_fn=<MseLossBackward>)
  31433. loss_train: 0.31217335909605026
  31434. step: 20
  31435. running loss: 0.015608667954802513
  31436. Train Steps: 20/90 Loss: 0.0156 torch.Size([8, 600, 800])
  31437. torch.Size([8, 8])
  31438. tensor([[0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938],
  31439. [0.6151, 0.4085, 0.8750, 0.4367, 0.3887, 0.4367, 0.5066, 0.5846],
  31440. [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329],
  31441. [0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058],
  31442. [0.6227, 0.4083, 0.8938, 0.4800, 0.3800, 0.2950, 0.5737, 0.5350],
  31443. [ nan, nan, 0.7240, 0.2722, 0.3900, 0.2567, 0.5168, 0.5933],
  31444. [ nan, nan, 0.7525, 0.2291, 0.3838, 0.3017, 0.6050, 0.5667],
  31445. [0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933]],
  31446. device='cuda:0', dtype=torch.float64)
  31447. predictions are: tensor([[ 0.8436, -0.2661, 1.7293, 0.3506, -0.4195, -0.0204, 0.8849, -0.0659],
  31448. [ 0.7133, -0.2538, 1.7154, -0.3090, -0.4786, -0.2038, 0.2791, 0.3428],
  31449. [ 0.8705, -0.2223, 1.6207, -0.7515, -0.6418, -0.4919, 0.7847, 0.1622],
  31450. [ 0.7293, -0.2986, 1.6384, -0.3508, -0.5085, -0.0364, 0.4600, 0.0077],
  31451. [ 0.8734, -0.1393, 1.8251, -0.0868, -0.4825, -0.8209, 0.6283, 0.1497],
  31452. [-2.1821, -2.1859, 1.0531, -1.0705, -0.4182, -1.0759, 0.1864, 0.3801],
  31453. [-1.7686, -1.9233, 1.2301, -1.1572, -0.3819, -0.9255, 0.5486, 0.3583],
  31454. [ 0.7109, -0.2645, 1.6211, 0.3448, -0.0315, -0.1749, 0.4552, 0.3760]],
  31455. device='cuda:0', grad_fn=<AddmmBackward>)
  31456. landmarks are: tensor([[[ 0.6395, -0.4213, 1.7037, 0.3623, -0.4326, -0.1073, 0.6560,
  31457. -0.0049],
  31458. [ 0.5553, -0.3992, 1.7557, -0.2690, -0.4903, -0.2690, 0.0542,
  31459. 0.4145],
  31460. [ 0.5680, -0.4393, 1.5920, -0.6672, -0.6453, -0.5457, 0.5149,
  31461. 0.1753],
  31462. [ 0.5245, -0.4361, 1.5940, -0.2921, -0.5480, -0.0919, 0.2432,
  31463. 0.0502],
  31464. [ 0.5901, -0.4000, 1.8423, -0.0688, -0.5307, -0.9233, 0.3642,
  31465. 0.1852],
  31466. [-2.2859, -2.2859, 1.0580, -1.0288, -0.4845, -1.1004, 0.1011,
  31467. 0.4543],
  31468. [-2.2859, -2.2859, 1.1898, -1.2278, -0.5134, -0.8925, 0.5085,
  31469. 0.3315],
  31470. [ 0.5959, -0.3579, 1.6055, 0.3623, -0.0573, -0.2074, 0.3122,
  31471. 0.4547]]], device='cuda:0')
  31472. loss_train_step before backward: tensor(0.0208, device='cuda:0', grad_fn=<MseLossBackward>)
  31473. loss_train_step after backward: tensor(0.0208, device='cuda:0', grad_fn=<MseLossBackward>)
  31474. loss_train: 0.33298695273697376
  31475. step: 21
  31476. running loss: 0.015856521558903512
  31477. Train Steps: 21/90 Loss: 0.0159 torch.Size([8, 600, 800])
  31478. torch.Size([8, 8])
  31479. tensor([[0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118],
  31480. [0.6200, 0.4098, 0.8237, 0.2917, 0.4012, 0.2967, 0.6000, 0.5683],
  31481. [0.6131, 0.4064, 0.8638, 0.5200, 0.4788, 0.4783, 0.5258, 0.5867],
  31482. [0.6058, 0.3986, 0.8324, 0.4626, 0.3838, 0.4983, 0.5147, 0.5466],
  31483. [0.6165, 0.4106, 0.7575, 0.1733, 0.3838, 0.2650, 0.5680, 0.5116],
  31484. [ nan, nan, 0.6688, 0.2513, 0.4113, 0.2117, 0.5193, 0.5933],
  31485. [0.6275, 0.4024, 0.8600, 0.2283, 0.5350, 0.1800, 0.7074, 0.5413],
  31486. [0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6188, 0.5433]],
  31487. device='cuda:0', dtype=torch.float64)
  31488. predictions are: tensor([[ 0.5701, -0.3610, 1.1925, -0.5998, -0.3476, -1.1460, 0.3494, 0.4502],
  31489. [ 0.5325, -0.4449, 1.5287, -0.8693, -0.5007, -0.7125, 0.5934, 0.2793],
  31490. [ 0.4868, -0.4511, 1.6993, 0.1740, -0.1953, 0.0165, 0.2025, 0.3025],
  31491. [ 0.3532, -0.5528, 1.6190, -0.0188, -0.5111, 0.0868, 0.3957, 0.1781],
  31492. [ 0.5768, -0.4132, 1.2803, -1.2153, -0.5702, -1.0083, 0.5767, -0.0181],
  31493. [-2.8392, -2.6520, 0.9217, -1.1096, -0.4011, -1.2152, 0.2435, 0.3523],
  31494. [ 0.8260, -0.2496, 1.7667, -1.0595, 0.0952, -1.3712, 1.0684, 0.1354],
  31495. [ 0.4822, -0.4867, 1.8582, 0.0209, -0.5483, 0.1714, 0.6321, 0.0669]],
  31496. device='cuda:0', grad_fn=<AddmmBackward>)
  31497. landmarks are: tensor([[[ 0.6077, -0.3226, 0.9993, -0.6462, -0.2651, -1.3082, 0.2946,
  31498. 0.5401],
  31499. [ 0.5778, -0.3930, 1.5189, -0.9387, -0.4326, -0.9156, 0.4855,
  31500. 0.3392],
  31501. [ 0.5462, -0.4090, 1.7037, 0.1159, -0.0746, -0.0765, 0.1427,
  31502. 0.4239],
  31503. [ 0.5124, -0.4446, 1.5587, -0.1493, -0.5134, 0.0159, 0.0912,
  31504. 0.2386],
  31505. [ 0.5618, -0.3895, 1.2129, -1.4853, -0.5134, -1.0619, 0.3378,
  31506. 0.0772],
  31507. [-2.2859, -2.2859, 0.8033, -1.1250, -0.3864, -1.3082, 0.1126,
  31508. 0.4543],
  31509. [ 0.6125, -0.4273, 1.6864, -1.2313, 0.1852, -1.4545, 0.9814,
  31510. 0.2142],
  31511. [ 0.5773, -0.4105, 1.8192, -0.1304, -0.5307, 0.0467, 0.5721,
  31512. 0.2237]]], device='cuda:0')
  31513. loss_train_step before backward: tensor(0.0193, device='cuda:0', grad_fn=<MseLossBackward>)
  31514. loss_train_step after backward: tensor(0.0193, device='cuda:0', grad_fn=<MseLossBackward>)
  31515. loss_train: 0.3522569965571165
  31516. step: 22
  31517. running loss: 0.016011681661687115
  31518.  
  31519. Train Steps: 22/90 Loss: 0.0160 torch.Size([8, 600, 800])
  31520. torch.Size([8, 8])
  31521. tensor([[0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895],
  31522. [ nan, nan, 0.7625, 0.2433, 0.3713, 0.2867, 0.5235, 0.5220],
  31523. [0.6212, 0.4159, 0.8675, 0.5783, 0.4088, 0.4317, 0.5613, 0.5917],
  31524. [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517],
  31525. [0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400],
  31526. [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5837, 0.5500],
  31527. [0.6197, 0.4118, 0.8688, 0.5517, 0.4037, 0.5233, 0.5875, 0.5600],
  31528. [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207]],
  31529. device='cuda:0', dtype=torch.float64)
  31530. predictions are: tensor([[ 0.4190, -0.5163, 1.2306, -1.3445, -0.2406, -1.4644, 0.4430, 0.0780],
  31531. [-2.6366, -2.5332, 1.1538, -1.0783, -0.4702, -1.0118, 0.2247, 0.2517],
  31532. [ 0.5058, -0.4009, 1.6188, 0.4193, -0.4515, -0.2694, 0.3354, 0.4193],
  31533. [ 0.4812, -0.4535, 1.7116, -0.0511, -0.2676, 0.2715, 0.4280, 0.3389],
  31534. [ 0.3823, -0.4933, 1.4640, -1.0505, -0.2502, -1.2271, 0.5584, 0.2454],
  31535. [ 0.5660, -0.4228, 1.7460, -0.1628, -0.4159, 0.1009, 0.4308, 0.1760],
  31536. [ 0.6304, -0.3654, 1.6921, 0.3137, -0.4992, 0.1335, 0.4860, 0.2379],
  31537. [ 0.4978, -0.4602, 1.4664, -1.0301, -0.3268, -1.2302, 0.7269, 0.1300]],
  31538. device='cuda:0', grad_fn=<AddmmBackward>)
  31539. landmarks are: tensor([[[ 0.5828, -0.4417, 1.2476, -1.3929, -0.1727, -1.5700, 0.4694,
  31540. -0.0248],
  31541. [-2.2859, -2.2859, 1.2360, -1.1620, -0.5711, -0.9618, 0.1322,
  31542. 0.1253],
  31543. [ 0.5836, -0.3649, 1.7210, 0.3854, -0.3979, -0.2921, 0.3065,
  31544. 0.4470],
  31545. [ 0.5897, -0.3804, 1.7788, -0.0226, -0.2016, 0.3007, 0.6125,
  31546. 0.2622],
  31547. [ 0.5958, -0.3818, 1.5536, -1.1081, -0.2074, -1.3390, 0.5663,
  31548. 0.2083],
  31549. [ 0.5604, -0.4620, 1.7961, -0.1997, -0.3229, 0.1082, 0.4104,
  31550. 0.2545],
  31551. [ 0.5762, -0.3840, 1.7268, 0.2622, -0.4210, 0.1313, 0.4277,
  31552. 0.3007],
  31553. [ 0.6132, -0.4248, 1.5247, -1.1620, -0.2882, -1.3159, 0.6545,
  31554. 0.1193]]], device='cuda:0')
  31555. loss_train_step before backward: tensor(0.0088, device='cuda:0', grad_fn=<MseLossBackward>)
  31556. loss_train_step after backward: tensor(0.0088, device='cuda:0', grad_fn=<MseLossBackward>)
  31557. loss_train: 0.36108022555708885
  31558. step: 23
  31559. running loss: 0.01569914024161256
  31560. Train Steps: 23/90 Loss: 0.0157 torch.Size([8, 600, 800])
  31561. torch.Size([8, 8])
  31562. tensor([[0.6260, 0.4214, 0.8538, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983],
  31563. [0.6265, 0.4071, 0.8875, 0.3367, 0.3975, 0.3350, 0.6312, 0.5250],
  31564. [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717],
  31565. [0.6176, 0.4017, 0.8788, 0.5100, 0.4075, 0.5250, 0.5913, 0.5033],
  31566. [0.6260, 0.4120, 0.8013, 0.2350, 0.4888, 0.1533, 0.6281, 0.4895],
  31567. [0.6109, 0.4015, 0.7668, 0.3639, 0.3513, 0.3667, 0.5200, 0.5641],
  31568. [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478],
  31569. [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208]],
  31570. device='cuda:0', dtype=torch.float64)
  31571. predictions are: tensor([[ 5.3397e-01, -4.4429e-01, 1.6449e+00, 3.7041e-01, -4.7376e-01,
  31572. -5.6033e-01, 1.8569e-01, 5.2465e-01],
  31573. [ 4.3509e-01, -5.2979e-01, 1.8832e+00, -7.4444e-01, -4.3511e-01,
  31574. -6.0733e-01, 6.5849e-01, 1.9896e-01],
  31575. [ 6.8811e-01, -3.3548e-01, 9.0652e-01, -1.1236e+00, -4.8432e-01,
  31576. -1.0859e+00, 2.8385e-01, 3.4144e-01],
  31577. [ 2.5779e-01, -7.2311e-01, 1.8132e+00, 1.1183e-01, -3.2180e-01,
  31578. 2.0947e-01, 4.4570e-01, 1.2317e-01],
  31579. [ 1.6847e-01, -7.1877e-01, 1.4626e+00, -1.2881e+00, 1.9532e-02,
  31580. -1.5018e+00, 6.9486e-01, 1.8900e-01],
  31581. [ 5.0582e-01, -4.5344e-01, 1.3273e+00, -6.3100e-01, -6.1505e-01,
  31582. -5.4265e-01, 1.8403e-01, 2.8166e-01],
  31583. [-1.5590e-03, -8.3832e-01, 1.8570e+00, -6.9928e-01, -4.9101e-01,
  31584. -7.3952e-01, 5.2433e-01, 3.0981e-01],
  31585. [ 6.4362e-01, -3.8651e-01, 1.5309e+00, -1.0483e+00, -3.1909e-01,
  31586. -1.2067e+00, 4.5029e-01, 1.2443e-01]], device='cuda:0',
  31587. grad_fn=<AddmmBackward>)
  31588. landmarks are: tensor([[[ 0.6055, -0.3393, 1.6575, 0.2545, -0.5942, -0.5461, 0.2949,
  31589. 0.4778],
  31590. [ 0.6078, -0.4057, 1.8134, -0.7309, -0.4499, -0.7386, 0.6298,
  31591. 0.1390],
  31592. [ 0.5633, -0.3883, 0.8648, -1.0288, -0.5192, -1.1081, 0.2776,
  31593. 0.3546],
  31594. [ 0.5668, -0.4306, 1.7730, 0.0697, -0.4037, 0.1390, 0.4450,
  31595. 0.0390],
  31596. [ 0.6053, -0.3830, 1.4150, -1.2005, -0.0284, -1.5777, 0.6154,
  31597. -0.0250],
  31598. [ 0.5357, -0.4315, 1.2557, -0.6051, -0.6635, -0.5923, 0.1159,
  31599. 0.3198],
  31600. [ 0.5792, -0.4052, 1.8214, -0.6587, -0.5384, -0.8924, 0.4381,
  31601. 0.2442],
  31602. [ 0.5885, -0.4429, 1.4266, -0.9926, -0.4383, -1.2313, 0.4228,
  31603. 0.1195]]], device='cuda:0')
  31604. loss_train_step before backward: tensor(0.0212, device='cuda:0', grad_fn=<MseLossBackward>)
  31605. loss_train_step after backward: tensor(0.0212, device='cuda:0', grad_fn=<MseLossBackward>)
  31606. loss_train: 0.38229384645819664
  31607. step: 24
  31608. running loss: 0.015928910269091528
  31609. Train Steps: 24/90 Loss: 0.0159 torch.Size([8, 600, 800])
  31610. torch.Size([8, 8])
  31611. tensor([[0.6200, 0.4118, 0.8287, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799],
  31612. [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633],
  31613. [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483],
  31614. [0.6170, 0.4102, 0.7468, 0.3695, 0.3463, 0.3767, 0.5238, 0.5823],
  31615. [0.6175, 0.4013, 0.8900, 0.4500, 0.4375, 0.5850, 0.6175, 0.5383],
  31616. [0.6250, 0.4054, 0.8770, 0.4723, 0.4662, 0.5367, 0.6162, 0.5433],
  31617. [0.6336, 0.4086, 0.8900, 0.3950, 0.3900, 0.2950, 0.6504, 0.5066],
  31618. [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5413, 0.5717]],
  31619. device='cuda:0', dtype=torch.float64)
  31620. predictions are: tensor([[ 0.5587, -0.4226, 1.6160, -0.5043, -0.5054, -1.1615, 0.1804, 0.3907],
  31621. [ 0.5253, -0.4484, 1.8616, 0.1801, -0.5132, -0.3568, 0.1869, 0.3852],
  31622. [ 0.4857, -0.4937, 1.0900, -1.4514, -0.3567, -1.3147, 0.4973, 0.2626],
  31623. [ 0.3560, -0.5321, 1.2411, -0.6366, -0.6568, -0.6283, 0.1718, 0.4453],
  31624. [ 0.4202, -0.5697, 1.8400, -0.1959, -0.2509, 0.3437, 0.5042, 0.1765],
  31625. [ 0.6044, -0.4654, 1.8214, -0.0966, -0.1333, 0.1111, 0.4976, 0.2188],
  31626. [ 0.5153, -0.4811, 1.9231, -0.4981, -0.4574, -1.0707, 0.7030, 0.0785],
  31627. [ 0.4016, -0.6046, 1.7845, -0.0465, -0.3021, -0.1544, 0.2956, 0.2958]],
  31628. device='cuda:0', grad_fn=<AddmmBackward>)
  31629. landmarks are: tensor([[[ 0.5779, -0.3840, 1.5420, -0.4306, -0.5423, -0.9772, 0.2041,
  31630. 0.3928],
  31631. [ 0.5879, -0.3591, 1.8018, 0.1236, -0.5596, -0.1612, 0.3469,
  31632. 0.3161],
  31633. [ 0.5784, -0.4085, 1.0859, -1.3929, -0.4037, -1.1158, 0.5605,
  31634. 0.2468],
  31635. [ 0.5639, -0.3911, 1.1634, -0.5794, -0.6866, -0.5461, 0.1334,
  31636. 0.4036],
  31637. [ 0.5664, -0.4321, 1.8249, -0.2074, -0.2651, 0.4162, 0.5663,
  31638. 0.2006],
  31639. [ 0.6009, -0.4135, 1.7651, -0.1043, -0.1323, 0.1929, 0.5605,
  31640. 0.2237],
  31641. [ 0.6405, -0.3984, 1.8249, -0.4614, -0.4845, -0.9233, 0.7182,
  31642. 0.0539],
  31643. [ 0.5432, -0.4462, 1.7557, -0.0380, -0.3113, -0.0765, 0.2141,
  31644. 0.3546]]], device='cuda:0')
  31645. loss_train_step before backward: tensor(0.0074, device='cuda:0', grad_fn=<MseLossBackward>)
  31646. loss_train_step after backward: tensor(0.0074, device='cuda:0', grad_fn=<MseLossBackward>)
  31647. loss_train: 0.38968273205682635
  31648. step: 25
  31649. running loss: 0.015587309282273054
  31650. Train Steps: 25/90 Loss: 0.0156 torch.Size([8, 600, 800])
  31651. torch.Size([8, 8])
  31652. tensor([[0.6139, 0.4019, 0.7137, 0.2150, 0.4375, 0.1533, 0.5293, 0.5006],
  31653. [0.6357, 0.4118, 0.8400, 0.2500, 0.5413, 0.1633, 0.6725, 0.5586],
  31654. [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717],
  31655. [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283],
  31656. [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250],
  31657. [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967],
  31658. [0.6263, 0.4030, 0.9000, 0.4767, 0.3800, 0.5167, 0.6415, 0.4771],
  31659. [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290]],
  31660. device='cuda:0', dtype=torch.float64)
  31661. predictions are: tensor([[ 0.1355, -0.7047, 1.0245, -1.2254, -0.2885, -1.5247, 0.0463, 0.1693],
  31662. [ 0.5920, -0.4166, 1.6155, -1.0675, 0.1942, -1.4872, 0.6072, 0.3212],
  31663. [ 0.5402, -0.4191, 1.1201, -1.1080, -0.5136, -0.9845, 0.2833, 0.4162],
  31664. [ 0.3849, -0.5270, 1.8260, -0.2820, -0.6391, -0.4763, 0.2228, 0.3612],
  31665. [ 0.4151, -0.5447, 1.8440, 0.0276, -0.4044, 0.1932, 0.3751, 0.2620],
  31666. [ 0.3779, -0.5567, 1.8440, 0.0828, -0.5188, -0.7754, 0.3708, 0.2907],
  31667. [ 0.4404, -0.5536, 1.8745, -0.0533, -0.5073, 0.0109, 0.4361, 0.1189],
  31668. [ 0.6545, -0.3770, 1.2548, -1.1872, -0.2788, -1.2400, 0.5030, 0.2161]],
  31669. device='cuda:0', grad_fn=<AddmmBackward>)
  31670. landmarks are: tensor([[[ 0.5496, -0.4295, 1.0109, -1.2928, -0.2651, -1.5777, 0.1589,
  31671. 0.0261],
  31672. [ 0.6504, -0.3840, 1.5940, -1.1312, 0.2141, -1.5315, 0.8205,
  31673. 0.2944],
  31674. [ 0.5746, -0.3882, 1.1436, -1.2005, -0.4903, -1.0157, 0.4393,
  31675. 0.3546],
  31676. [ 0.5880, -0.3887, 1.8423, -0.3306, -0.6231, -0.5230, 0.4046,
  31677. 0.1544],
  31678. [ 0.5991, -0.3803, 1.8018, -0.0534, -0.3460, 0.1852, 0.5374,
  31679. 0.1390],
  31680. [ 0.6054, -0.3767, 1.7788, 0.0774, -0.5711, -0.7694, 0.5887,
  31681. 0.0081],
  31682. [ 0.6072, -0.4247, 1.8711, -0.0842, -0.5307, 0.1005, 0.6771,
  31683. -0.0821],
  31684. [ 0.6125, -0.4273, 1.2807, -1.3253, -0.2574, -1.2542, 0.6864,
  31685. 0.1575]]], device='cuda:0')
  31686. loss_train_step before backward: tensor(0.0168, device='cuda:0', grad_fn=<MseLossBackward>)
  31687. loss_train_step after backward: tensor(0.0168, device='cuda:0', grad_fn=<MseLossBackward>)
  31688. loss_train: 0.4064830797724426
  31689. step: 26
  31690. running loss: 0.015633964606632408
  31691.  
  31692. Train Steps: 26/90 Loss: 0.0156 torch.Size([8, 600, 800])
  31693. torch.Size([8, 8])
  31694. tensor([[0.6273, 0.4105, 0.8988, 0.4517, 0.3912, 0.2550, 0.5894, 0.4811],
  31695. [0.6069, 0.3975, 0.8625, 0.5083, 0.4388, 0.5483, 0.5650, 0.4967],
  31696. [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050],
  31697. [0.6265, 0.4091, 0.8950, 0.3533, 0.3600, 0.3967, 0.6295, 0.4901],
  31698. [0.6193, 0.3930, 0.8949, 0.4437, 0.3852, 0.5435, 0.6263, 0.5263],
  31699. [0.6164, 0.4119, 0.7913, 0.2650, 0.3538, 0.3500, 0.5614, 0.5038],
  31700. [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413],
  31701. [ nan, nan, 0.7512, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617]],
  31702. device='cuda:0', dtype=torch.float64)
  31703. predictions are: tensor([[ 4.7468e-01, -4.2405e-01, 1.7528e+00, -2.6740e-01, -3.7613e-01,
  31704. -1.1936e+00, 2.4123e-01, 2.4698e-01],
  31705. [ 7.2586e-01, -3.2563e-01, 1.5593e+00, 8.7796e-02, -1.8891e-01,
  31706. 8.3703e-02, 2.8491e-02, 2.2675e-01],
  31707. [ 5.2334e-01, -4.4256e-01, 1.7059e+00, -2.4735e-01, -4.9585e-01,
  31708. -5.1626e-01, 2.3367e-01, 1.9561e-01],
  31709. [ 6.6523e-01, -3.2866e-01, 1.7513e+00, -6.5725e-01, -5.1157e-01,
  31710. -6.0597e-01, 5.1856e-01, 2.0810e-01],
  31711. [ 7.9341e-01, -2.9291e-01, 1.7165e+00, -1.9395e-01, -4.3940e-01,
  31712. -2.4226e-03, 4.1576e-01, 2.1084e-01],
  31713. [ 4.6099e-01, -4.5544e-01, 1.2891e+00, -1.0755e+00, -5.3392e-01,
  31714. -7.8686e-01, 2.7304e-01, 1.8630e-01],
  31715. [ 3.8093e-01, -5.4834e-01, 1.8468e+00, -3.5344e-01, -2.1268e-01,
  31716. -1.0234e+00, 6.6850e-01, 4.1560e-01],
  31717. [-2.5304e+00, -2.3983e+00, 9.7836e-01, -1.2369e+00, -2.7478e-01,
  31718. -1.2620e+00, 2.4886e-01, 3.5089e-01]], device='cuda:0',
  31719. grad_fn=<AddmmBackward>)
  31720. landmarks are: tensor([[[ 0.6116, -0.3898, 1.8654, -0.1997, -0.4788, -1.1081, 0.4367,
  31721. -0.0637],
  31722. [ 0.5174, -0.4497, 1.6979, 0.0620, -0.2594, 0.2468, 0.3238,
  31723. 0.0082],
  31724. [ 0.5844, -0.4466, 1.8423, -0.1997, -0.5942, -0.3998, 0.4219,
  31725. 0.0467],
  31726. [ 0.6077, -0.3965, 1.8480, -0.6539, -0.6231, -0.4537, 0.6216,
  31727. -0.0220],
  31728. [ 0.5746, -0.4706, 1.8476, -0.2365, -0.5068, 0.2245, 0.6069,
  31729. 0.1449],
  31730. [ 0.5614, -0.3834, 1.3688, -1.0619, -0.6520, -0.6693, 0.3069,
  31731. 0.0412],
  31732. [ 0.6224, -0.4345, 1.9404, -0.2921, -0.3171, -0.8771, 1.0655,
  31733. 0.2142],
  31734. [-2.2859, -2.2859, 1.1841, -1.3082, -0.3055, -1.3621, 0.3007,
  31735. 0.3084]]], device='cuda:0')
  31736. loss_train_step before backward: tensor(0.0215, device='cuda:0', grad_fn=<MseLossBackward>)
  31737. loss_train_step after backward: tensor(0.0215, device='cuda:0', grad_fn=<MseLossBackward>)
  31738. loss_train: 0.427944693248719
  31739. step: 27
  31740. running loss: 0.015849803453656258
  31741. Train Steps: 27/90 Loss: 0.0158 torch.Size([8, 600, 800])
  31742. torch.Size([8, 8])
  31743. tensor([[0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083],
  31744. [0.6197, 0.4118, 0.8688, 0.5517, 0.4037, 0.5233, 0.5875, 0.5600],
  31745. [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267],
  31746. [0.6128, 0.4118, 0.8638, 0.5333, 0.4625, 0.5267, 0.5193, 0.5475],
  31747. [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217],
  31748. [0.6214, 0.4112, 0.7838, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083],
  31749. [0.6339, 0.4102, 0.9088, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390],
  31750. [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575]],
  31751. device='cuda:0', dtype=torch.float64)
  31752. predictions are: tensor([[ 0.4570, -0.5229, 1.7586, -0.0554, -0.5715, -0.5387, 0.2527, 0.4641],
  31753. [ 0.6345, -0.3882, 1.7560, 0.2261, -0.4069, -0.0963, 0.2293, 0.2826],
  31754. [ 0.6074, -0.4412, 1.6149, 0.2421, -0.3680, -0.4795, 0.1157, 0.2424],
  31755. [ 0.6530, -0.3447, 1.7055, 0.0860, -0.1344, -0.2192, -0.1011, 0.2670],
  31756. [ 0.4935, -0.5104, 1.8558, -0.0979, -0.4405, -0.0116, 0.3747, 0.1274],
  31757. [ 0.2789, -0.6232, 1.3619, -1.4057, -0.5798, -1.1723, 0.3074, 0.1082],
  31758. [ 0.5152, -0.4956, 1.8940, -0.1752, -0.4296, -0.0322, 0.8230, 0.2762],
  31759. [ 0.9435, -0.1969, 1.7871, -0.7499, -0.3294, -0.5972, 0.8296, 0.2128]],
  31760. device='cuda:0', grad_fn=<AddmmBackward>)
  31761. landmarks are: tensor([[[ 0.5713, -0.4105, 1.7557, 0.0467, -0.6520, -0.2690, 0.3988,
  31762. 0.5239],
  31763. [ 0.5762, -0.3840, 1.7268, 0.2622, -0.4210, 0.1313, 0.4277,
  31764. 0.3007],
  31765. [ 0.5891, -0.4550, 1.5132, 0.3546, -0.3691, -0.1535, 0.3815,
  31766. 0.1467],
  31767. [ 0.5447, -0.3838, 1.7037, 0.1775, -0.1497, 0.1467, 0.1128,
  31768. 0.2431],
  31769. [ 0.6047, -0.4209, 1.7557, -0.0303, -0.4845, 0.2545, 0.6587,
  31770. 0.1236],
  31771. [ 0.5845, -0.3864, 1.3342, -1.3082, -0.6000, -0.8386, 0.3353,
  31772. 0.0620],
  31773. [ 0.6421, -0.3912, 1.9115, -0.0842, -0.4730, 0.1544, 1.1824,
  31774. 0.2035],
  31775. [ 0.6158, -0.4249, 1.8654, -0.9002, -0.3229, -0.3537, 0.9667,
  31776. 0.2890]]], device='cuda:0')
  31777. loss_train_step before backward: tensor(0.0249, device='cuda:0', grad_fn=<MseLossBackward>)
  31778. loss_train_step after backward: tensor(0.0249, device='cuda:0', grad_fn=<MseLossBackward>)
  31779. loss_train: 0.4528018026612699
  31780. step: 28
  31781. running loss: 0.01617149295218821
  31782. Train Steps: 28/90 Loss: 0.0162 torch.Size([8, 600, 800])
  31783. torch.Size([8, 8])
  31784. tensor([[0.6164, 0.4076, 0.8838, 0.4117, 0.3713, 0.5550, 0.6238, 0.5350],
  31785. [ nan, nan, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552],
  31786. [0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400],
  31787. [0.6339, 0.4102, 0.9088, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390],
  31788. [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167],
  31789. [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5637, 0.5633],
  31790. [0.6100, 0.4016, 0.8600, 0.5067, 0.4612, 0.5233, 0.5086, 0.5519],
  31791. [0.6207, 0.4081, 0.7662, 0.2067, 0.3962, 0.3200, 0.6312, 0.5300]],
  31792. device='cuda:0', dtype=torch.float64)
  31793. predictions are: tensor([[ 0.7551, -0.2866, 1.7308, -0.2989, -0.4932, 0.0477, 0.4730, 0.0802],
  31794. [-2.5839, -2.4586, 1.5701, -1.1338, 0.0178, -1.2622, 0.7895, 0.2292],
  31795. [ 0.4940, -0.4422, 1.4511, -1.0838, -0.1897, -1.4387, 0.3430, 0.1239],
  31796. [ 0.6857, -0.3570, 1.8496, -0.0496, -0.4596, 0.0302, 0.8218, 0.1607],
  31797. [ 0.3040, -0.5759, 1.3551, -0.7724, -0.6065, -0.6449, 0.2698, 0.4399],
  31798. [ 0.7371, -0.2424, 1.7519, 0.0958, -0.6039, -0.7225, 0.1211, 0.2988],
  31799. [ 0.6786, -0.3373, 1.6232, 0.1325, -0.1826, -0.1288, -0.1010, 0.1699],
  31800. [ 0.4422, -0.4763, 1.2160, -1.1084, -0.4471, -1.0021, 0.4260, 0.1637]],
  31801. device='cuda:0', grad_fn=<AddmmBackward>)
  31802. landmarks are: tensor([[[ 0.5614, -0.4032, 1.7961, -0.3844, -0.5711, 0.2776, 0.5952,
  31803. 0.1852],
  31804. [-2.2859, -2.2859, 1.7557, -1.1466, 0.0871, -1.0773, 1.1239,
  31805. 0.2783],
  31806. [ 0.5958, -0.3818, 1.5536, -1.1081, -0.2074, -1.3390, 0.5663,
  31807. 0.2083],
  31808. [ 0.6421, -0.3912, 1.9115, -0.0842, -0.4730, 0.1544, 1.1824,
  31809. 0.2035],
  31810. [ 0.5768, -0.3899, 1.3861, -0.7771, -0.5885, -0.5461, 0.5028,
  31811. 0.5624],
  31812. [ 0.5902, -0.3493, 1.7961, -0.0072, -0.5942, -0.5615, 0.3180,
  31813. 0.3161],
  31814. [ 0.5318, -0.4310, 1.6864, 0.0543, -0.1554, 0.1313, 0.0635,
  31815. 0.2634],
  31816. [ 0.5809, -0.4011, 1.2533, -1.3313, -0.4557, -0.8079, 0.6298,
  31817. 0.1621]]], device='cuda:0')
  31818. loss_train_step before backward: tensor(0.0203, device='cuda:0', grad_fn=<MseLossBackward>)
  31819. loss_train_step after backward: tensor(0.0203, device='cuda:0', grad_fn=<MseLossBackward>)
  31820. loss_train: 0.47314505791291595
  31821. step: 29
  31822. running loss: 0.01631534682458331
  31823. Train Steps: 29/90 Loss: 0.0163 torch.Size([8, 600, 800])
  31824. torch.Size([8, 8])
  31825. tensor([[0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6362, 0.5200],
  31826. [0.6280, 0.4055, 0.8600, 0.5317, 0.3800, 0.4700, 0.6275, 0.5133],
  31827. [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395],
  31828. [0.6179, 0.3961, 0.8347, 0.6020, 0.3887, 0.4624, 0.5714, 0.5373],
  31829. [ nan, nan, 0.8363, 0.3317, 0.3563, 0.3367, 0.5329, 0.5142],
  31830. [0.6151, 0.4085, 0.8750, 0.4367, 0.3887, 0.4367, 0.5066, 0.5846],
  31831. [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646],
  31832. [0.6286, 0.4078, 0.8063, 0.2267, 0.4788, 0.1533, 0.5953, 0.4913]],
  31833. device='cuda:0', dtype=torch.float64)
  31834. predictions are: tensor([[ 0.6479, -0.3862, 1.7959, -0.6655, -0.4468, 0.0505, 0.8091, 0.0969],
  31835. [ 0.6897, -0.3558, 1.7876, 0.1890, -0.4869, -0.1190, 0.6193, 0.0967],
  31836. [ 0.6158, -0.3547, 1.7291, 0.0648, -0.5078, -0.6661, 0.3796, 0.2100],
  31837. [ 0.7697, -0.3150, 1.5880, 0.3262, -0.4156, -0.3101, 0.4204, 0.1966],
  31838. [-2.2058, -2.1965, 1.4134, -1.0115, -0.5480, -0.8214, 0.3142, 0.1668],
  31839. [ 0.5060, -0.3896, 1.7413, -0.3907, -0.4397, -0.3701, 0.1312, 0.3688],
  31840. [ 0.6487, -0.3205, 1.4211, -0.8362, -0.5193, -0.9339, 0.0720, 0.2005],
  31841. [ 0.5592, -0.4004, 1.3866, -1.2150, -0.0363, -1.5537, 0.4363, 0.0868]],
  31842. device='cuda:0', grad_fn=<AddmmBackward>)
  31843. landmarks are: tensor([[[ 0.5841, -0.4199, 1.7961, -0.6693, -0.6231, 0.1082, 0.6529,
  31844. 0.1159],
  31845. [ 0.6148, -0.4130, 1.6864, 0.1698, -0.5307, -0.1150, 0.6125,
  31846. 0.0851],
  31847. [ 0.5697, -0.4514, 1.7420, 0.2672, -0.6055, -0.6312, 0.3449,
  31848. 0.2058],
  31849. [ 0.5680, -0.4562, 1.5697, 0.4947, -0.4904, -0.1503, 0.3536,
  31850. 0.1956],
  31851. [-2.2859, -2.2859, 1.5767, -0.7540, -0.6404, -0.7309, 0.1753,
  31852. 0.0893],
  31853. [ 0.5553, -0.3992, 1.7557, -0.2690, -0.4903, -0.2690, 0.0542,
  31854. 0.4145],
  31855. [ 0.5483, -0.4105, 1.4208, -0.8002, -0.6000, -0.9002, 0.0511,
  31856. 0.3220],
  31857. [ 0.6177, -0.4022, 1.4381, -1.2390, -0.0746, -1.5777, 0.4639,
  31858. -0.0168]]], device='cuda:0')
  31859. loss_train_step before backward: tensor(0.0082, device='cuda:0', grad_fn=<MseLossBackward>)
  31860. loss_train_step after backward: tensor(0.0082, device='cuda:0', grad_fn=<MseLossBackward>)
  31861. loss_train: 0.48136191023513675
  31862. step: 30
  31863. running loss: 0.016045397007837893
  31864.  
  31865. Train Steps: 30/90 Loss: 0.0160 torch.Size([8, 600, 800])
  31866. torch.Size([8, 8])
  31867. tensor([[0.6214, 0.4175, 0.7300, 0.2883, 0.4338, 0.2167, 0.5698, 0.5773],
  31868. [0.6200, 0.4049, 0.8638, 0.5617, 0.4125, 0.5100, 0.6013, 0.5317],
  31869. [0.6268, 0.4029, 0.8500, 0.2683, 0.3937, 0.3500, 0.6860, 0.5297],
  31870. [0.6136, 0.4060, 0.8025, 0.3217, 0.3650, 0.3000, 0.5060, 0.5646],
  31871. [0.6127, 0.4115, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495],
  31872. [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5988, 0.5667],
  31873. [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5988, 0.5283],
  31874. [0.6307, 0.4045, 0.8025, 0.5833, 0.3775, 0.4867, 0.6892, 0.5459]],
  31875. device='cuda:0', dtype=torch.float64)
  31876. predictions are: tensor([[ 0.5437, -0.4190, 1.2397, -0.9936, -0.2858, -1.3147, 0.3011, 0.3873],
  31877. [ 0.4616, -0.5160, 1.8487, 0.2746, -0.3216, 0.0470, 0.4399, 0.0933],
  31878. [ 0.7601, -0.3446, 1.7578, -1.0177, -0.4505, -0.7159, 0.8616, 0.0667],
  31879. [ 0.6262, -0.3732, 1.5641, -0.8014, -0.5630, -0.9078, 0.0116, 0.1624],
  31880. [ 0.4152, -0.5154, 1.2256, -1.0323, -0.5391, -0.9245, 0.1490, 0.2274],
  31881. [ 0.7800, -0.3039, 1.8411, -0.6873, -0.4801, -0.6690, 0.4628, 0.2354],
  31882. [ 0.6368, -0.4049, 1.9289, 0.1807, -0.4706, -0.0622, 0.4105, 0.1088],
  31883. [ 0.6706, -0.3962, 1.6757, 0.2004, -0.4664, -0.1117, 0.8264, 0.1407]],
  31884. device='cuda:0', grad_fn=<AddmmBackward>)
  31885. landmarks are: tensor([[[ 0.5841, -0.3574, 1.0859, -0.9541, -0.2824, -1.2851, 0.3460,
  31886. 0.3808],
  31887. [ 0.5776, -0.4159, 1.7037, 0.3084, -0.3806, 0.0697, 0.4912,
  31888. 0.1698],
  31889. [ 0.6092, -0.4249, 1.6402, -1.0465, -0.4672, -0.6693, 0.8827,
  31890. 0.1608],
  31891. [ 0.5483, -0.4105, 1.4208, -0.8002, -0.6000, -0.9002, 0.0511,
  31892. 0.3220],
  31893. [ 0.5442, -0.3855, 1.0224, -0.9541, -0.6115, -0.9233, 0.1745,
  31894. 0.2522],
  31895. [ 0.5783, -0.3933, 1.6748, -0.6154, -0.5769, -0.6462, 0.4797,
  31896. 0.3315],
  31897. [ 0.5603, -0.4319, 1.7788, 0.1775, -0.5538, -0.0611, 0.4797,
  31898. 0.1544],
  31899. [ 0.6273, -0.4177, 1.4208, 0.4085, -0.5423, -0.0380, 0.8973,
  31900. 0.2356]]], device='cuda:0')
  31901. loss_train_step before backward: tensor(0.0089, device='cuda:0', grad_fn=<MseLossBackward>)
  31902. loss_train_step after backward: tensor(0.0089, device='cuda:0', grad_fn=<MseLossBackward>)
  31903. loss_train: 0.49028233671560884
  31904. step: 31
  31905. running loss: 0.015815559248890606
  31906. Train Steps: 31/90 Loss: 0.0158 torch.Size([8, 600, 800])
  31907. torch.Size([8, 8])
  31908. tensor([[0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200],
  31909. [0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689],
  31910. [0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343],
  31911. [0.6179, 0.3993, 0.8925, 0.4789, 0.3879, 0.4900, 0.6041, 0.5279],
  31912. [0.6277, 0.4103, 0.8087, 0.5717, 0.4188, 0.4750, 0.5663, 0.6083],
  31913. [0.6126, 0.4039, 0.8237, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138],
  31914. [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012],
  31915. [0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767]],
  31916. device='cuda:0', dtype=torch.float64)
  31917. predictions are: tensor([[ 0.5666, -0.4195, 1.8396, -1.0422, -0.1928, -1.2118, 0.5643, 0.1019],
  31918. [ 0.6211, -0.3755, 1.3698, -0.6980, -0.6557, -0.4697, 0.1725, 0.2209],
  31919. [ 0.5803, -0.4509, 1.9152, -0.6383, -0.3041, -0.8855, 0.8861, 0.0336],
  31920. [ 0.7501, -0.3202, 1.8613, -0.1640, -0.4751, 0.0098, 0.5408, 0.0257],
  31921. [ 0.6964, -0.3375, 1.5899, 0.3226, -0.3722, -0.0324, 0.4169, 0.4082],
  31922. [ 0.4888, -0.4642, 1.5771, -0.5611, -0.5993, -0.5054, 0.4021, 0.4605],
  31923. [ 0.7757, -0.3521, 1.8301, 0.2087, -0.5843, -0.1622, 0.5861, 0.0187],
  31924. [ 0.6856, -0.3810, 1.8420, -0.0025, -0.4428, -0.0025, 0.2644, 0.1522]],
  31925. device='cuda:0', grad_fn=<AddmmBackward>)
  31926. landmarks are: tensor([[[ 0.6089, -0.3948, 1.7383, -0.8617, -0.2536, -1.2390, 0.6009,
  31927. 0.1159],
  31928. [ 0.5327, -0.4381, 1.2880, -0.6308, -0.6866, -0.5307, 0.2658,
  31929. 0.3417],
  31930. [ 0.6125, -0.4369, 1.9173, -0.5384, -0.2594, -0.8386, 0.9741,
  31931. 0.1821],
  31932. [ 0.5680, -0.4417, 1.8365, -0.0740, -0.4941, -0.0227, 0.5045,
  31933. 0.1525],
  31934. [ 0.6134, -0.3910, 1.4497, 0.3546, -0.3517, -0.0919, 0.3296,
  31935. 0.5239],
  31936. [ 0.5438, -0.4206, 1.5189, -0.4537, -0.6115, -0.6231, 0.4365,
  31937. 0.5491],
  31938. [ 0.6249, -0.4352, 1.8018, 0.2545, -0.6173, -0.1997, 0.6401,
  31939. 0.0291],
  31940. [ 0.5610, -0.4381, 1.7730, 0.0390, -0.4326, -0.1458, 0.1794,
  31941. 0.3777]]], device='cuda:0')
  31942. loss_train_step before backward: tensor(0.0072, device='cuda:0', grad_fn=<MseLossBackward>)
  31943. loss_train_step after backward: tensor(0.0072, device='cuda:0', grad_fn=<MseLossBackward>)
  31944. loss_train: 0.49751824559643865
  31945. step: 32
  31946. running loss: 0.015547445174888708
  31947. Train Steps: 32/90 Loss: 0.0155 torch.Size([8, 600, 800])
  31948. torch.Size([8, 8])
  31949. tensor([[0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261],
  31950. [0.6173, 0.4114, 0.7325, 0.2500, 0.4213, 0.1917, 0.5338, 0.5700],
  31951. [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207],
  31952. [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817],
  31953. [0.6357, 0.4118, 0.8400, 0.2500, 0.5413, 0.1633, 0.6725, 0.5586],
  31954. [0.6346, 0.4092, 0.7712, 0.5917, 0.4037, 0.4767, 0.7343, 0.5725],
  31955. [0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583],
  31956. [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209]],
  31957. device='cuda:0', dtype=torch.float64)
  31958. predictions are: tensor([[ 0.7685, -0.2554, 1.8790, -0.0951, -0.5094, 0.1064, 0.3453, 0.0740],
  31959. [ 0.2122, -0.6466, 1.1111, -1.1771, -0.4408, -1.2101, 0.1642, 0.2664],
  31960. [ 0.5185, -0.4993, 1.5777, -1.2015, -0.3875, -1.1538, 0.6529, 0.0108],
  31961. [ 0.6000, -0.4405, 1.6702, 0.3551, -0.3816, 0.2087, 0.2137, -0.0261],
  31962. [ 0.5316, -0.4604, 1.7136, -1.2135, 0.0900, -1.4193, 0.7995, 0.1499],
  31963. [ 0.7085, -0.3692, 1.4491, 0.1493, -0.6077, 0.0302, 0.9614, 0.2498],
  31964. [ 0.6815, -0.3494, 1.7495, -0.4778, -0.7476, -0.3255, 0.4961, 0.2621],
  31965. [ 0.6221, -0.3105, 1.7768, -0.2247, -0.3396, -0.9656, 0.3741, 0.4538]],
  31966. device='cuda:0', grad_fn=<AddmmBackward>)
  31967. landmarks are: tensor([[[ 0.5746, -0.3623, 1.7961, -0.1150, -0.3691, -0.0380, 0.2208,
  31968. 0.1439],
  31969. [ 0.5652, -0.3858, 1.0975, -1.1312, -0.3402, -1.4006, 0.1794,
  31970. 0.3469],
  31971. [ 0.6132, -0.4248, 1.5247, -1.1620, -0.2882, -1.3159, 0.6545,
  31972. 0.1193],
  31973. [ 0.5193, -0.4699, 1.5767, 0.4008, -0.2420, 0.0774, 0.1178,
  31974. -0.0610],
  31975. [ 0.6504, -0.3840, 1.5940, -1.1312, 0.2141, -1.5315, 0.8205,
  31976. 0.2944],
  31977. [ 0.6454, -0.3960, 1.2764, 0.4470, -0.4210, -0.0842, 1.1057,
  31978. 0.3585],
  31979. [ 0.5761, -0.3966, 1.6171, -0.4845, -0.6346, -0.4691, 0.4739,
  31980. 0.2930],
  31981. [ 0.6174, -0.3118, 1.6402, -0.2074, -0.1958, -1.0927, 0.2267,
  31982. 0.5822]]], device='cuda:0')
  31983. loss_train_step before backward: tensor(0.0149, device='cuda:0', grad_fn=<MseLossBackward>)
  31984. loss_train_step after backward: tensor(0.0149, device='cuda:0', grad_fn=<MseLossBackward>)
  31985. loss_train: 0.5123806581832469
  31986. step: 33
  31987. running loss: 0.01552668661161354
  31988. Train Steps: 33/90 Loss: 0.0155 torch.Size([8, 600, 800])
  31989. torch.Size([8, 8])
  31990. tensor([[0.6042, 0.3990, 0.6831, 0.2875, 0.3500, 0.3133, 0.5143, 0.5510],
  31991. [0.6307, 0.4045, 0.8025, 0.5833, 0.3775, 0.4867, 0.6892, 0.5459],
  31992. [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633],
  31993. [0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204],
  31994. [0.6252, 0.4158, 0.8988, 0.4083, 0.3788, 0.4783, 0.6225, 0.5633],
  31995. [0.6038, 0.3946, 0.8413, 0.4883, 0.3563, 0.4550, 0.5266, 0.4693],
  31996. [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367],
  31997. [0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350]],
  31998. device='cuda:0', dtype=torch.float64)
  31999. predictions are: tensor([[ 0.6101, -0.3861, 1.0465, -1.0171, -0.5287, -0.9288, 0.1636, 0.3035],
  32000. [ 0.7595, -0.3106, 1.5950, 0.3544, -0.4964, -0.0664, 0.8690, 0.1603],
  32001. [ 0.5309, -0.4411, 1.7549, -0.7236, -0.5169, -0.8883, 0.3679, 0.1458],
  32002. [ 0.7765, -0.3054, 1.7333, 0.3088, -0.3553, 0.1623, 0.7348, 0.0983],
  32003. [ 0.6907, -0.3104, 1.9679, -0.2774, -0.5666, -0.0387, 0.7282, 0.2893],
  32004. [ 0.5866, -0.3977, 1.7523, 0.0307, -0.5445, -0.1638, 0.2781, 0.1210],
  32005. [ 0.5182, -0.4533, 1.4693, -0.9037, -0.5433, -0.7623, 0.4990, 0.2418],
  32006. [ 0.3984, -0.5048, 1.4839, -1.1185, -0.1541, -1.3903, 0.4739, 0.2252]],
  32007. device='cuda:0', grad_fn=<AddmmBackward>)
  32008. landmarks are: tensor([[[ 0.5049, -0.4428, 0.8692, -0.9581, -0.6693, -0.8386, 0.0897,
  32009. 0.2589],
  32010. [ 0.6273, -0.4177, 1.4208, 0.4085, -0.5423, -0.0380, 0.8973,
  32011. 0.2356],
  32012. [ 0.5944, -0.4008, 1.6748, -0.9002, -0.5711, -0.8848, 0.2776,
  32013. 0.3161],
  32014. [ 0.6042, -0.4273, 1.7198, 0.2184, -0.3478, 0.1149, 0.8062,
  32015. 0.1176],
  32016. [ 0.6017, -0.3654, 1.8654, -0.3998, -0.5365, -0.0765, 0.5894,
  32017. 0.3161],
  32018. [ 0.5030, -0.4631, 1.5998, -0.0303, -0.6404, -0.1843, 0.1465,
  32019. -0.1181],
  32020. [ 0.5697, -0.4442, 1.3529, -0.9515, -0.5774, -0.7801, 0.5253,
  32021. 0.1931],
  32022. [ 0.5768, -0.4031, 1.3838, -1.1527, -0.2188, -1.4216, 0.4379,
  32023. 0.1850]]], device='cuda:0')
  32024. loss_train_step before backward: tensor(0.0088, device='cuda:0', grad_fn=<MseLossBackward>)
  32025. loss_train_step after backward: tensor(0.0088, device='cuda:0', grad_fn=<MseLossBackward>)
  32026. loss_train: 0.5212206044234335
  32027. step: 34
  32028. running loss: 0.01533001777715981
  32029.  
  32030. Train Steps: 34/90 Loss: 0.0153 torch.Size([8, 600, 800])
  32031. torch.Size([8, 8])
  32032. tensor([[0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650],
  32033. [ nan, nan, 0.7612, 0.3250, 0.4037, 0.2533, 0.5438, 0.5767],
  32034. [0.6200, 0.3993, 0.8519, 0.4923, 0.3962, 0.4717, 0.6013, 0.5433],
  32035. [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617],
  32036. [0.6226, 0.4103, 0.8575, 0.3450, 0.4388, 0.2067, 0.5787, 0.5383],
  32037. [0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153],
  32038. [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250],
  32039. [0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719]],
  32040. device='cuda:0', dtype=torch.float64)
  32041. predictions are: tensor([[ 0.7662, -0.2341, 1.5781, -0.2909, -0.6309, -0.7669, 0.4134, 0.3638],
  32042. [-1.5719, -1.7703, 1.2770, -0.8794, -0.4331, -1.0268, 0.3747, 0.3496],
  32043. [ 0.7161, -0.3501, 1.6823, 0.1277, -0.5211, -0.0683, 0.6721, 0.1472],
  32044. [ 0.6450, -0.3428, 1.7448, 0.0048, -0.3701, 0.0794, 0.4939, 0.2061],
  32045. [ 0.6709, -0.3055, 1.6780, -0.6860, -0.2785, -1.2521, 0.4511, 0.2015],
  32046. [ 0.6167, -0.3962, 1.6691, 0.1726, -0.2480, 0.1969, 0.6600, 0.0819],
  32047. [ 0.7405, -0.2789, 1.8078, 0.0298, -0.4354, 0.2576, 0.7490, 0.0836],
  32048. [ 0.5992, -0.3914, 1.5422, -0.4336, -0.6631, -0.2964, 0.3682, 0.2405]],
  32049. device='cuda:0', grad_fn=<AddmmBackward>)
  32050. landmarks are: tensor([[[ 0.5894, -0.3503, 1.6402, -0.3614, -0.5827, -0.7925, 0.3238,
  32051. 0.3238],
  32052. [-2.2859, -2.2859, 1.2303, -0.7848, -0.4210, -1.1158, 0.2256,
  32053. 0.3777],
  32054. [ 0.5777, -0.4415, 1.6491, -0.0122, -0.4557, -0.1073, 0.4912,
  32055. 0.2237],
  32056. [ 0.5637, -0.4143, 1.7519, -0.0787, -0.3055, -0.0149, 0.3758,
  32057. 0.3084],
  32058. [ 0.5898, -0.3908, 1.6748, -0.6924, -0.2594, -1.3313, 0.3873,
  32059. 0.2006],
  32060. [ 0.5548, -0.4682, 1.7309, 0.0966, -0.1394, 0.1757, 0.5045,
  32061. 0.0942],
  32062. [ 0.5991, -0.3803, 1.8018, -0.0534, -0.3460, 0.1852, 0.5374,
  32063. 0.1390],
  32064. [ 0.5367, -0.4294, 1.5709, -0.4999, -0.6693, -0.3075, 0.2455,
  32065. 0.3559]]], device='cuda:0')
  32066. loss_train_step before backward: tensor(0.0195, device='cuda:0', grad_fn=<MseLossBackward>)
  32067. loss_train_step after backward: tensor(0.0195, device='cuda:0', grad_fn=<MseLossBackward>)
  32068. loss_train: 0.5407532160170376
  32069. step: 35
  32070. running loss: 0.015450091886201075
  32071. Train Steps: 35/90 Loss: 0.0155 torch.Size([8, 600, 800])
  32072. torch.Size([8, 8])
  32073. tensor([[0.6282, 0.4034, 0.7830, 0.2080, 0.4532, 0.2080, 0.6404, 0.5323],
  32074. [0.6034, 0.4011, 0.7350, 0.2533, 0.3438, 0.3367, 0.5516, 0.5084],
  32075. [0.6273, 0.4100, 0.7137, 0.2133, 0.4000, 0.2650, 0.6075, 0.5633],
  32076. [ nan, nan, 0.7192, 0.2346, 0.4037, 0.2050, 0.5138, 0.5650],
  32077. [0.6286, 0.4097, 0.8107, 0.2414, 0.4425, 0.2483, 0.6745, 0.5385],
  32078. [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329],
  32079. [0.6219, 0.3934, 0.8688, 0.5267, 0.4313, 0.4967, 0.5988, 0.4983],
  32080. [0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283]],
  32081. device='cuda:0', dtype=torch.float64)
  32082. predictions are: tensor([[ 0.5688, -0.4102, 1.3362, -1.0214, -0.2337, -1.2387, 0.5933, 0.2205],
  32083. [ 0.5588, -0.4107, 1.1360, -0.9133, -0.6114, -0.6363, 0.2814, 0.1120],
  32084. [ 0.4255, -0.4792, 1.0315, -1.0336, -0.4536, -1.0128, 0.4469, 0.3514],
  32085. [-2.0111, -2.0337, 1.0375, -1.0228, -0.4102, -1.0963, 0.2566, 0.2898],
  32086. [ 0.5824, -0.3836, 1.4738, -0.8859, -0.2546, -1.1444, 0.8461, 0.2488],
  32087. [ 0.6641, -0.3520, 1.6249, -0.4234, -0.6472, -0.4652, 0.5174, 0.2492],
  32088. [ 0.7156, -0.3546, 1.7393, 0.3704, -0.3914, 0.1171, 0.5668, -0.0235],
  32089. [ 0.6197, -0.3639, 1.9170, -0.2692, -0.5723, -0.3346, 0.6970, 0.2217]],
  32090. device='cuda:0', grad_fn=<AddmmBackward>)
  32091. landmarks are: tensor([[[ 0.6158, -0.4225, 1.3307, -1.3253, -0.1924, -1.3252, 0.6721,
  32092. 0.1727],
  32093. [ 0.5009, -0.4333, 1.1090, -1.1158, -0.6982, -0.7309, 0.2617,
  32094. 0.0622],
  32095. [ 0.6115, -0.3924, 1.0109, -1.3005, -0.4383, -1.0619, 0.5201,
  32096. 0.3161],
  32097. [-2.2859, -2.2859, 1.0361, -1.2021, -0.4210, -1.3390, 0.0871,
  32098. 0.3238],
  32099. [ 0.6174, -0.3936, 1.4586, -1.1709, -0.2420, -1.1389, 0.8296,
  32100. 0.2012],
  32101. [ 0.5680, -0.4393, 1.5920, -0.6672, -0.6453, -0.5457, 0.5149,
  32102. 0.1753],
  32103. [ 0.5864, -0.4690, 1.7268, 0.1467, -0.2940, 0.0082, 0.4797,
  32104. 0.0159],
  32105. [ 0.6073, -0.4058, 1.8885, -0.4999, -0.5942, -0.4768, 0.6413,
  32106. 0.1544]]], device='cuda:0')
  32107. loss_train_step before backward: tensor(0.0145, device='cuda:0', grad_fn=<MseLossBackward>)
  32108. loss_train_step after backward: tensor(0.0145, device='cuda:0', grad_fn=<MseLossBackward>)
  32109. loss_train: 0.5552862132899463
  32110. step: 36
  32111. running loss: 0.015424617035831843
  32112. Train Steps: 36/90 Loss: 0.0154 torch.Size([8, 600, 800])
  32113. torch.Size([8, 8])
  32114. tensor([[0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833],
  32115. [0.6336, 0.4191, 0.8938, 0.5167, 0.3937, 0.3517, 0.7343, 0.5748],
  32116. [0.6179, 0.4008, 0.7505, 0.2678, 0.4368, 0.1891, 0.5831, 0.5263],
  32117. [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217],
  32118. [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374],
  32119. [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567],
  32120. [0.6339, 0.4102, 0.9088, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390],
  32121. [0.6200, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183]],
  32122. device='cuda:0', dtype=torch.float64)
  32123. predictions are: tensor([[ 0.6712, -0.3344, 0.8852, -0.8259, -0.5781, -0.8849, 0.2836, 0.4461],
  32124. [ 0.6448, -0.3840, 1.7873, 0.1943, -0.3306, -0.5015, 0.9702, 0.3116],
  32125. [ 0.6883, -0.3586, 1.1804, -1.1055, -0.3317, -1.3468, 0.3378, 0.1692],
  32126. [ 0.6501, -0.3811, 1.7949, 0.0383, -0.4651, 0.4288, 0.6228, 0.0843],
  32127. [ 0.5457, -0.4620, 1.6727, -0.5455, -0.5818, -0.7265, 0.4503, 0.1704],
  32128. [ 0.5400, -0.4457, 1.6360, -0.1206, -0.5851, -0.3272, 0.2805, 0.3501],
  32129. [ 0.6369, -0.3821, 1.8384, -0.0908, -0.4472, 0.3384, 1.0631, 0.2493],
  32130. [ 0.4643, -0.4946, 1.4443, -0.9013, -0.4899, -0.8918, 0.3928, 0.1161]],
  32131. device='cuda:0', grad_fn=<AddmmBackward>)
  32132. landmarks are: tensor([[[ 0.5614, -0.4080, 0.7741, -0.8848, -0.5423, -0.9156, 0.3584,
  32133. 0.4085],
  32134. [ 0.6405, -0.3503, 1.8423, 0.1005, -0.4672, -0.6616, 1.1057,
  32135. 0.3692],
  32136. [ 0.5680, -0.4345, 1.1806, -1.0490, -0.2683, -1.4127, 0.4074,
  32137. 0.1449],
  32138. [ 0.6047, -0.4209, 1.7557, -0.0303, -0.4845, 0.2545, 0.6587,
  32139. 0.1236],
  32140. [ 0.5777, -0.4416, 1.7044, -0.5827, -0.5962, -0.8361, 0.4862,
  32141. 0.1963],
  32142. [ 0.5432, -0.4336, 1.7095, -0.1766, -0.5942, -0.4845, 0.3007,
  32143. 0.2853],
  32144. [ 0.6421, -0.3912, 1.9115, -0.0842, -0.4730, 0.1544, 1.1824,
  32145. 0.2035],
  32146. [ 0.5781, -0.4129, 1.4142, -0.9057, -0.5115, -0.9937, 0.4621,
  32147. 0.1080]]], device='cuda:0')
  32148. loss_train_step before backward: tensor(0.0055, device='cuda:0', grad_fn=<MseLossBackward>)
  32149. loss_train_step after backward: tensor(0.0055, device='cuda:0', grad_fn=<MseLossBackward>)
  32150. loss_train: 0.5607698620297015
  32151. step: 37
  32152. running loss: 0.015155942217018959
  32153. Train Steps: 37/90 Loss: 0.0152 torch.Size([8, 600, 800])
  32154. torch.Size([8, 8])
  32155. tensor([[0.6085, 0.4008, 0.8588, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283],
  32156. [0.6256, 0.4199, 0.8638, 0.5800, 0.3987, 0.4383, 0.5600, 0.5950],
  32157. [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817],
  32158. [0.6275, 0.4081, 0.8063, 0.2017, 0.4825, 0.1583, 0.6156, 0.4869],
  32159. [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683],
  32160. [0.6250, 0.4103, 0.8950, 0.4400, 0.3912, 0.5650, 0.6050, 0.5133],
  32161. [0.6308, 0.3990, 0.8688, 0.5183, 0.3950, 0.4983, 0.6388, 0.5133],
  32162. [0.6127, 0.4115, 0.7163, 0.2883, 0.3625, 0.2950, 0.5327, 0.5495]],
  32163. device='cuda:0', dtype=torch.float64)
  32164. predictions are: tensor([[ 0.5668, -0.4175, 1.6320, 0.0660, -0.1801, -0.0683, 0.3313, 0.2419],
  32165. [ 0.5908, -0.3833, 1.6107, 0.3380, -0.5683, -0.2802, 0.4929, 0.5329],
  32166. [ 0.4183, -0.5385, 1.5019, 0.3227, -0.3926, 0.0037, 0.3500, 0.1237],
  32167. [ 0.6590, -0.4055, 1.4546, -1.4122, -0.1601, -1.6383, 0.7569, 0.0479],
  32168. [ 0.5319, -0.4480, 1.6853, -0.0332, -0.2487, -0.0303, 0.4169, 0.3119],
  32169. [ 0.5453, -0.4142, 1.7662, -0.3417, -0.5738, 0.2874, 0.7600, 0.2159],
  32170. [ 0.5614, -0.4496, 1.7109, 0.0886, -0.5932, -0.0110, 0.9051, 0.1374],
  32171. [ 0.5893, -0.4026, 1.0282, -1.0175, -0.7196, -0.9125, 0.2854, 0.3452]],
  32172. device='cuda:0', grad_fn=<AddmmBackward>)
  32173. landmarks are: tensor([[[ 0.5245, -0.4347, 1.6806, 0.1159, 0.0046, 0.0129, 0.1044,
  32174. 0.1544],
  32175. [ 0.6038, -0.3464, 1.7037, 0.3931, -0.4441, -0.2613, 0.3007,
  32176. 0.4624],
  32177. [ 0.5193, -0.4699, 1.5767, 0.4008, -0.2420, 0.0774, 0.1178,
  32178. -0.0610],
  32179. [ 0.6124, -0.4010, 1.4381, -1.3544, -0.0573, -1.5546, 0.5573,
  32180. -0.0369],
  32181. [ 0.5359, -0.4370, 1.7095, -0.0303, -0.0804, -0.0380, 0.1044,
  32182. 0.3392],
  32183. [ 0.6010, -0.3909, 1.8480, -0.2536, -0.4788, 0.3238, 0.5085,
  32184. 0.0851],
  32185. [ 0.6275, -0.4430, 1.7268, 0.1082, -0.4614, 0.0159, 0.6644,
  32186. 0.0851],
  32187. [ 0.5442, -0.3855, 1.0224, -0.9541, -0.6115, -0.9233, 0.1745,
  32188. 0.2522]]], device='cuda:0')
  32189. loss_train_step before backward: tensor(0.0117, device='cuda:0', grad_fn=<MseLossBackward>)
  32190. loss_train_step after backward: tensor(0.0117, device='cuda:0', grad_fn=<MseLossBackward>)
  32191. loss_train: 0.5724910940043628
  32192. step: 38
  32193. running loss: 0.01506555510537797
  32194.  
  32195. Train Steps: 38/90 Loss: 0.0151 torch.Size([8, 600, 800])
  32196. torch.Size([8, 8])
  32197. tensor([[0.6102, 0.4005, 0.8688, 0.5100, 0.4813, 0.5400, 0.5404, 0.5064],
  32198. [0.6274, 0.4117, 0.8100, 0.5801, 0.4000, 0.4583, 0.5582, 0.6118],
  32199. [0.6193, 0.4050, 0.7313, 0.2433, 0.4075, 0.2117, 0.5649, 0.5656],
  32200. [0.6226, 0.4125, 0.8800, 0.4900, 0.4512, 0.5600, 0.6275, 0.5517],
  32201. [0.6268, 0.4052, 0.8175, 0.2250, 0.4688, 0.1917, 0.6375, 0.5267],
  32202. [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377],
  32203. [0.6068, 0.3963, 0.8650, 0.4317, 0.4037, 0.5083, 0.5253, 0.4999],
  32204. [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290]],
  32205. device='cuda:0', dtype=torch.float64)
  32206. predictions are: tensor([[ 0.5787, -0.4111, 1.6932, 0.0974, -0.2177, 0.0952, 0.3047, 0.1366],
  32207. [ 0.3868, -0.5064, 1.4697, 0.3451, -0.5825, -0.1846, 0.5391, 0.5011],
  32208. [ 0.5121, -0.4486, 1.0925, -1.1847, -0.4936, -1.2274, 0.4309, 0.2811],
  32209. [ 0.5534, -0.3974, 1.7306, -0.0548, -0.2913, 0.2534, 0.6129, 0.3512],
  32210. [ 0.6356, -0.4325, 1.4172, -1.2141, -0.2466, -1.5455, 0.7136, 0.1865],
  32211. [ 0.6097, -0.4405, 1.6402, 0.2651, -0.5443, -0.1034, 0.6511, 0.1993],
  32212. [ 0.5489, -0.4402, 1.6364, -0.2779, -0.5507, 0.0573, 0.3002, 0.0735],
  32213. [ 0.5650, -0.4668, 1.2234, -1.2461, -0.3808, -1.2864, 0.7092, 0.1740]],
  32214. device='cuda:0', grad_fn=<AddmmBackward>)
  32215. landmarks are: tensor([[[ 0.5328, -0.4361, 1.7268, 0.0697, -0.0630, 0.2083, 0.2103,
  32216. 0.0532],
  32217. [ 0.6121, -0.3844, 1.4556, 0.3936, -0.4383, -0.1689, 0.2925,
  32218. 0.5401],
  32219. [ 0.5746, -0.4153, 1.0917, -1.1620, -0.4037, -1.3082, 0.3234,
  32220. 0.3267],
  32221. [ 0.5897, -0.3804, 1.7788, -0.0226, -0.2016, 0.3007, 0.6125,
  32222. 0.2622],
  32223. [ 0.6092, -0.4143, 1.4901, -1.2467, -0.1208, -1.4006, 0.6587,
  32224. 0.1467],
  32225. [ 0.5775, -0.4607, 1.6741, 0.1962, -0.4036, -0.1212, 0.4588,
  32226. 0.1979],
  32227. [ 0.5168, -0.4556, 1.7095, -0.2921, -0.4210, 0.0620, 0.1404,
  32228. 0.0231],
  32229. [ 0.6125, -0.4273, 1.2807, -1.3253, -0.2574, -1.2542, 0.6864,
  32230. 0.1575]]], device='cuda:0')
  32231. loss_train_step before backward: tensor(0.0072, device='cuda:0', grad_fn=<MseLossBackward>)
  32232. loss_train_step after backward: tensor(0.0072, device='cuda:0', grad_fn=<MseLossBackward>)
  32233. loss_train: 0.5796950873918831
  32234. step: 39
  32235. running loss: 0.014863976599791875
  32236. Train Steps: 39/90 Loss: 0.0149 torch.Size([8, 600, 800])
  32237. torch.Size([8, 8])
  32238. tensor([[0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148],
  32239. [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200],
  32240. [0.6259, 0.4156, 0.8812, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960],
  32241. [0.6110, 0.3984, 0.8750, 0.4933, 0.4625, 0.4950, 0.5578, 0.5676],
  32242. [0.6104, 0.4029, 0.8738, 0.4900, 0.4088, 0.4533, 0.5070, 0.5510],
  32243. [0.6273, 0.4105, 0.8988, 0.4517, 0.3912, 0.2550, 0.5894, 0.4811],
  32244. [0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153],
  32245. [0.6203, 0.4078, 0.8800, 0.5083, 0.3900, 0.5000, 0.6100, 0.5583]],
  32246. device='cuda:0', dtype=torch.float64)
  32247. predictions are: tensor([[ 0.4518, -0.5226, 1.5759, 0.0545, -0.5464, -0.0713, 0.5453, 0.1484],
  32248. [ 0.5803, -0.3962, 1.6064, -0.1914, -0.6144, -0.1679, 0.5165, 0.2878],
  32249. [ 0.5458, -0.4359, 1.6392, -0.9572, -0.1208, -1.4111, 0.6136, 0.1508],
  32250. [ 0.5890, -0.4114, 1.5700, -0.1026, -0.1931, 0.0397, 0.3496, 0.4049],
  32251. [ 0.5104, -0.4378, 1.5383, -0.0344, -0.4701, -0.0894, 0.2003, 0.3153],
  32252. [ 0.5813, -0.4163, 1.6899, -0.3295, -0.5267, -1.1116, 0.5762, 0.1261],
  32253. [ 0.4640, -0.5030, 1.4995, 0.0286, -0.2245, 0.1700, 0.5783, 0.2438],
  32254. [ 0.5282, -0.4285, 1.5917, -0.0356, -0.4829, 0.0265, 0.5645, 0.2748]],
  32255. device='cuda:0', grad_fn=<AddmmBackward>)
  32256. landmarks are: tensor([[[ 5.7725e-01, -4.3156e-01, 1.7399e+00, 1.2871e-01, -5.1531e-01,
  32257. -8.1749e-02, 4.3131e-01, 9.1941e-02],
  32258. [ 5.8655e-01, -3.9731e-01, 1.8423e+00, -6.8822e-02, -5.1917e-01,
  32259. -2.3048e-01, 4.1617e-01, 1.1594e-01],
  32260. [ 6.0520e-01, -3.6628e-01, 1.7845e+00, -8.1555e-01, -8.0370e-02,
  32261. -1.4237e+00, 5.8660e-01, 5.0889e-03],
  32262. [ 5.3637e-01, -4.4573e-01, 1.7557e+00, -7.2363e-03, -1.4965e-01,
  32263. 4.6189e-04, 2.9049e-01, 3.3573e-01],
  32264. [ 5.3343e-01, -4.2517e-01, 1.7499e+00, -2.2633e-02, -3.9792e-01,
  32265. -1.9199e-01, 5.5769e-02, 2.5891e-01],
  32266. [ 6.1161e-01, -3.8976e-01, 1.8654e+00, -1.9969e-01, -4.7875e-01,
  32267. -1.1081e+00, 4.3668e-01, -6.3661e-02],
  32268. [ 5.5484e-01, -4.6823e-01, 1.7309e+00, 9.6578e-02, -1.3942e-01,
  32269. 1.7573e-01, 5.0451e-01, 9.4188e-02],
  32270. [ 5.7939e-01, -4.0231e-01, 1.7788e+00, 6.2048e-02, -4.8453e-01,
  32271. 2.3557e-02, 5.3164e-01, 2.9299e-01]]], device='cuda:0')
  32272. loss_train_step before backward: tensor(0.0105, device='cuda:0', grad_fn=<MseLossBackward>)
  32273. loss_train_step after backward: tensor(0.0105, device='cuda:0', grad_fn=<MseLossBackward>)
  32274. loss_train: 0.5902246865443885
  32275. step: 40
  32276. running loss: 0.014755617163609713
  32277. Train Steps: 40/90 Loss: 0.0148 torch.Size([8, 600, 800])
  32278. torch.Size([8, 8])
  32279. tensor([[0.6204, 0.4110, 0.7913, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367],
  32280. [0.6040, 0.4002, 0.7338, 0.2267, 0.3975, 0.2100, 0.5231, 0.4778],
  32281. [0.6127, 0.4118, 0.8650, 0.5083, 0.4088, 0.5367, 0.5300, 0.5456],
  32282. [ nan, nan, 0.6935, 0.1930, 0.4150, 0.2250, 0.5450, 0.5650],
  32283. [0.6157, 0.4102, 0.8513, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890],
  32284. [0.6218, 0.4185, 0.7338, 0.2650, 0.4625, 0.1950, 0.5687, 0.5800],
  32285. [0.6250, 0.4110, 0.7238, 0.2067, 0.4263, 0.1883, 0.5625, 0.5633],
  32286. [0.6203, 0.4073, 0.8189, 0.2398, 0.4400, 0.2054, 0.5929, 0.5501]],
  32287. device='cuda:0', dtype=torch.float64)
  32288. predictions are: tensor([[ 0.6170, -0.4152, 1.4426, -0.9766, -0.4147, -0.9239, 0.6348, 0.1055],
  32289. [ 0.6492, -0.3891, 1.0826, -1.1782, -0.4910, -1.1525, 0.3329, -0.0206],
  32290. [ 0.5339, -0.4025, 1.6845, 0.1729, -0.3956, 0.3507, 0.3930, 0.1802],
  32291. [-1.7459, -1.8893, 1.0771, -1.0959, -0.2213, -1.1079, 0.4450, 0.2548],
  32292. [ 0.4224, -0.4749, 1.6062, -0.3164, -0.6632, -0.3488, 0.2345, 0.2554],
  32293. [ 0.5959, -0.3686, 1.0819, -0.8909, -0.2060, -1.2524, 0.4956, 0.3669],
  32294. [ 0.4865, -0.4776, 1.0770, -1.1057, -0.3190, -1.2749, 0.5400, 0.2507],
  32295. [ 0.5668, -0.4227, 1.4136, -0.9592, -0.2372, -1.2127, 0.6334, 0.2023]],
  32296. device='cuda:0', grad_fn=<AddmmBackward>)
  32297. landmarks are: tensor([[[ 0.5796, -0.3878, 1.3688, -1.0542, -0.4095, -1.1312, 0.5894,
  32298. 0.1929],
  32299. [ 0.5037, -0.4375, 1.1032, -1.2390, -0.4499, -1.3159, 0.1301,
  32300. -0.0791],
  32301. [ 0.5443, -0.3840, 1.7095, 0.0620, -0.3979, 0.1929, 0.1622,
  32302. 0.2341],
  32303. [-2.2859, -2.2859, 0.9175, -1.3947, -0.3691, -1.2467, 0.2314,
  32304. 0.3238],
  32305. [ 0.5581, -0.3912, 1.6460, -0.5230, -0.6173, -0.5923, 0.0681,
  32306. 0.4348],
  32307. [ 0.5862, -0.3530, 1.1032, -1.0619, -0.1497, -1.3852, 0.3411,
  32308. 0.3931],
  32309. [ 0.6010, -0.3875, 1.0570, -1.3313, -0.3171, -1.4160, 0.3122,
  32310. 0.3161],
  32311. [ 0.5792, -0.4048, 1.4965, -1.1781, -0.2534, -1.3371, 0.4528,
  32312. 0.2549]]], device='cuda:0')
  32313. loss_train_step before backward: tensor(0.0225, device='cuda:0', grad_fn=<MseLossBackward>)
  32314. loss_train_step after backward: tensor(0.0225, device='cuda:0', grad_fn=<MseLossBackward>)
  32315. loss_train: 0.6126899658702314
  32316. step: 41
  32317. running loss: 0.014943657704151986
  32318. Train Steps: 41/90 Loss: 0.0149 torch.Size([8, 600, 800])
  32319. torch.Size([8, 8])
  32320. tensor([[0.6200, 0.4098, 0.8237, 0.2917, 0.4012, 0.2967, 0.6000, 0.5683],
  32321. [0.6125, 0.4010, 0.8650, 0.4567, 0.3663, 0.3900, 0.5600, 0.5567],
  32322. [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117],
  32323. [0.6254, 0.3993, 0.8988, 0.4767, 0.3987, 0.5517, 0.6955, 0.5285],
  32324. [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208],
  32325. [0.6176, 0.4030, 0.8850, 0.4850, 0.3688, 0.4050, 0.5312, 0.5783],
  32326. [0.6204, 0.4110, 0.7913, 0.2667, 0.4062, 0.2500, 0.6225, 0.5367],
  32327. [0.6197, 0.3930, 0.8793, 0.4736, 0.4152, 0.5464, 0.6308, 0.5235]],
  32328. device='cuda:0', dtype=torch.float64)
  32329. predictions are: tensor([[ 0.7854, -0.2991, 1.3389, -0.9233, -0.3140, -0.9295, 0.4933, 0.3211],
  32330. [ 0.4851, -0.4815, 1.5754, -0.1447, -0.5025, -0.4698, 0.2705, 0.3721],
  32331. [ 0.4202, -0.5532, 1.6440, -0.7332, -0.4721, -0.9302, 0.4453, 0.0863],
  32332. [ 0.5706, -0.4244, 1.7125, -0.0341, -0.3242, 0.3171, 0.8492, 0.1820],
  32333. [ 0.4740, -0.5121, 1.3715, -1.0159, -0.2918, -1.2467, 0.4170, 0.1064],
  32334. [ 0.6055, -0.4053, 1.6585, 0.0100, -0.4876, -0.3127, 0.2278, 0.2346],
  32335. [ 0.6458, -0.3923, 1.3370, -1.0757, -0.2966, -1.1584, 0.5205, 0.1678],
  32336. [ 0.4434, -0.5174, 1.6545, -0.0345, -0.2625, 0.2169, 0.5201, 0.1693]],
  32337. device='cuda:0', grad_fn=<AddmmBackward>)
  32338. landmarks are: tensor([[[ 0.5778, -0.3930, 1.5189, -0.9387, -0.4326, -0.9156, 0.4855,
  32339. 0.3392],
  32340. [ 0.5432, -0.4336, 1.7095, -0.1766, -0.5942, -0.4845, 0.3007,
  32341. 0.2853],
  32342. [ 0.5799, -0.4329, 1.7210, -0.7694, -0.5711, -0.8771, 0.3988,
  32343. 0.0774],
  32344. [ 0.6026, -0.4417, 1.8654, -0.0842, -0.4441, 0.2622, 0.9265,
  32345. 0.1554],
  32346. [ 0.5885, -0.4429, 1.4266, -0.9926, -0.4383, -1.2313, 0.4228,
  32347. 0.1195],
  32348. [ 0.5667, -0.4246, 1.8018, -0.0457, -0.5827, -0.4152, 0.1679,
  32349. 0.3854],
  32350. [ 0.5796, -0.3878, 1.3688, -1.0542, -0.4095, -1.1312, 0.5894,
  32351. 0.1929],
  32352. [ 0.5762, -0.4706, 1.7754, -0.0984, -0.3680, 0.2380, 0.6277,
  32353. 0.1322]]], device='cuda:0')
  32354. loss_train_step before backward: tensor(0.0071, device='cuda:0', grad_fn=<MseLossBackward>)
  32355. loss_train_step after backward: tensor(0.0071, device='cuda:0', grad_fn=<MseLossBackward>)
  32356. loss_train: 0.6198296854272485
  32357. step: 42
  32358. running loss: 0.014757849653029726
  32359.  
  32360. Train Steps: 42/90 Loss: 0.0148 torch.Size([8, 600, 800])
  32361. torch.Size([8, 8])
  32362. tensor([[0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833],
  32363. [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367],
  32364. [0.6102, 0.4020, 0.8638, 0.3717, 0.3625, 0.5017, 0.6038, 0.5500],
  32365. [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649],
  32366. [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413],
  32367. [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973],
  32368. [0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719],
  32369. [0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5988, 0.5700]],
  32370. device='cuda:0', dtype=torch.float64)
  32371. predictions are: tensor([[ 0.6620, -0.3627, 0.8645, -0.8953, -0.4624, -1.0615, 0.2567, 0.4448],
  32372. [ 0.5483, -0.4709, 1.3646, -0.9783, -0.4312, -0.9092, 0.4041, 0.2567],
  32373. [ 0.4148, -0.5180, 1.6699, -0.6309, -0.4510, -0.0706, 0.5017, 0.2699],
  32374. [ 0.6890, -0.3866, 1.6303, 0.5067, -0.3901, -0.0298, 0.3656, -0.0725],
  32375. [ 0.2361, -0.6919, 2.0030, -0.9078, 0.1716, -1.4575, 1.0266, 0.2301],
  32376. [ 0.6916, -0.4029, 1.8254, 0.1274, -0.4539, -0.1698, 0.5178, -0.0051],
  32377. [ 0.3982, -0.5432, 1.5778, -0.5003, -0.5175, -0.3889, 0.1091, 0.2928],
  32378. [ 0.6824, -0.3615, 1.3945, -0.8919, -0.4078, -0.7204, 0.3479, 0.2505]],
  32379. device='cuda:0', grad_fn=<AddmmBackward>)
  32380. landmarks are: tensor([[[ 0.5614, -0.4080, 0.7741, -0.8848, -0.5423, -0.9156, 0.3584,
  32381. 0.4085],
  32382. [ 0.5697, -0.4442, 1.3529, -0.9515, -0.5774, -0.7801, 0.5253,
  32383. 0.1931],
  32384. [ 0.5324, -0.4293, 1.7037, -0.5692, -0.6115, 0.0313, 0.5028,
  32385. 0.2545],
  32386. [ 0.6105, -0.4360, 1.6171, 0.5162, -0.5076, 0.0159, 0.5171,
  32387. -0.1385],
  32388. [ 0.6108, -0.4201, 1.8711, -0.7848, -0.0053, -1.2236, 1.0362,
  32389. 0.2142],
  32390. [ 0.6240, -0.4321, 1.8423, 0.1852, -0.5885, -0.1612, 0.6962,
  32391. 0.0111],
  32392. [ 0.5367, -0.4294, 1.5709, -0.4999, -0.6693, -0.3075, 0.2455,
  32393. 0.3559],
  32394. [ 0.5960, -0.3888, 1.4840, -1.0095, -0.6115, -0.6231, 0.4797,
  32395. 0.3469]]], device='cuda:0')
  32396. loss_train_step before backward: tensor(0.0128, device='cuda:0', grad_fn=<MseLossBackward>)
  32397. loss_train_step after backward: tensor(0.0128, device='cuda:0', grad_fn=<MseLossBackward>)
  32398. loss_train: 0.632617355324328
  32399. step: 43
  32400. running loss: 0.014712031519170417
  32401. Train Steps: 43/90 Loss: 0.0147 torch.Size([8, 600, 800])
  32402. torch.Size([8, 8])
  32403. tensor([[0.6357, 0.4139, 0.8450, 0.5883, 0.3775, 0.4950, 0.6488, 0.5297],
  32404. [0.6211, 0.4069, 0.8750, 0.5117, 0.4150, 0.4900, 0.5875, 0.5167],
  32405. [0.6185, 0.4129, 0.8900, 0.4567, 0.3937, 0.5417, 0.5734, 0.5110],
  32406. [0.6230, 0.4152, 0.7588, 0.2283, 0.4012, 0.2883, 0.6200, 0.5767],
  32407. [0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038],
  32408. [0.6251, 0.4163, 0.8662, 0.4467, 0.3625, 0.3567, 0.6038, 0.5533],
  32409. [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833],
  32410. [0.6286, 0.4086, 0.8408, 0.2801, 0.4163, 0.2800, 0.6725, 0.5393]],
  32411. device='cuda:0', dtype=torch.float64)
  32412. predictions are: tensor([[ 0.3379, -0.5900, 1.6437, 0.3095, -0.4413, -0.1503, 0.5774, 0.1200],
  32413. [ 0.3663, -0.5576, 1.8159, -0.0059, -0.3329, -0.1042, 0.3667, 0.1079],
  32414. [ 0.6056, -0.3637, 1.8744, -0.2903, -0.3508, 0.0614, 0.2247, 0.1056],
  32415. [ 0.6311, -0.4130, 1.2440, -1.1452, -0.4054, -1.1107, 0.4740, 0.3398],
  32416. [ 0.5598, -0.4292, 1.7782, -0.1508, -0.1658, -0.0379, 0.0746, -0.0257],
  32417. [ 0.6271, -0.3747, 1.7128, -0.3489, -0.4974, -0.8574, 0.3628, 0.2517],
  32418. [ 0.6131, -0.4009, 0.8760, -0.9219, -0.4928, -1.0583, 0.2549, 0.4428],
  32419. [ 0.4163, -0.5562, 1.5630, -0.9899, -0.1836, -1.1978, 0.7956, 0.2543]],
  32420. device='cuda:0', grad_fn=<AddmmBackward>)
  32421. landmarks are: tensor([[[ 6.5036e-01, -3.7434e-01, 1.6171e+00, 4.3156e-01, -5.4226e-01,
  32422. 4.6189e-04, 7.1085e-01, 1.6077e-01],
  32423. [ 5.8279e-01, -4.0662e-01, 1.7557e+00, 7.7444e-02, -3.6905e-01,
  32424. -2.2633e-02, 4.2771e-01, 1.0054e-01],
  32425. [ 5.7113e-01, -3.7875e-01, 1.8249e+00, -1.7660e-01, -4.6721e-01,
  32426. 2.1601e-01, 3.6246e-01, 7.4222e-02],
  32427. [ 5.9151e-01, -3.6821e-01, 1.2187e+00, -1.2313e+00, -4.3256e-01,
  32428. -9.5412e-01, 5.7783e-01, 3.7768e-01],
  32429. [ 5.2494e-01, -4.4734e-01, 1.7326e+00, -9.1917e-02, -2.0162e-01,
  32430. 1.5443e-01, 1.7328e-01, 4.1158e-02],
  32431. [ 6.0162e-01, -3.6328e-01, 1.7152e+00, -2.2279e-01, -6.1155e-01,
  32432. -6.3849e-01, 5.0277e-01, 2.6990e-01],
  32433. [ 5.6143e-01, -4.0805e-01, 7.7413e-01, -8.8483e-01, -5.4226e-01,
  32434. -9.1563e-01, 3.5843e-01, 4.0847e-01],
  32435. [ 6.1742e-01, -3.9842e-01, 1.5975e+00, -9.9214e-01, -3.6328e-01,
  32436. -9.9261e-01, 8.2047e-01, 2.0505e-01]]], device='cuda:0')
  32437. loss_train_step before backward: tensor(0.0122, device='cuda:0', grad_fn=<MseLossBackward>)
  32438. loss_train_step after backward: tensor(0.0122, device='cuda:0', grad_fn=<MseLossBackward>)
  32439. loss_train: 0.6447693314403296
  32440. step: 44
  32441. running loss: 0.014653848441825672
  32442. Train Steps: 44/90 Loss: 0.0147 torch.Size([8, 600, 800])
  32443. torch.Size([8, 8])
  32444. tensor([[0.6189, 0.4049, 0.8888, 0.4417, 0.4213, 0.5200, 0.5988, 0.5633],
  32445. [ nan, nan, 0.6900, 0.1917, 0.3937, 0.2367, 0.5240, 0.5246],
  32446. [0.6186, 0.4154, 0.6825, 0.2633, 0.4150, 0.2300, 0.5713, 0.5517],
  32447. [0.6193, 0.4108, 0.7438, 0.2700, 0.3650, 0.3683, 0.6238, 0.5717],
  32448. [0.6257, 0.4024, 0.8612, 0.5352, 0.4361, 0.5253, 0.6680, 0.5166],
  32449. [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343],
  32450. [0.6268, 0.4061, 0.8350, 0.2433, 0.4575, 0.2283, 0.6350, 0.5300],
  32451. [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550]],
  32452. device='cuda:0', dtype=torch.float64)
  32453. predictions are: tensor([[ 0.6149, -0.3487, 1.7814, -0.2349, -0.4272, 0.1006, 0.3636, 0.1894],
  32454. [-2.3590, -2.3096, 1.0946, -1.1491, -0.4777, -1.3035, 0.0823, 0.1859],
  32455. [ 0.5637, -0.4276, 0.9312, -1.0148, -0.4159, -1.2547, 0.2622, 0.2948],
  32456. [ 0.5444, -0.4137, 1.1649, -0.9891, -0.6327, -0.5801, 0.3558, 0.2762],
  32457. [ 0.5295, -0.4331, 1.6803, 0.1782, -0.3518, 0.1204, 0.6046, 0.0747],
  32458. [ 0.5971, -0.4446, 1.7483, -0.9914, -0.0197, -1.3498, 0.8464, 0.0536],
  32459. [ 0.6495, -0.3742, 1.5860, -1.0997, -0.2983, -1.3387, 0.5325, 0.0691],
  32460. [ 0.3837, -0.5090, 1.7551, -0.1794, -0.0704, -0.0786, 0.2862, 0.2237]],
  32461. device='cuda:0', grad_fn=<AddmmBackward>)
  32462. landmarks are: tensor([[[ 0.5726, -0.4159, 1.8192, -0.2459, -0.3402, 0.1159, 0.4797,
  32463. 0.3161],
  32464. [-2.2859, -2.2859, 0.9012, -1.4006, -0.4672, -1.1928, 0.1342,
  32465. 0.1373],
  32466. [ 0.5713, -0.3671, 0.8665, -1.0696, -0.3691, -1.2236, 0.3527,
  32467. 0.2622],
  32468. [ 0.5747, -0.3886, 1.1494, -1.0388, -0.6000, -0.5846, 0.5952,
  32469. 0.3546],
  32470. [ 0.6042, -0.4273, 1.6920, 0.1860, -0.2717, 0.1406, 0.7997,
  32471. 0.1004],
  32472. [ 0.6125, -0.4321, 1.8018, -0.9772, 0.0120, -1.2543, 0.9741,
  32473. 0.1821],
  32474. [ 0.6093, -0.4104, 1.5709, -1.1620, -0.1727, -1.2313, 0.6471,
  32475. 0.1621],
  32476. [ 0.5351, -0.4321, 1.7326, -0.1381, 0.0463, 0.1049, 0.4162,
  32477. 0.2776]]], device='cuda:0')
  32478. loss_train_step before backward: tensor(0.0080, device='cuda:0', grad_fn=<MseLossBackward>)
  32479. loss_train_step after backward: tensor(0.0080, device='cuda:0', grad_fn=<MseLossBackward>)
  32480. loss_train: 0.652790330350399
  32481. step: 45
  32482. running loss: 0.014506451785564423
  32483. Train Steps: 45/90 Loss: 0.0145 torch.Size([8, 600, 800])
  32484. torch.Size([8, 8])
  32485. tensor([[0.6201, 0.4055, 0.7717, 0.1919, 0.4250, 0.2079, 0.5871, 0.5044],
  32486. [0.6230, 0.4152, 0.7588, 0.2283, 0.4012, 0.2883, 0.6200, 0.5767],
  32487. [0.6205, 0.4062, 0.8337, 0.2683, 0.3675, 0.4283, 0.6338, 0.5250],
  32488. [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500],
  32489. [0.6299, 0.4008, 0.8450, 0.5350, 0.4213, 0.5000, 0.6350, 0.5100],
  32490. [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600],
  32491. [0.6284, 0.4093, 0.8900, 0.4700, 0.3650, 0.3850, 0.6212, 0.5167],
  32492. [0.6262, 0.4163, 0.8850, 0.5183, 0.3763, 0.4150, 0.6025, 0.5500]],
  32493. device='cuda:0', dtype=torch.float64)
  32494. predictions are: tensor([[ 0.4966, -0.4927, 1.2699, -1.3492, -0.2894, -1.4053, 0.3474, 0.0317],
  32495. [ 0.7391, -0.3165, 1.2395, -1.1042, -0.4034, -1.0465, 0.4538, 0.3428],
  32496. [ 0.3934, -0.5337, 1.5320, -1.0319, -0.4831, -0.3558, 0.4587, 0.2339],
  32497. [ 0.6288, -0.3821, 1.3381, -1.0664, -0.5571, -0.4490, 0.3945, 0.2473],
  32498. [ 0.5720, -0.4417, 1.6718, 0.1874, -0.2753, -0.1033, 0.4886, 0.1259],
  32499. [ 0.5163, -0.4307, 1.8413, 0.3058, -0.3082, -0.0951, 0.2531, 0.2563],
  32500. [ 0.4386, -0.5311, 1.9441, -0.1007, -0.5351, -0.5640, 0.4021, 0.0828],
  32501. [ 0.6740, -0.3230, 1.7757, 0.1000, -0.4700, -0.4806, 0.4102, 0.2311]],
  32502. device='cuda:0', grad_fn=<AddmmBackward>)
  32503. landmarks are: tensor([[[ 0.5785, -0.4129, 1.2785, -1.3996, -0.3227, -1.3259, 0.4258,
  32504. 0.0438],
  32505. [ 0.5915, -0.3682, 1.2187, -1.2313, -0.4326, -0.9541, 0.5778,
  32506. 0.3777],
  32507. [ 0.5799, -0.4099, 1.5651, -1.0465, -0.5885, -0.3075, 0.6413,
  32508. 0.1390],
  32509. [ 0.5763, -0.4147, 1.3226, -1.0619, -0.6635, -0.4152, 0.5374,
  32510. 0.2545],
  32511. [ 0.6236, -0.4344, 1.6171, 0.1852, -0.3402, 0.0236, 0.6471,
  32512. 0.0697],
  32513. [ 0.5883, -0.3594, 1.7557, 0.2545, -0.4152, -0.0611, 0.3353,
  32514. 0.3007],
  32515. [ 0.6164, -0.3956, 1.8249, -0.1150, -0.6000, -0.5076, 0.5836,
  32516. 0.1005],
  32517. [ 0.6066, -0.3632, 1.8018, 0.1082, -0.5480, -0.3691, 0.4970,
  32518. 0.2545]]], device='cuda:0')
  32519. loss_train_step before backward: tensor(0.0074, device='cuda:0', grad_fn=<MseLossBackward>)
  32520. loss_train_step after backward: tensor(0.0074, device='cuda:0', grad_fn=<MseLossBackward>)
  32521. loss_train: 0.6602126411162317
  32522. step: 46
  32523. running loss: 0.01435244871991808
  32524.  
  32525. Train Steps: 46/90 Loss: 0.0144 torch.Size([8, 600, 800])
  32526. torch.Size([8, 8])
  32527. tensor([[0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933],
  32528. [0.6125, 0.4076, 0.8488, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505],
  32529. [0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507],
  32530. [0.6263, 0.4039, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804],
  32531. [0.6159, 0.4085, 0.6900, 0.2283, 0.4088, 0.1950, 0.5123, 0.5397],
  32532. [0.6218, 0.4137, 0.7263, 0.2233, 0.4075, 0.2650, 0.6212, 0.5783],
  32533. [0.6260, 0.4106, 0.8025, 0.2583, 0.4550, 0.1867, 0.6281, 0.4869],
  32534. [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550]],
  32535. device='cuda:0', dtype=torch.float64)
  32536. predictions are: tensor([[ 0.3742, -0.5066, 1.7244, 0.3687, -0.0828, -0.2045, 0.3595, 0.3976],
  32537. [ 0.5415, -0.4194, 1.6979, -0.4614, -0.7064, -0.5221, -0.0144, 0.2012],
  32538. [ 0.8284, -0.2728, 0.9806, -1.3618, -0.4925, -1.1502, 0.3888, 0.2431],
  32539. [ 0.6283, -0.4036, 2.0047, -0.3204, -0.6244, -0.0814, 0.6591, -0.0324],
  32540. [ 0.4902, -0.4794, 0.9705, -1.2138, -0.4393, -1.3999, 0.1626, 0.1567],
  32541. [ 0.3520, -0.5710, 1.1238, -1.1896, -0.5240, -0.9904, 0.5083, 0.3460],
  32542. [ 0.6551, -0.3947, 1.5409, -1.1691, -0.2441, -1.3841, 0.6094, -0.0491],
  32543. [ 0.4091, -0.5144, 1.8871, -0.2143, -0.0562, -0.0024, 0.4478, 0.2442]],
  32544. device='cuda:0', grad_fn=<AddmmBackward>)
  32545. landmarks are: tensor([[[ 0.5959, -0.3579, 1.6055, 0.3623, -0.0573, -0.2074, 0.3122,
  32546. 0.4547],
  32547. [ 0.5433, -0.4032, 1.6344, -0.4922, -0.5769, -0.5846, 0.0357,
  32548. 0.2567],
  32549. [ 0.5185, -0.4252, 0.9647, -1.2928, -0.4788, -1.2390, 0.2617,
  32550. 0.2576],
  32551. [ 0.6072, -0.4206, 1.8711, -0.2536, -0.6115, -0.1304, 0.6812,
  32552. -0.0670],
  32553. [ 0.5591, -0.3990, 0.9012, -1.2313, -0.3979, -1.3852, 0.0804,
  32554. 0.2071],
  32555. [ 0.5863, -0.3749, 1.0686, -1.2543, -0.4037, -1.0619, 0.5836,
  32556. 0.3854],
  32557. [ 0.6058, -0.3892, 1.4208, -1.0927, -0.1843, -1.4237, 0.6154,
  32558. -0.0370],
  32559. [ 0.5351, -0.4321, 1.7326, -0.1381, 0.0463, 0.1049, 0.4162,
  32560. 0.2776]]], device='cuda:0')
  32561. loss_train_step before backward: tensor(0.0087, device='cuda:0', grad_fn=<MseLossBackward>)
  32562. loss_train_step after backward: tensor(0.0087, device='cuda:0', grad_fn=<MseLossBackward>)
  32563. loss_train: 0.66893858788535
  32564. step: 47
  32565. running loss: 0.014232735912454254
  32566. Train Steps: 47/90 Loss: 0.0142 torch.Size([8, 600, 800])
  32567. torch.Size([8, 8])
  32568. tensor([[0.6346, 0.4165, 0.9138, 0.3983, 0.3875, 0.4317, 0.7469, 0.5471],
  32569. [0.6226, 0.4098, 0.8912, 0.4100, 0.4025, 0.2383, 0.5763, 0.5367],
  32570. [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290],
  32571. [0.6185, 0.4067, 0.8838, 0.4450, 0.4037, 0.4733, 0.5213, 0.5142],
  32572. [0.6361, 0.4076, 0.8862, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297],
  32573. [ nan, nan, 0.7512, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617],
  32574. [0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283],
  32575. [0.6188, 0.4099, 0.7400, 0.2433, 0.3962, 0.2750, 0.6162, 0.5467]],
  32576. device='cuda:0', dtype=torch.float64)
  32577. predictions are: tensor([[ 0.4051, -0.5136, 1.8833, -0.4045, -0.3325, -0.2155, 0.8739, 0.2710],
  32578. [ 0.6273, -0.3338, 1.7346, -0.3649, -0.3575, -1.0852, 0.2452, 0.2073],
  32579. [ 0.7074, -0.3350, 1.1613, -1.3132, -0.2587, -1.1944, 0.5886, 0.1082],
  32580. [ 0.5802, -0.3619, 1.6835, -0.1829, -0.4105, -0.0615, -0.0204, 0.1170],
  32581. [ 0.7498, -0.3204, 1.7780, 0.1821, -0.5136, -0.1024, 0.6707, 0.1807],
  32582. [-2.0482, -2.0954, 1.0732, -1.2049, -0.3920, -1.2881, 0.1151, 0.2374],
  32583. [ 0.6764, -0.3235, 1.8600, -0.5183, -0.5283, -0.4335, 0.4900, 0.1452],
  32584. [ 0.7032, -0.3257, 1.0644, -1.2792, -0.4724, -0.9075, 0.4405, 0.1871]],
  32585. device='cuda:0', grad_fn=<AddmmBackward>)
  32586. landmarks are: tensor([[[ 0.6454, -0.3623, 1.9346, -0.4460, -0.4961, -0.2921, 1.1642,
  32587. 0.2409],
  32588. [ 0.5900, -0.3932, 1.8307, -0.3921, -0.4268, -1.1851, 0.3758,
  32589. 0.1929],
  32590. [ 0.6125, -0.4273, 1.2807, -1.3253, -0.2574, -1.2542, 0.6864,
  32591. 0.1575],
  32592. [ 0.5708, -0.4075, 1.7961, -0.2305, -0.4210, -0.0996, 0.1219,
  32593. 0.0893],
  32594. [ 0.6520, -0.4032, 1.8076, 0.1852, -0.5711, -0.1381, 0.7876,
  32595. 0.1608],
  32596. [-2.2859, -2.2859, 1.1841, -1.3082, -0.3055, -1.3621, 0.3007,
  32597. 0.3084],
  32598. [ 0.6073, -0.4058, 1.8885, -0.4999, -0.5942, -0.4768, 0.6413,
  32599. 0.1544],
  32600. [ 0.5725, -0.3926, 1.1321, -1.1620, -0.4557, -1.0157, 0.5605,
  32601. 0.2391]]], device='cuda:0')
  32602. loss_train_step before backward: tensor(0.0102, device='cuda:0', grad_fn=<MseLossBackward>)
  32603. loss_train_step after backward: tensor(0.0102, device='cuda:0', grad_fn=<MseLossBackward>)
  32604. loss_train: 0.6791246416978538
  32605. step: 48
  32606. running loss: 0.014148430035371954
  32607. Train Steps: 48/90 Loss: 0.0141 torch.Size([8, 600, 800])
  32608. torch.Size([8, 8])
  32609. tensor([[0.6296, 0.4076, 0.8400, 0.5583, 0.3700, 0.4367, 0.6876, 0.5494],
  32610. [0.6200, 0.3978, 0.8900, 0.4550, 0.3775, 0.5200, 0.6150, 0.5367],
  32611. [0.6129, 0.4063, 0.8738, 0.5250, 0.4313, 0.4733, 0.5230, 0.5874],
  32612. [0.6229, 0.4198, 0.7662, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783],
  32613. [0.6120, 0.4014, 0.6863, 0.2817, 0.3700, 0.2783, 0.5513, 0.5667],
  32614. [0.6200, 0.3993, 0.8519, 0.4923, 0.3962, 0.4717, 0.6013, 0.5433],
  32615. [0.6302, 0.4007, 0.8850, 0.5500, 0.3613, 0.4517, 0.6335, 0.5012],
  32616. [0.6124, 0.4069, 0.8314, 0.5001, 0.3738, 0.4650, 0.5167, 0.5402]],
  32617. device='cuda:0', dtype=torch.float64)
  32618. predictions are: tensor([[ 0.6016, -0.4220, 1.7538, 0.1703, -0.5186, -0.2837, 0.8657, 0.1236],
  32619. [ 0.4566, -0.5206, 1.9691, -0.3438, -0.5333, 0.1213, 0.4083, 0.1729],
  32620. [ 0.6168, -0.3684, 1.7965, 0.0535, -0.2900, -0.0179, 0.1753, 0.3332],
  32621. [ 0.4819, -0.4374, 1.2815, -1.2242, -0.1292, -1.3188, 0.3799, 0.3378],
  32622. [ 0.5563, -0.3974, 1.0200, -1.2427, -0.5874, -0.9715, 0.2051, 0.2670],
  32623. [ 0.6416, -0.3941, 1.8055, -0.1221, -0.4340, -0.1021, 0.5248, 0.1909],
  32624. [ 0.6706, -0.4074, 1.8877, 0.1244, -0.5731, -0.2325, 0.6664, 0.0643],
  32625. [ 0.6742, -0.3256, 1.6654, -0.1106, -0.4677, -0.1163, 0.1835, 0.2330]],
  32626. device='cuda:0', grad_fn=<AddmmBackward>)
  32627. landmarks are: tensor([[[ 0.6224, -0.4032, 1.5940, 0.2930, -0.5769, -0.2690, 0.8900,
  32628. 0.2516],
  32629. [ 0.5776, -0.4484, 1.8249, -0.1843, -0.5423, 0.1159, 0.5547,
  32630. 0.1929],
  32631. [ 0.5453, -0.4091, 1.7499, 0.1390, -0.2940, -0.0996, 0.1300,
  32632. 0.4272],
  32633. [ 0.5912, -0.3467, 1.2533, -1.0388, -0.1150, -1.3005, 0.3353,
  32634. 0.3854],
  32635. [ 0.5410, -0.4321, 0.8838, -0.9849, -0.5769, -1.0003, 0.2603,
  32636. 0.3315],
  32637. [ 0.5777, -0.4415, 1.6491, -0.0122, -0.4557, -0.1073, 0.4912,
  32638. 0.2237],
  32639. [ 0.6249, -0.4352, 1.8018, 0.2545, -0.6173, -0.1997, 0.6401,
  32640. 0.0291],
  32641. [ 0.5425, -0.4067, 1.5543, 0.0241, -0.5596, -0.1381, 0.1005,
  32642. 0.2093]]], device='cuda:0')
  32643. loss_train_step before backward: tensor(0.0076, device='cuda:0', grad_fn=<MseLossBackward>)
  32644. loss_train_step after backward: tensor(0.0076, device='cuda:0', grad_fn=<MseLossBackward>)
  32645. loss_train: 0.6867347755469382
  32646. step: 49
  32647. running loss: 0.01401499541932527
  32648. Train Steps: 49/90 Loss: 0.0140 torch.Size([8, 600, 800])
  32649. torch.Size([8, 8])
  32650. tensor([[0.6048, 0.3928, 0.8538, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719],
  32651. [0.6200, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913],
  32652. [0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162],
  32653. [0.6333, 0.4037, 0.8638, 0.5733, 0.4012, 0.4717, 0.6369, 0.4938],
  32654. [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350],
  32655. [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5988, 0.5283],
  32656. [0.6190, 0.4135, 0.8000, 0.4883, 0.3566, 0.3647, 0.5613, 0.5900],
  32657. [ nan, nan, 0.7512, 0.2117, 0.4288, 0.2000, 0.5600, 0.5617]],
  32658. device='cuda:0', dtype=torch.float64)
  32659. predictions are: tensor([[ 0.5894, -0.3801, 1.6447, 0.1531, -0.4084, 0.0639, 0.3153, 0.1972],
  32660. [ 0.6713, -0.3791, 1.7616, 0.0157, -0.3227, 0.1243, 0.4278, 0.0049],
  32661. [ 0.6451, -0.3785, 1.8642, -0.5306, -0.5431, -0.1355, 0.8136, 0.2384],
  32662. [ 0.7246, -0.3287, 1.7098, 0.2364, -0.4094, -0.0973, 0.7035, 0.0385],
  32663. [ 0.7551, -0.2297, 1.7543, -0.4982, -0.5698, -0.4273, 0.2666, 0.2370],
  32664. [ 0.6142, -0.3619, 1.7747, 0.0742, -0.4392, -0.0329, 0.5170, 0.2477],
  32665. [ 0.6024, -0.3271, 1.5167, -0.2986, -0.5872, -0.5988, 0.2826, 0.4324],
  32666. [-2.1414, -2.1944, 1.1244, -1.3578, -0.3550, -1.3435, 0.2133, 0.2454]],
  32667. device='cuda:0', grad_fn=<AddmmBackward>)
  32668. landmarks are: tensor([[[ 0.5079, -0.4714, 1.6575, 0.2237, -0.4961, 0.0774, 0.1465,
  32669. -0.1061],
  32670. [ 0.5776, -0.4784, 1.7730, 0.1236, -0.4037, 0.0697, 0.5132,
  32671. -0.0168],
  32672. [ 0.6059, -0.4442, 1.8921, -0.3422, -0.6230, -0.1461, 0.7489,
  32673. 0.0985],
  32674. [ 0.6395, -0.4213, 1.7037, 0.3623, -0.4326, -0.1073, 0.6560,
  32675. -0.0049],
  32676. [ 0.6072, -0.3239, 1.8423, -0.3537, -0.4961, -0.3921, 0.2083,
  32677. 0.1852],
  32678. [ 0.5603, -0.4319, 1.7788, 0.1775, -0.5538, -0.0611, 0.4797,
  32679. 0.1544],
  32680. [ 0.5732, -0.3761, 1.4092, -0.0303, -0.6388, -0.6012, 0.3065,
  32681. 0.4393],
  32682. [-2.2859, -2.2859, 1.1841, -1.3082, -0.3055, -1.3621, 0.3007,
  32683. 0.3084]]], device='cuda:0')
  32684. loss_train_step before backward: tensor(0.0087, device='cuda:0', grad_fn=<MseLossBackward>)
  32685. loss_train_step after backward: tensor(0.0087, device='cuda:0', grad_fn=<MseLossBackward>)
  32686. loss_train: 0.6954759764485061
  32687. step: 50
  32688. running loss: 0.013909519528970122
  32689.  
  32690. Train Steps: 50/90 Loss: 0.0139 torch.Size([8, 600, 800])
  32691. torch.Size([8, 8])
  32692. tensor([[ nan, nan, 0.7225, 0.2167, 0.3987, 0.2283, 0.5427, 0.5181],
  32693. [0.6230, 0.4152, 0.7588, 0.2283, 0.4012, 0.2883, 0.6200, 0.5767],
  32694. [0.6179, 0.3993, 0.8925, 0.4789, 0.3879, 0.4900, 0.6041, 0.5279],
  32695. [0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274],
  32696. [0.6208, 0.4082, 0.8538, 0.3067, 0.3588, 0.3717, 0.6112, 0.5517],
  32697. [0.6055, 0.4015, 0.7425, 0.2033, 0.4113, 0.1883, 0.5217, 0.4823],
  32698. [0.6259, 0.4133, 0.8200, 0.2317, 0.5025, 0.1533, 0.6250, 0.4947],
  32699. [0.6128, 0.4115, 0.7934, 0.3778, 0.3450, 0.4033, 0.5337, 0.5456]],
  32700. device='cuda:0', dtype=torch.float64)
  32701. predictions are: tensor([[-2.0417, -2.1145, 0.9707, -1.2916, -0.5134, -1.1903, 0.1369, 0.1892],
  32702. [ 0.6440, -0.3374, 1.2099, -1.0695, -0.4862, -0.8650, 0.6079, 0.3799],
  32703. [ 0.6411, -0.3589, 1.8196, -0.0150, -0.5245, 0.0290, 0.5910, 0.1434],
  32704. [ 0.5620, -0.3825, 1.6482, -0.7810, -0.0810, -1.1873, 0.6928, 0.2009],
  32705. [ 0.5983, -0.3833, 1.6245, -0.7267, -0.6526, -0.4460, 0.5589, 0.2570],
  32706. [ 0.3943, -0.4981, 1.0297, -1.1695, -0.4064, -1.2838, 0.2265, 0.0506],
  32707. [ 0.5851, -0.3882, 1.4568, -1.1265, -0.0250, -1.3781, 0.6049, 0.1001],
  32708. [ 0.5941, -0.3497, 1.4244, -0.4968, -0.6755, -0.3567, 0.3419, 0.2971]],
  32709. device='cuda:0', grad_fn=<AddmmBackward>)
  32710. landmarks are: tensor([[[-2.2859e+00, -2.2859e+00, 1.0513e+00, -1.2851e+00, -4.4411e-01,
  32711. -1.2313e+00, 2.2057e-01, 1.0729e-01],
  32712. [ 5.9151e-01, -3.6821e-01, 1.2187e+00, -1.2313e+00, -4.3256e-01,
  32713. -9.5412e-01, 5.7783e-01, 3.7768e-01],
  32714. [ 5.6801e-01, -4.4175e-01, 1.8365e+00, -7.4042e-02, -4.9414e-01,
  32715. -2.2653e-02, 5.0451e-01, 1.5252e-01],
  32716. [ 6.5365e-01, -3.7194e-01, 1.6979e+00, -8.6174e-01, -1.6859e-02,
  32717. -1.3621e+00, 6.9257e-01, 1.5008e-01],
  32718. [ 5.8135e-01, -4.0031e-01, 1.6575e+00, -8.6944e-01, -6.2887e-01,
  32719. -5.6921e-01, 5.3741e-01, 2.6220e-01],
  32720. [ 5.1085e-01, -4.3164e-01, 1.1436e+00, -1.3467e+00, -3.8637e-01,
  32721. -1.4160e+00, 1.2393e-01, -5.8033e-02],
  32722. [ 6.0514e-01, -3.7714e-01, 1.5016e+00, -1.2159e+00, 3.5104e-02,
  32723. -1.5777e+00, 6.0099e-01, -9.2270e-04],
  32724. [ 5.4440e-01, -3.8522e-01, 1.3786e+00, -5.4087e-01, -6.9238e-01,
  32725. -4.2294e-01, 1.7915e-01, 2.3412e-01]]], device='cuda:0')
  32726. loss_train_step before backward: tensor(0.0076, device='cuda:0', grad_fn=<MseLossBackward>)
  32727. loss_train_step after backward: tensor(0.0076, device='cuda:0', grad_fn=<MseLossBackward>)
  32728. loss_train: 0.7030893294140697
  32729. step: 51
  32730. running loss: 0.013786065282628816
  32731. Train Steps: 51/90 Loss: 0.0138 torch.Size([8, 600, 800])
  32732. torch.Size([8, 8])
  32733. tensor([[0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050],
  32734. [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351],
  32735. [0.6159, 0.4085, 0.6900, 0.2283, 0.4088, 0.1950, 0.5123, 0.5397],
  32736. [0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283],
  32737. [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378],
  32738. [0.6364, 0.4144, 0.8625, 0.3083, 0.4913, 0.2000, 0.6448, 0.5274],
  32739. [0.6196, 0.4088, 0.8888, 0.4583, 0.4500, 0.5683, 0.6138, 0.5883],
  32740. [0.6275, 0.4024, 0.7722, 0.2080, 0.4392, 0.2234, 0.6435, 0.5290]],
  32741. device='cuda:0', dtype=torch.float64)
  32742. predictions are: tensor([[ 0.6494, -0.3643, 1.7888, -0.1043, -0.6878, -0.2725, 0.3839, 0.0475],
  32743. [ 0.5098, -0.4389, 1.2404, -1.0108, -0.3349, -1.3609, 0.3059, 0.1627],
  32744. [ 0.3660, -0.5140, 0.8387, -1.1964, -0.4575, -1.3360, 0.0789, 0.2320],
  32745. [ 0.6436, -0.3578, 1.8885, -0.4588, -0.6509, -0.4069, 0.5721, 0.1984],
  32746. [ 0.5306, -0.4581, 1.7036, 0.0555, -0.3979, 0.1843, 0.7566, 0.2434],
  32747. [ 0.4884, -0.4214, 1.6386, -0.8014, -0.1471, -1.2295, 0.6100, 0.2176],
  32748. [ 0.5705, -0.3704, 1.7832, -0.1546, -0.3066, 0.3843, 0.5200, 0.3995],
  32749. [ 0.5586, -0.4240, 1.1823, -1.2662, -0.3523, -1.1681, 0.6554, 0.1749]],
  32750. device='cuda:0', grad_fn=<AddmmBackward>)
  32751. landmarks are: tensor([[[ 0.5844, -0.4466, 1.8423, -0.1997, -0.5942, -0.3998, 0.4219,
  32752. 0.0467],
  32753. [ 0.5664, -0.4321, 1.2862, -1.0003, -0.2189, -1.4608, 0.3883,
  32754. 0.1855],
  32755. [ 0.5591, -0.3990, 0.9012, -1.2313, -0.3979, -1.3852, 0.0804,
  32756. 0.2071],
  32757. [ 0.6073, -0.4058, 1.8885, -0.4999, -0.5942, -0.4768, 0.6413,
  32758. 0.1544],
  32759. [ 0.6141, -0.4345, 1.6864, -0.0303, -0.2882, 0.1544, 0.9521,
  32760. 0.1982],
  32761. [ 0.6537, -0.3719, 1.6979, -0.8617, -0.0169, -1.3621, 0.6926,
  32762. 0.1501],
  32763. [ 0.5759, -0.3978, 1.8192, -0.1689, -0.2074, 0.3392, 0.5490,
  32764. 0.4316],
  32765. [ 0.6125, -0.4273, 1.2807, -1.3253, -0.2574, -1.2542, 0.6864,
  32766. 0.1575]]], device='cuda:0')
  32767. loss_train_step before backward: tensor(0.0058, device='cuda:0', grad_fn=<MseLossBackward>)
  32768. loss_train_step after backward: tensor(0.0058, device='cuda:0', grad_fn=<MseLossBackward>)
  32769. loss_train: 0.7088540904223919
  32770. step: 52
  32771. running loss: 0.013631809431199845
  32772. Train Steps: 52/90 Loss: 0.0136 torch.Size([8, 600, 800])
  32773. torch.Size([8, 8])
  32774. tensor([[0.6197, 0.4091, 0.8800, 0.4783, 0.3538, 0.4767, 0.5950, 0.5550],
  32775. [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5413, 0.5717],
  32776. [0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748],
  32777. [0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266],
  32778. [0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869],
  32779. [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896],
  32780. [0.6124, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485],
  32781. [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750]],
  32782. device='cuda:0', dtype=torch.float64)
  32783. predictions are: tensor([[ 0.6867, -0.3530, 1.8893, -0.0427, -0.6081, -0.0703, 0.5564, 0.2224],
  32784. [ 0.6031, -0.4232, 1.8380, -0.0181, -0.3122, 0.0297, 0.3452, 0.1512],
  32785. [ 0.5968, -0.4435, 1.6900, 0.2695, -0.5246, -0.3441, 1.0880, 0.1359],
  32786. [ 0.5225, -0.4026, 1.1519, -1.2220, -0.3218, -1.0692, 0.5546, 0.5218],
  32787. [ 0.4358, -0.4869, 1.6258, -1.2949, -0.0756, -1.2650, 0.5801, -0.0843],
  32788. [ 0.5571, -0.4031, 1.5258, -0.5435, -0.6655, -0.8282, 0.1144, 0.2918],
  32789. [ 0.5077, -0.4292, 1.0523, -0.9469, -0.5805, -0.9220, 0.1200, 0.1878],
  32790. [ 0.4917, -0.4569, 1.0927, -1.1920, -0.5139, -0.9124, 0.5999, 0.2974]],
  32791. device='cuda:0', grad_fn=<AddmmBackward>)
  32792. landmarks are: tensor([[[ 0.5763, -0.3963, 1.7788, -0.0765, -0.6520, -0.0842, 0.4624,
  32793. 0.2776],
  32794. [ 0.5432, -0.4462, 1.7557, -0.0380, -0.3113, -0.0765, 0.2141,
  32795. 0.3546],
  32796. [ 0.6421, -0.3647, 1.5940, 0.3084, -0.5192, -0.3691, 1.1057,
  32797. 0.3692],
  32798. [ 0.5746, -0.4021, 1.0801, -1.1312, -0.3229, -1.1081, 0.4803,
  32799. 0.6084],
  32800. [ 0.6127, -0.3944, 1.5189, -1.2467, -0.1323, -1.4622, 0.5646,
  32801. -0.0369],
  32802. [ 0.5485, -0.3997, 1.4445, -0.4895, -0.6000, -0.8309, 0.1878,
  32803. 0.4374],
  32804. [ 0.5425, -0.3998, 0.9263, -0.8683, -0.6000, -1.0157, 0.0990,
  32805. 0.2476],
  32806. [ 0.5805, -0.3818, 1.0282, -1.1774, -0.4903, -0.9310, 0.5894,
  32807. 0.3700]]], device='cuda:0')
  32808. loss_train_step before backward: tensor(0.0067, device='cuda:0', grad_fn=<MseLossBackward>)
  32809. loss_train_step after backward: tensor(0.0067, device='cuda:0', grad_fn=<MseLossBackward>)
  32810. loss_train: 0.7155375052243471
  32811. step: 53
  32812. running loss: 0.013500707645742398
  32813. Train Steps: 53/90 Loss: 0.0135 torch.Size([8, 600, 800])
  32814. torch.Size([8, 8])
  32815. tensor([[0.6249, 0.4142, 0.8350, 0.3283, 0.3613, 0.3700, 0.6188, 0.5400],
  32816. [0.6216, 0.4167, 0.8588, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667],
  32817. [ nan, nan, 0.8363, 0.3317, 0.3563, 0.3367, 0.5329, 0.5142],
  32818. [0.6329, 0.4196, 0.9238, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748],
  32819. [0.6236, 0.4081, 0.8575, 0.3000, 0.3713, 0.3033, 0.5550, 0.5633],
  32820. [0.6149, 0.4054, 0.6713, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695],
  32821. [0.6163, 0.4114, 0.7650, 0.2017, 0.3763, 0.2867, 0.5631, 0.5071],
  32822. [ nan, nan, 0.7425, 0.2117, 0.3937, 0.2433, 0.5438, 0.5567]],
  32823. device='cuda:0', dtype=torch.float64)
  32824. predictions are: tensor([[ 0.6952, -0.2909, 1.5572, -0.7564, -0.6099, -0.5721, 0.5670, 0.2324],
  32825. [ 1.0167, -0.1225, 1.6744, 0.3887, -0.4468, 0.1252, 0.6714, 0.2086],
  32826. [-1.5770, -1.8032, 1.3967, -0.9103, -0.6525, -0.7907, 0.2326, 0.1734],
  32827. [ 0.7683, -0.2966, 1.9082, -0.1257, -0.2844, -0.8902, 1.1414, 0.2348],
  32828. [ 0.6591, -0.3276, 1.5648, -0.8208, -0.4872, -0.8877, 0.3777, 0.1544],
  32829. [ 0.7440, -0.2497, 0.8397, -1.2438, -0.3175, -1.3303, 0.1661, 0.2908],
  32830. [ 0.8056, -0.2475, 1.1913, -1.2971, -0.4608, -0.9584, 0.5213, 0.0536],
  32831. [-1.8084, -1.9542, 1.1200, -1.2545, -0.4184, -1.0982, 0.2681, 0.2478]],
  32832. device='cuda:0', grad_fn=<AddmmBackward>)
  32833. landmarks are: tensor([[[ 0.6006, -0.3728, 1.5709, -0.7694, -0.6173, -0.5769, 0.5721,
  32834. 0.2083],
  32835. [ 0.5853, -0.3614, 1.6806, 0.2930, -0.4499, 0.1005, 0.3815,
  32836. 0.3315],
  32837. [-2.2859, -2.2859, 1.5767, -0.7540, -0.6404, -0.7309, 0.1753,
  32838. 0.0893],
  32839. [ 0.6372, -0.3479, 1.9808, -0.1689, -0.3171, -0.9310, 1.1057,
  32840. 0.3692],
  32841. [ 0.5944, -0.4008, 1.6748, -0.9002, -0.5711, -0.8848, 0.2776,
  32842. 0.3161],
  32843. [ 0.5544, -0.4133, 0.8145, -1.2082, -0.4268, -1.3544, 0.1221,
  32844. 0.3446],
  32845. [ 0.5606, -0.3856, 1.2476, -1.3544, -0.5480, -0.9618, 0.3152,
  32846. 0.0562],
  32847. [-2.2859, -2.2859, 1.1436, -1.3082, -0.4672, -1.1620, 0.2256,
  32848. 0.2853]]], device='cuda:0')
  32849. loss_train_step before backward: tensor(0.0291, device='cuda:0', grad_fn=<MseLossBackward>)
  32850. loss_train_step after backward: tensor(0.0291, device='cuda:0', grad_fn=<MseLossBackward>)
  32851. loss_train: 0.7446684222668409
  32852. step: 54
  32853. running loss: 0.013790155967904462
  32854.  
  32855. Train Steps: 54/90 Loss: 0.0138 torch.Size([8, 600, 800])
  32856. torch.Size([8, 8])
  32857. tensor([[0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400],
  32858. [0.6176, 0.4030, 0.8850, 0.4850, 0.3688, 0.4050, 0.5312, 0.5783],
  32859. [0.6192, 0.3980, 0.7078, 0.2750, 0.4250, 0.2100, 0.5450, 0.5783],
  32860. [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973],
  32861. [0.6137, 0.4035, 0.8850, 0.4417, 0.3900, 0.4283, 0.5449, 0.5617],
  32862. [0.6043, 0.4022, 0.6887, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136],
  32863. [0.6081, 0.3950, 0.8538, 0.4667, 0.3850, 0.4917, 0.5342, 0.4954],
  32864. [0.6162, 0.4014, 0.8800, 0.5333, 0.3750, 0.4817, 0.5988, 0.5283]],
  32865. device='cuda:0', dtype=torch.float64)
  32866. predictions are: tensor([[ 7.3151e-01, -3.2304e-01, 1.8139e+00, -4.3675e-01, -6.3154e-01,
  32867. -1.5793e-01, 7.0981e-01, 2.7233e-01],
  32868. [ 7.2082e-01, -3.2855e-01, 1.7849e+00, 1.5959e-03, -5.5553e-01,
  32869. -3.2249e-01, 3.4367e-01, 2.8201e-01],
  32870. [-1.0985e+00, -1.4875e+00, 1.0323e+00, -1.1355e+00, -2.6686e-01,
  32871. -1.3477e+00, 3.6026e-01, 3.8651e-01],
  32872. [ 7.9492e-01, -3.2222e-01, 1.8341e+00, 2.6112e-01, -5.6936e-01,
  32873. -1.4523e-01, 8.4170e-01, 6.2565e-02],
  32874. [ 7.0293e-01, -3.1763e-01, 1.8348e+00, -1.5597e-01, -4.7494e-01,
  32875. -2.8655e-01, 4.0996e-01, 2.7960e-01],
  32876. [ 5.9462e-01, -3.7201e-01, 8.9894e-01, -1.3596e+00, -4.7997e-01,
  32877. -1.1826e+00, 4.1370e-01, 1.1764e-01],
  32878. [ 5.3425e-01, -4.5420e-01, 1.7221e+00, -1.7401e-01, -5.0341e-01,
  32879. -2.0689e-02, 3.2310e-01, 3.1694e-02],
  32880. [ 5.7662e-01, -4.1861e-01, 1.7637e+00, 2.6550e-01, -4.6343e-01,
  32881. -2.3559e-02, 5.5708e-01, 2.2800e-01]], device='cuda:0',
  32882. grad_fn=<AddmmBackward>)
  32883. landmarks are: tensor([[[ 0.5778, -0.4393, 1.8018, -0.4614, -0.6693, -0.1381, 0.5490,
  32884. 0.2083],
  32885. [ 0.5667, -0.4246, 1.8018, -0.0457, -0.5827, -0.4152, 0.1679,
  32886. 0.3854],
  32887. [ 0.5742, -0.4474, 0.9834, -1.0159, -0.3229, -1.3159, 0.2314,
  32888. 0.3854],
  32889. [ 0.6240, -0.4321, 1.8423, 0.1852, -0.5885, -0.1612, 0.6962,
  32890. 0.0111],
  32891. [ 0.5488, -0.4221, 1.8018, -0.2459, -0.4845, -0.3075, 0.2309,
  32892. 0.3087],
  32893. [ 0.5053, -0.4281, 0.8954, -1.3698, -0.5423, -1.1389, 0.2453,
  32894. 0.0862],
  32895. [ 0.5227, -0.4615, 1.6575, -0.1304, -0.5076, -0.0149, 0.1815,
  32896. 0.0021],
  32897. [ 0.5603, -0.4319, 1.7788, 0.1775, -0.5538, -0.0611, 0.4797,
  32898. 0.1544]]], device='cuda:0')
  32899. loss_train_step before backward: tensor(0.0676, device='cuda:0', grad_fn=<MseLossBackward>)
  32900. loss_train_step after backward: tensor(0.0676, device='cuda:0', grad_fn=<MseLossBackward>)
  32901. loss_train: 0.8122478108853102
  32902. step: 55
  32903. running loss: 0.01476814201609655
  32904. Train Steps: 55/90 Loss: 0.0148 torch.Size([8, 600, 800])
  32905. torch.Size([8, 8])
  32906. tensor([[0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320],
  32907. [0.6068, 0.3963, 0.8650, 0.4317, 0.4037, 0.5083, 0.5253, 0.4999],
  32908. [0.6197, 0.3986, 0.8800, 0.4617, 0.4188, 0.4783, 0.5687, 0.5550],
  32909. [0.6198, 0.4076, 0.7945, 0.2453, 0.4475, 0.1871, 0.5897, 0.5350],
  32910. [0.6205, 0.4016, 0.8350, 0.2717, 0.3987, 0.2550, 0.5787, 0.5133],
  32911. [0.6227, 0.4193, 0.8838, 0.4933, 0.3663, 0.3733, 0.5637, 0.5633],
  32912. [0.6224, 0.4179, 0.8700, 0.5683, 0.4037, 0.4683, 0.5650, 0.5600],
  32913. [0.6260, 0.4214, 0.8538, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983]],
  32914. device='cuda:0', dtype=torch.float64)
  32915. predictions are: tensor([[ 0.6733, -0.3537, 1.3374, -1.3534, -0.3475, -1.1263, 0.9403, 0.1226],
  32916. [ 0.4557, -0.5294, 1.6551, -0.3730, -0.5073, 0.0383, 0.2753, 0.0239],
  32917. [ 0.4782, -0.5015, 1.6879, -0.2401, -0.4075, -0.0528, 0.4632, 0.2047],
  32918. [ 0.5682, -0.3980, 1.3429, -1.2565, -0.1921, -1.4071, 0.6248, 0.1764],
  32919. [ 0.6195, -0.3795, 1.5597, -1.1509, -0.4353, -1.0833, 0.5308, 0.0891],
  32920. [ 0.5529, -0.4028, 1.6849, -0.0855, -0.6844, -0.5984, 0.4197, 0.4058],
  32921. [ 0.5639, -0.4079, 1.6461, 0.3015, -0.4696, -0.1384, 0.3921, 0.3661],
  32922. [ 0.4199, -0.5031, 1.5452, 0.2457, -0.6153, -0.5789, 0.3617, 0.4921]],
  32923. device='cuda:0', grad_fn=<AddmmBackward>)
  32924. landmarks are: tensor([[[ 0.6141, -0.4153, 1.4208, -1.2697, -0.2940, -1.0234, 0.8644,
  32925. 0.1715],
  32926. [ 0.5168, -0.4556, 1.7095, -0.2921, -0.4210, 0.0620, 0.1404,
  32927. 0.0231],
  32928. [ 0.5763, -0.4450, 1.7788, -0.1535, -0.3517, -0.0765, 0.3411,
  32929. 0.2776],
  32930. [ 0.5768, -0.4031, 1.3838, -1.1527, -0.2188, -1.4216, 0.4379,
  32931. 0.1850],
  32932. [ 0.5800, -0.4312, 1.5709, -1.0311, -0.4441, -1.1081, 0.3873,
  32933. 0.0851],
  32934. [ 0.5902, -0.3493, 1.7961, -0.0072, -0.5942, -0.5615, 0.3180,
  32935. 0.3161],
  32936. [ 0.5891, -0.3557, 1.7326, 0.3392, -0.4210, -0.1227, 0.3238,
  32937. 0.3007],
  32938. [ 0.6055, -0.3393, 1.6575, 0.2545, -0.5942, -0.5461, 0.2949,
  32939. 0.4778]]], device='cuda:0')
  32940. loss_train_step before backward: tensor(0.0060, device='cuda:0', grad_fn=<MseLossBackward>)
  32941. loss_train_step after backward: tensor(0.0060, device='cuda:0', grad_fn=<MseLossBackward>)
  32942. loss_train: 0.8182361843064427
  32943. step: 56
  32944. running loss: 0.01461136043404362
  32945. Train Steps: 56/90 Loss: 0.0146 torch.Size([8, 600, 800])
  32946. torch.Size([8, 8])
  32947. tensor([[0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400],
  32948. [0.6059, 0.4002, 0.7562, 0.2767, 0.3538, 0.3033, 0.5529, 0.5455],
  32949. [ nan, nan, 0.8888, 0.3100, 0.5262, 0.2817, 0.7145, 0.6003],
  32950. [0.6246, 0.4126, 0.8850, 0.4833, 0.4200, 0.5350, 0.6112, 0.5250],
  32951. [0.6213, 0.4131, 0.8438, 0.3550, 0.3513, 0.4400, 0.5716, 0.5123],
  32952. [0.6250, 0.4236, 0.8638, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799],
  32953. [0.6095, 0.3970, 0.8688, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940],
  32954. [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817]],
  32955. device='cuda:0', dtype=torch.float64)
  32956. predictions are: tensor([[ 0.5556, -0.4120, 1.6844, -0.5043, -0.7045, -0.3261, 0.6481, 0.2448],
  32957. [ 0.5690, -0.3721, 1.1256, -1.0251, -0.6030, -1.0134, 0.4253, 0.1987],
  32958. [-2.3533, -2.3378, 1.6712, -0.8950, 0.0409, -1.2024, 1.0081, 0.3700],
  32959. [ 0.5131, -0.4205, 1.6990, -0.0872, -0.4344, 0.0332, 0.6334, 0.1912],
  32960. [ 0.4841, -0.4213, 1.5142, -0.6901, -0.7282, -0.4796, 0.3208, 0.1235],
  32961. [ 0.6184, -0.3216, 1.5878, -0.5615, -0.4717, -0.9574, 0.4515, 0.4288],
  32962. [ 0.4499, -0.4935, 1.6163, -0.1788, -0.1629, -0.1551, 0.3060, 0.1087],
  32963. [ 0.5053, -0.4721, 1.4499, 0.3496, -0.3156, -0.0978, 0.2096, 0.0571]],
  32964. device='cuda:0', grad_fn=<AddmmBackward>)
  32965. landmarks are: tensor([[[ 0.5778, -0.4393, 1.8018, -0.4614, -0.6693, -0.1381, 0.5490,
  32966. 0.2083],
  32967. [ 0.5129, -0.4374, 1.2072, -1.0080, -0.6520, -0.8848, 0.2679,
  32968. 0.2335],
  32969. [-2.2859, -2.2859, 1.8192, -0.8540, 0.1448, -0.9849, 1.0143,
  32970. 0.4867],
  32971. [ 0.5991, -0.3803, 1.8018, -0.0534, -0.3460, 0.1852, 0.5374,
  32972. 0.1390],
  32973. [ 0.5841, -0.3778, 1.6113, -0.6462, -0.6635, -0.2536, 0.3542,
  32974. 0.0802],
  32975. [ 0.6008, -0.3293, 1.7037, -0.5461, -0.4152, -0.8309, 0.3234,
  32976. 0.3928],
  32977. [ 0.5292, -0.4520, 1.7268, -0.0842, -0.0413, -0.0324, 0.1116,
  32978. -0.0039],
  32979. [ 0.5193, -0.4699, 1.5767, 0.4008, -0.2420, 0.0774, 0.1178,
  32980. -0.0610]]], device='cuda:0')
  32981. loss_train_step before backward: tensor(0.0095, device='cuda:0', grad_fn=<MseLossBackward>)
  32982. loss_train_step after backward: tensor(0.0095, device='cuda:0', grad_fn=<MseLossBackward>)
  32983. loss_train: 0.8277576519176364
  32984. step: 57
  32985. running loss: 0.014522064068730463
  32986. Train Steps: 57/90 Loss: 0.0145 torch.Size([8, 600, 800])
  32987. torch.Size([8, 8])
  32988. tensor([[0.6132, 0.4118, 0.8200, 0.3633, 0.3563, 0.5400, 0.5787, 0.5136],
  32989. [0.6199, 0.4071, 0.7600, 0.2117, 0.4037, 0.2767, 0.6138, 0.5550],
  32990. [0.6200, 0.3998, 0.8850, 0.3950, 0.3500, 0.4650, 0.6138, 0.5400],
  32991. [0.6307, 0.4029, 0.8650, 0.5200, 0.3763, 0.4017, 0.7311, 0.5366],
  32992. [0.6332, 0.4118, 0.9238, 0.4267, 0.4012, 0.4733, 0.7525, 0.5436],
  32993. [0.6204, 0.4007, 0.7838, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167],
  32994. [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6063, 0.5617],
  32995. [0.6206, 0.4123, 0.7175, 0.2400, 0.3887, 0.2933, 0.6225, 0.5750]],
  32996. device='cuda:0', dtype=torch.float64)
  32997. predictions are: tensor([[ 0.3653, -0.5301, 1.4886, -0.5390, -0.5928, 0.0507, 0.2721, 0.2364],
  32998. [ 0.5125, -0.4335, 1.2703, -1.2746, -0.3055, -1.0842, 0.5741, 0.3149],
  32999. [ 0.4835, -0.5004, 1.7939, -0.4004, -0.6551, -0.2711, 0.3656, 0.2429],
  33000. [ 0.5630, -0.4785, 1.7487, 0.2648, -0.5154, -0.5629, 0.8131, 0.0676],
  33001. [ 0.5086, -0.5020, 1.8854, -0.2616, -0.4566, -0.1313, 0.9231, 0.2956],
  33002. [ 0.6456, -0.3688, 1.2962, -1.1110, -0.1280, -1.6545, 0.3240, 0.0716],
  33003. [ 0.6136, -0.3784, 1.8307, 0.0766, -0.5950, -0.2757, 0.2350, 0.2931],
  33004. [ 0.6073, -0.3740, 1.0229, -1.0454, -0.4761, -1.0501, 0.4470, 0.4177]],
  33005. device='cuda:0', grad_fn=<AddmmBackward>)
  33006. landmarks are: tensor([[[ 0.5466, -0.3840, 1.5016, -0.6077, -0.6404, 0.2083, 0.3871,
  33007. 0.0862],
  33008. [ 0.5775, -0.4054, 1.2245, -1.3082, -0.4210, -1.0080, 0.5490,
  33009. 0.2776],
  33010. [ 0.5778, -0.4393, 1.8018, -0.4614, -0.6693, -0.1381, 0.5490,
  33011. 0.2083],
  33012. [ 0.6273, -0.4249, 1.7095, 0.1159, -0.5480, -0.4306, 1.0910,
  33013. 0.1928],
  33014. [ 0.6388, -0.3840, 1.9808, -0.3152, -0.4326, -0.0996, 1.1897,
  33015. 0.2249],
  33016. [ 0.5796, -0.4351, 1.3342, -1.3159, -0.2189, -1.4853, 0.4046,
  33017. 0.1005],
  33018. [ 0.5797, -0.3965, 1.8480, -0.0765, -0.6173, -0.1535, 0.5143,
  33019. 0.3084],
  33020. [ 0.5805, -0.3818, 1.0282, -1.1774, -0.4903, -0.9310, 0.5894,
  33021. 0.3700]]], device='cuda:0')
  33022. loss_train_step before backward: tensor(0.0118, device='cuda:0', grad_fn=<MseLossBackward>)
  33023. loss_train_step after backward: tensor(0.0118, device='cuda:0', grad_fn=<MseLossBackward>)
  33024. loss_train: 0.8395349802449346
  33025. step: 58
  33026. running loss: 0.014474741038705769
  33027.  
  33028. Train Steps: 58/90 Loss: 0.0145 torch.Size([8, 600, 800])
  33029. torch.Size([8, 8])
  33030. tensor([[0.6198, 0.4101, 0.8838, 0.5283, 0.3763, 0.5267, 0.5913, 0.5567],
  33031. [0.6133, 0.4066, 0.6787, 0.2617, 0.3800, 0.2433, 0.5147, 0.5358],
  33032. [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5413, 0.5433],
  33033. [0.6185, 0.4129, 0.8900, 0.4567, 0.3937, 0.5417, 0.5734, 0.5110],
  33034. [0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758],
  33035. [0.6213, 0.4001, 0.7712, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183],
  33036. [0.6151, 0.4125, 0.8738, 0.4417, 0.3575, 0.3783, 0.5138, 0.5483],
  33037. [0.6160, 0.4093, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808]],
  33038. device='cuda:0', dtype=torch.float64)
  33039. predictions are: tensor([[ 0.4477, -0.5367, 1.7819, 0.2281, -0.5142, 0.0859, 0.5084, 0.3027],
  33040. [ 0.6758, -0.3454, 0.9215, -1.0019, -0.4842, -1.2423, 0.1739, 0.2234],
  33041. [ 0.5677, -0.3958, 1.6321, -0.6215, -0.5877, -0.8491, 0.4111, 0.2329],
  33042. [ 0.5721, -0.4261, 1.8749, -0.1511, -0.4139, 0.1686, 0.4451, 0.1869],
  33043. [ 0.5606, -0.4834, 1.8408, -0.0444, -0.3999, 0.0658, 0.7831, 0.1033],
  33044. [ 0.5758, -0.4490, 1.2640, -1.3102, -0.1981, -1.5140, 0.5301, 0.1225],
  33045. [ 0.4960, -0.4478, 1.6744, -0.1980, -0.6295, -0.6053, 0.1823, 0.3388],
  33046. [ 0.3174, -0.5879, 1.6093, -0.2038, -0.5482, -0.2282, 0.3039, 0.4434]],
  33047. device='cuda:0', grad_fn=<AddmmBackward>)
  33048. landmarks are: tensor([[[ 0.5770, -0.3918, 1.7961, 0.1544, -0.5480, 0.1467, 0.4450,
  33049. 0.2853],
  33050. [ 0.5470, -0.4081, 0.8492, -1.0773, -0.5307, -1.1620, 0.0912,
  33051. 0.1890],
  33052. [ 0.5894, -0.3479, 1.7730, -0.6847, -0.5538, -0.7155, 0.2141,
  33053. 0.2237],
  33054. [ 0.5711, -0.3788, 1.8249, -0.1766, -0.4672, 0.2160, 0.3625,
  33055. 0.0742],
  33056. [ 0.6072, -0.4250, 1.8249, -0.0072, -0.4037, 0.1082, 0.6730,
  33057. -0.0881],
  33058. [ 0.5838, -0.4379, 1.2764, -1.3082, -0.2824, -1.4545, 0.4162,
  33059. 0.1082],
  33060. [ 0.5551, -0.3807, 1.7499, -0.2459, -0.6346, -0.5384, 0.0871,
  33061. 0.2468],
  33062. [ 0.5592, -0.3956, 1.5543, -0.2456, -0.5885, -0.1689, 0.1392,
  33063. 0.3968]]], device='cuda:0')
  33064. loss_train_step before backward: tensor(0.0073, device='cuda:0', grad_fn=<MseLossBackward>)
  33065. loss_train_step after backward: tensor(0.0073, device='cuda:0', grad_fn=<MseLossBackward>)
  33066. loss_train: 0.8468739879317582
  33067. step: 59
  33068. running loss: 0.01435379640562302
  33069. Train Steps: 59/90 Loss: 0.0144 torch.Size([8, 600, 800])
  33070. torch.Size([8, 8])
  33071. tensor([[0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672],
  33072. [0.6153, 0.4119, 0.8463, 0.3833, 0.3600, 0.3200, 0.5106, 0.5563],
  33073. [0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058],
  33074. [0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750],
  33075. [0.6264, 0.4035, 0.8888, 0.4883, 0.4050, 0.5217, 0.6361, 0.4791],
  33076. [0.6138, 0.4101, 0.8800, 0.5083, 0.4637, 0.5950, 0.5587, 0.5077],
  33077. [0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6262, 0.5167],
  33078. [0.6148, 0.3996, 0.8488, 0.3867, 0.3488, 0.4067, 0.5863, 0.5000]],
  33079. device='cuda:0', dtype=torch.float64)
  33080. predictions are: tensor([[ 0.5716, -0.4683, 1.6745, 0.2683, -0.6068, -0.3342, 0.6065, 0.1156],
  33081. [ 0.6136, -0.3583, 1.5699, -0.5214, -0.5876, -0.8559, 0.1447, 0.3616],
  33082. [ 0.3707, -0.5752, 1.5958, -0.3524, -0.5343, -0.1191, 0.1477, 0.1774],
  33083. [ 0.6514, -0.3415, 1.3220, -0.9513, -0.3702, -1.1556, 0.1929, 0.4177],
  33084. [ 0.5917, -0.4493, 1.7983, -0.0735, -0.4208, 0.0744, 0.7328, 0.0889],
  33085. [ 0.4227, -0.5170, 1.7463, 0.0612, -0.0829, 0.3032, 0.4088, 0.2015],
  33086. [ 0.4966, -0.4800, 1.8956, -0.3544, -0.4724, -0.7064, 0.7027, 0.3024],
  33087. [ 0.4521, -0.4958, 1.6725, -0.5166, -0.7062, -0.4615, 0.4067, 0.1989]],
  33088. device='cuda:0', grad_fn=<AddmmBackward>)
  33089. landmarks are: tensor([[[ 0.6128, -0.4375, 1.7788, 0.2699, -0.6346, -0.2536, 0.5463,
  33090. -0.1278],
  33091. [ 0.5561, -0.3834, 1.6229, -0.5153, -0.6231, -0.8079, 0.0727,
  33092. 0.2837],
  33093. [ 0.5245, -0.4361, 1.5940, -0.2921, -0.5480, -0.0919, 0.2432,
  33094. 0.0502],
  33095. [ 0.5695, -0.3927, 1.3226, -0.9002, -0.4672, -1.1928, 0.1737,
  33096. 0.3700],
  33097. [ 0.6074, -0.4223, 1.8192, -0.0303, -0.4152, 0.1236, 0.6524,
  33098. -0.0731],
  33099. [ 0.5491, -0.3918, 1.7788, 0.0620, -0.1439, 0.4624, 0.2946,
  33100. 0.0592],
  33101. [ 0.6094, -0.3947, 1.8885, -0.2998, -0.5769, -0.6770, 0.6067,
  33102. 0.1005],
  33103. [ 0.5538, -0.4400, 1.6344, -0.4999, -0.6751, -0.4075, 0.4219,
  33104. 0.0236]]], device='cuda:0')
  33105. loss_train_step before backward: tensor(0.0074, device='cuda:0', grad_fn=<MseLossBackward>)
  33106. loss_train_step after backward: tensor(0.0074, device='cuda:0', grad_fn=<MseLossBackward>)
  33107. loss_train: 0.8542463732883334
  33108. step: 60
  33109. running loss: 0.014237439554805557
  33110. Train Steps: 60/90 Loss: 0.0142 torch.Size([8, 600, 800])
  33111. torch.Size([8, 8])
  33112. tensor([[0.6193, 0.4108, 0.7438, 0.2700, 0.3650, 0.3683, 0.6238, 0.5717],
  33113. [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426],
  33114. [0.6164, 0.4076, 0.8838, 0.4117, 0.3713, 0.5550, 0.6238, 0.5350],
  33115. [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938],
  33116. [0.6350, 0.4043, 0.8738, 0.5650, 0.3850, 0.4750, 0.6401, 0.4950],
  33117. [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413],
  33118. [0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742],
  33119. [0.6267, 0.4094, 0.8712, 0.3083, 0.4400, 0.2267, 0.6250, 0.5200]],
  33120. device='cuda:0', dtype=torch.float64)
  33121. predictions are: tensor([[ 0.4369, -0.4964, 1.1146, -1.0326, -0.6546, -0.4766, 0.3411, 0.4066],
  33122. [ 0.6392, -0.3629, 1.1748, -1.1745, -0.2520, -1.3513, 0.2541, 0.2668],
  33123. [ 0.3895, -0.5462, 1.7361, -0.3636, -0.5520, 0.3711, 0.4176, 0.3004],
  33124. [ 0.4097, -0.5550, 1.7841, 0.0594, -0.5848, -0.4577, 0.3989, 0.0598],
  33125. [ 0.4349, -0.5610, 1.6784, 0.2591, -0.5321, 0.0387, 0.3806, 0.0563],
  33126. [ 0.5784, -0.4637, 1.9492, -0.3301, -0.2924, -0.8105, 0.9900, 0.2359],
  33127. [ 0.6130, -0.4002, 1.7304, -0.6834, -0.4741, -1.0336, 0.3487, 0.0620],
  33128. [ 0.6742, -0.3106, 1.7169, -0.9077, -0.2364, -1.1439, 0.4729, 0.2411]],
  33129. device='cuda:0', grad_fn=<AddmmBackward>)
  33130. landmarks are: tensor([[[ 5.7471e-01, -3.8861e-01, 1.1494e+00, -1.0388e+00, -6.0000e-01,
  33131. -5.8460e-01, 5.9515e-01, 3.5458e-01],
  33132. [ 5.7841e-01, -4.1532e-01, 1.2972e+00, -1.2541e+00, -2.2647e-01,
  33133. -1.4553e+00, 4.7413e-01, 2.2033e-01],
  33134. [ 5.6143e-01, -4.0323e-01, 1.7961e+00, -3.8445e-01, -5.7113e-01,
  33135. 2.7760e-01, 5.9515e-01, 1.8522e-01],
  33136. [ 6.2730e-01, -4.1045e-01, 1.8480e+00, 1.0824e-01, -5.5381e-01,
  33137. -5.0762e-01, 6.4140e-01, -4.8817e-03],
  33138. [ 6.4706e-01, -4.1832e-01, 1.7499e+00, 3.2379e-01, -5.0762e-01,
  33139. -9.1917e-02, 6.7064e-01, 4.6189e-04],
  33140. [ 6.2236e-01, -4.3453e-01, 1.9404e+00, -2.9207e-01, -3.1709e-01,
  33141. -8.7714e-01, 1.0655e+00, 2.1421e-01],
  33142. [ 6.1339e-01, -4.2179e-01, 1.7268e+00, -6.1540e-01, -4.7298e-01,
  33143. -1.0850e+00, 5.4635e-01, -9.5723e-02],
  33144. [ 6.0889e-01, -3.9477e-01, 1.7383e+00, -8.6174e-01, -2.5358e-01,
  33145. -1.2390e+00, 6.0092e-01, 1.1594e-01]]], device='cuda:0')
  33146. loss_train_step before backward: tensor(0.0121, device='cuda:0', grad_fn=<MseLossBackward>)
  33147. loss_train_step after backward: tensor(0.0121, device='cuda:0', grad_fn=<MseLossBackward>)
  33148. loss_train: 0.8663500072434545
  33149. step: 61
  33150. running loss: 0.014202459135138597
  33151. Train Steps: 61/90 Loss: 0.0142 torch.Size([8, 600, 800])
  33152. torch.Size([8, 8])
  33153. tensor([[0.6058, 0.3986, 0.8324, 0.4626, 0.3838, 0.4983, 0.5147, 0.5466],
  33154. [ nan, nan, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733],
  33155. [0.6102, 0.4020, 0.8638, 0.3717, 0.3625, 0.5017, 0.6038, 0.5500],
  33156. [0.6353, 0.4128, 0.9138, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991],
  33157. [0.6284, 0.4127, 0.8538, 0.5867, 0.4363, 0.5083, 0.6038, 0.5433],
  33158. [0.6179, 0.3993, 0.8925, 0.4789, 0.3879, 0.4900, 0.6041, 0.5279],
  33159. [0.6275, 0.4024, 0.8600, 0.2283, 0.5350, 0.1800, 0.7074, 0.5413],
  33160. [0.6129, 0.4063, 0.8738, 0.5250, 0.4313, 0.4733, 0.5230, 0.5874]],
  33161. device='cuda:0', dtype=torch.float64)
  33162. predictions are: tensor([[ 0.3651, -0.5405, 1.6259, -0.0620, -0.5820, 0.0492, 0.1450, 0.1149],
  33163. [-1.0391, -1.4939, 0.8902, -1.3470, -0.2780, -1.4584, 0.2304, 0.4114],
  33164. [ 0.5111, -0.4410, 1.7324, -0.5073, -0.7363, 0.0249, 0.4958, 0.1436],
  33165. [ 0.6049, -0.4107, 1.9642, -0.5052, -0.2794, -0.8110, 1.0207, 0.2991],
  33166. [ 0.5173, -0.4620, 1.6767, 0.4512, -0.4031, 0.0677, 0.3231, 0.0606],
  33167. [ 0.4464, -0.4915, 1.8005, -0.0707, -0.6219, -0.0269, 0.3503, 0.0424],
  33168. [ 0.9084, -0.1551, 1.7199, -1.1509, 0.0133, -1.4654, 0.9196, 0.0301],
  33169. [ 0.5109, -0.4292, 1.6963, 0.2100, -0.4320, -0.0112, -0.0428, 0.2638]],
  33170. device='cuda:0', grad_fn=<AddmmBackward>)
  33171. landmarks are: tensor([[[ 0.5124, -0.4446, 1.5587, -0.1493, -0.5134, 0.0159, 0.0912,
  33172. 0.2386],
  33173. [-2.2859, -2.2859, 0.7106, -1.4468, -0.2882, -1.4237, 0.2430,
  33174. 0.3623],
  33175. [ 0.5324, -0.4293, 1.7037, -0.5692, -0.6115, 0.0313, 0.5028,
  33176. 0.2545],
  33177. [ 0.6487, -0.3792, 1.9346, -0.6539, -0.1208, -0.7848, 1.0143,
  33178. 0.4814],
  33179. [ 0.6166, -0.3795, 1.6575, 0.4239, -0.2709, 0.0620, 0.5028,
  33180. 0.2237],
  33181. [ 0.5680, -0.4417, 1.8365, -0.0740, -0.4941, -0.0227, 0.5045,
  33182. 0.1525],
  33183. [ 0.6125, -0.4273, 1.6864, -1.2313, 0.1852, -1.4545, 0.9814,
  33184. 0.2142],
  33185. [ 0.5453, -0.4091, 1.7499, 0.1390, -0.2940, -0.0996, 0.1300,
  33186. 0.4272]]], device='cuda:0')
  33187. loss_train_step before backward: tensor(0.0454, device='cuda:0', grad_fn=<MseLossBackward>)
  33188. loss_train_step after backward: tensor(0.0454, device='cuda:0', grad_fn=<MseLossBackward>)
  33189. loss_train: 0.9117817068472505
  33190. step: 62
  33191. running loss: 0.014706156562052427
  33192.  
  33193. Train Steps: 62/90 Loss: 0.0147 torch.Size([8, 600, 800])
  33194. torch.Size([8, 8])
  33195. tensor([[0.6183, 0.4076, 0.8838, 0.4517, 0.3813, 0.4483, 0.5775, 0.5633],
  33196. [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356],
  33197. [0.6122, 0.3993, 0.8738, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927],
  33198. [0.6249, 0.4142, 0.8350, 0.3283, 0.3613, 0.3700, 0.6188, 0.5400],
  33199. [ nan, nan, 0.8463, 0.2550, 0.5850, 0.2133, 0.7129, 0.6072],
  33200. [0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862],
  33201. [0.6134, 0.4090, 0.6926, 0.2819, 0.3538, 0.3233, 0.5563, 0.5667],
  33202. [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167]],
  33203. device='cuda:0', dtype=torch.float64)
  33204. predictions are: tensor([[ 0.3876, -0.5186, 1.7843, -0.1677, -0.5246, -0.2142, 0.2486, 0.2402],
  33205. [ 0.5521, -0.4131, 1.4844, -0.9581, -0.3242, -1.1722, 0.6436, 0.1172],
  33206. [ 0.4796, -0.4917, 1.7374, -0.0860, -0.2325, 0.0316, 0.0268, -0.1138],
  33207. [ 0.5655, -0.4131, 1.6269, -0.6418, -0.6632, -0.6151, 0.5047, 0.1447],
  33208. [-2.0073, -2.1644, 1.5982, -0.9444, 0.2918, -1.1021, 1.0624, 0.3503],
  33209. [ 0.5611, -0.4074, 1.8383, -0.2921, -0.6295, -0.2790, 0.4994, -0.0903],
  33210. [ 0.4352, -0.4581, 1.0209, -0.8541, -0.6522, -0.8294, 0.1052, 0.2289],
  33211. [ 0.6101, -0.3655, 1.4231, -0.7075, -0.6191, -0.5375, 0.3774, 0.4063]],
  33212. device='cuda:0', grad_fn=<AddmmBackward>)
  33213. landmarks are: tensor([[[ 0.5700, -0.4032, 1.7961, -0.1997, -0.5249, -0.2151, 0.3815,
  33214. 0.3161],
  33215. [ 0.6174, -0.4225, 1.4975, -1.1709, -0.3174, -1.1806, 0.6539,
  33216. 0.1879],
  33217. [ 0.5417, -0.4417, 1.7499, -0.1304, -0.1994, -0.0324, 0.0951,
  33218. -0.0099],
  33219. [ 0.6006, -0.3728, 1.5709, -0.7694, -0.6173, -0.5769, 0.5721,
  33220. 0.2083],
  33221. [-2.2859, -2.2859, 1.6229, -1.1081, 0.4162, -1.3005, 1.0070,
  33222. 0.5188],
  33223. [ 0.6072, -0.4075, 1.8942, -0.3537, -0.5423, -0.1612, 0.6277,
  33224. -0.0400],
  33225. [ 0.5473, -0.3966, 0.9131, -0.9838, -0.6520, -0.7925, 0.2834,
  33226. 0.3315],
  33227. [ 0.5768, -0.3899, 1.3861, -0.7771, -0.5885, -0.5461, 0.5028,
  33228. 0.5624]]], device='cuda:0')
  33229. loss_train_step before backward: tensor(0.0091, device='cuda:0', grad_fn=<MseLossBackward>)
  33230. loss_train_step after backward: tensor(0.0091, device='cuda:0', grad_fn=<MseLossBackward>)
  33231. loss_train: 0.9208882143720984
  33232. step: 63
  33233. running loss: 0.014617273244001562
  33234. Train Steps: 63/90 Loss: 0.0146 torch.Size([8, 600, 800])
  33235. torch.Size([8, 8])
  33236. tensor([[0.6060, 0.3924, 0.8450, 0.5717, 0.4200, 0.5217, 0.5253, 0.4752],
  33237. [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667],
  33238. [ nan, nan, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533],
  33239. [0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524],
  33240. [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343],
  33241. [0.6271, 0.4005, 0.8450, 0.6067, 0.3850, 0.4983, 0.6069, 0.4649],
  33242. [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817],
  33243. [0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600]],
  33244. device='cuda:0', dtype=torch.float64)
  33245. predictions are: tensor([[ 5.7482e-01, -4.2917e-01, 1.6042e+00, 2.2142e-01, -3.1746e-01,
  33246. 1.4282e-01, 2.6080e-01, 1.7529e-02],
  33247. [ 6.0468e-01, -3.6197e-01, 1.5752e+00, -8.6071e-01, -7.0272e-01,
  33248. -5.9898e-01, 3.5001e-01, 2.9250e-01],
  33249. [-1.7935e+00, -2.0138e+00, 1.3754e+00, -1.0052e+00, -5.5799e-01,
  33250. -9.4388e-01, 3.6216e-01, 2.2459e-01],
  33251. [ 5.5556e-01, -3.8595e-01, 1.7956e+00, -1.1016e-01, -5.4520e-01,
  33252. -1.9756e-01, 1.6915e-01, 1.7724e-01],
  33253. [ 6.3051e-01, -3.9541e-01, 1.8195e+00, -1.0585e+00, -4.3913e-02,
  33254. -1.2640e+00, 9.3094e-01, -8.6043e-04],
  33255. [ 6.1835e-01, -3.9714e-01, 1.6594e+00, 3.2765e-01, -5.4559e-01,
  33256. 2.5476e-02, 5.1325e-01, -1.0676e-01],
  33257. [ 5.3566e-01, -3.9626e-01, 1.4632e+00, -9.3724e-01, -2.8451e-01,
  33258. -1.0030e+00, 3.8548e-01, 3.7415e-01],
  33259. [ 6.2579e-01, -3.6221e-01, 1.8387e+00, 1.5004e-01, -4.5995e-01,
  33260. 4.2121e-03, 3.9129e-01, 2.0803e-01]], device='cuda:0',
  33261. grad_fn=<AddmmBackward>)
  33262. landmarks are: tensor([[[ 0.5132, -0.4736, 1.6171, 0.3546, -0.3460, 0.1236, 0.1404,
  33263. -0.0911],
  33264. [ 0.5768, -0.3857, 1.5305, -0.7694, -0.6462, -0.6308, 0.3988,
  33265. 0.3315],
  33266. [-2.2859, -2.2859, 1.5074, -1.0388, -0.5423, -0.9849, 0.2199,
  33267. 0.2699],
  33268. [ 0.5529, -0.3811, 1.7788, -0.0380, -0.5307, -0.2074, 0.0727,
  33269. 0.2657],
  33270. [ 0.6125, -0.4321, 1.8018, -0.9772, 0.0120, -1.2543, 0.9741,
  33271. 0.1821],
  33272. [ 0.6105, -0.4360, 1.6171, 0.5162, -0.5076, 0.0159, 0.5171,
  33273. -0.1385],
  33274. [ 0.5964, -0.3380, 1.4785, -0.8386, -0.2420, -1.0619, 0.3238,
  33275. 0.4008],
  33276. [ 0.5883, -0.3594, 1.7557, 0.2545, -0.4152, -0.0611, 0.3353,
  33277. 0.3007]]], device='cuda:0')
  33278. loss_train_step before backward: tensor(0.0095, device='cuda:0', grad_fn=<MseLossBackward>)
  33279. loss_train_step after backward: tensor(0.0095, device='cuda:0', grad_fn=<MseLossBackward>)
  33280. loss_train: 0.9303545439615846
  33281. step: 64
  33282. running loss: 0.014536789749399759
  33283. Train Steps: 64/90 Loss: 0.0145 torch.Size([8, 600, 800])
  33284. torch.Size([8, 8])
  33285. tensor([[0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283],
  33286. [0.6098, 0.3991, 0.8638, 0.4717, 0.4263, 0.4967, 0.5212, 0.5650],
  33287. [0.6250, 0.3993, 0.9138, 0.4333, 0.3763, 0.5217, 0.6995, 0.5320],
  33288. [0.6111, 0.3995, 0.8788, 0.4567, 0.3813, 0.4833, 0.5450, 0.5700],
  33289. [0.6284, 0.4127, 0.8538, 0.5867, 0.4363, 0.5083, 0.6038, 0.5433],
  33290. [0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222],
  33291. [0.6250, 0.4103, 0.8950, 0.4400, 0.3912, 0.5650, 0.6050, 0.5133],
  33292. [0.6254, 0.3993, 0.8988, 0.4767, 0.3987, 0.5517, 0.6955, 0.5285]],
  33293. device='cuda:0', dtype=torch.float64)
  33294. predictions are: tensor([[ 0.5780, -0.3837, 1.8008, -0.3853, -0.6711, -0.5892, 0.4198, 0.1675],
  33295. [ 0.3405, -0.5710, 1.7168, -0.1329, -0.2891, -0.0377, 0.1252, 0.1673],
  33296. [ 0.5260, -0.4815, 1.9134, -0.3315, -0.5640, 0.0606, 0.8162, 0.1185],
  33297. [ 0.5334, -0.4503, 1.7712, -0.2396, -0.5370, -0.1090, 0.2359, 0.1811],
  33298. [ 0.4991, -0.4842, 1.6608, 0.3424, -0.2535, -0.0699, 0.4387, 0.1371],
  33299. [ 0.7833, -0.1901, 1.4691, -0.5761, -0.1424, -1.3454, 0.1815, 0.5077],
  33300. [ 0.6020, -0.3899, 1.8731, -0.2790, -0.4664, 0.2550, 0.5829, 0.0653],
  33301. [ 0.7122, -0.3456, 1.8752, -0.1186, -0.4541, 0.2055, 0.8527, 0.0780]],
  33302. device='cuda:0', grad_fn=<AddmmBackward>)
  33303. landmarks are: tensor([[[ 0.5880, -0.3887, 1.8423, -0.3306, -0.6231, -0.5230, 0.4046,
  33304. 0.1544],
  33305. [ 0.5308, -0.4425, 1.7037, -0.1073, -0.3171, 0.0082, 0.1217,
  33306. 0.3238],
  33307. [ 0.6010, -0.4417, 1.9346, -0.2844, -0.5480, 0.1236, 0.9448,
  33308. 0.1715],
  33309. [ 0.5368, -0.4406, 1.7730, -0.1766, -0.5249, -0.0534, 0.2314,
  33310. 0.3469],
  33311. [ 0.6166, -0.3795, 1.6575, 0.4239, -0.2709, 0.0620, 0.5028,
  33312. 0.2237],
  33313. [ 0.6237, -0.2983, 1.3919, -0.4691, -0.0457, -1.2313, 0.2453,
  33314. 0.5882],
  33315. [ 0.6010, -0.3909, 1.8480, -0.2536, -0.4788, 0.3238, 0.5085,
  33316. 0.0851],
  33317. [ 0.6026, -0.4417, 1.8654, -0.0842, -0.4441, 0.2622, 0.9265,
  33318. 0.1554]]], device='cuda:0')
  33319. loss_train_step before backward: tensor(0.0056, device='cuda:0', grad_fn=<MseLossBackward>)
  33320. loss_train_step after backward: tensor(0.0056, device='cuda:0', grad_fn=<MseLossBackward>)
  33321. loss_train: 0.9359597829170525
  33322. step: 65
  33323. running loss: 0.014399381275646961
  33324. Train Steps: 65/90 Loss: 0.0144 torch.Size([8, 600, 800])
  33325. torch.Size([8, 8])
  33326. tensor([[0.6087, 0.3976, 0.8337, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300],
  33327. [0.6199, 0.4093, 0.7913, 0.2533, 0.4288, 0.2467, 0.5975, 0.5700],
  33328. [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167],
  33329. [0.6275, 0.4048, 0.8488, 0.2883, 0.4463, 0.2033, 0.6321, 0.5155],
  33330. [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083],
  33331. [0.6109, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567],
  33332. [0.6064, 0.3953, 0.8738, 0.4417, 0.3663, 0.4683, 0.5511, 0.5416],
  33333. [0.6200, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913]],
  33334. device='cuda:0', dtype=torch.float64)
  33335. predictions are: tensor([[ 0.5275, -0.4615, 1.7064, -0.4898, -0.5600, -0.7662, 0.4385, 0.1653],
  33336. [ 0.6013, -0.4050, 1.4063, -1.1010, -0.3349, -0.9689, 0.5350, 0.2417],
  33337. [ 0.7457, -0.3026, 1.4962, -0.7248, -0.5378, -0.3920, 0.5202, 0.4871],
  33338. [ 0.6284, -0.3964, 1.7495, -0.9078, -0.1837, -1.2448, 0.6356, 0.0381],
  33339. [ 0.6329, -0.3909, 1.8370, 0.0075, -0.5290, -0.1658, 0.4625, 0.3722],
  33340. [ 0.4794, -0.4647, 1.0918, -1.3628, -0.4908, -1.0768, 0.3367, 0.2329],
  33341. [ 0.5490, -0.4589, 1.8760, -0.1951, -0.5087, -0.0391, 0.3459, 0.0651],
  33342. [ 0.5781, -0.4790, 1.8421, 0.1577, -0.2619, 0.1640, 0.4385, -0.0967]],
  33343. device='cuda:0', grad_fn=<AddmmBackward>)
  33344. landmarks are: tensor([[[ 0.5255, -0.4495, 1.5651, -0.4999, -0.5711, -0.8463, 0.4566,
  33345. 0.1621],
  33346. [ 0.5771, -0.3955, 1.3688, -1.1158, -0.3055, -1.1466, 0.4739,
  33347. 0.3469],
  33348. [ 0.5768, -0.3899, 1.3861, -0.7771, -0.5885, -0.5461, 0.5028,
  33349. 0.5624],
  33350. [ 0.6126, -0.4161, 1.6344, -0.9541, -0.2247, -1.3467, 0.6339,
  33351. 0.0953],
  33352. [ 0.5713, -0.4105, 1.7557, 0.0467, -0.6520, -0.2690, 0.3988,
  33353. 0.5239],
  33354. [ 0.5356, -0.4217, 1.0339, -1.4776, -0.5076, -1.1081, 0.4219,
  33355. 0.2853],
  33356. [ 0.5149, -0.4603, 1.7499, -0.2459, -0.5942, -0.1227, 0.2596,
  33357. 0.2155],
  33358. [ 0.5776, -0.4784, 1.7730, 0.1236, -0.4037, 0.0697, 0.5132,
  33359. -0.0168]]], device='cuda:0')
  33360. loss_train_step before backward: tensor(0.0063, device='cuda:0', grad_fn=<MseLossBackward>)
  33361. loss_train_step after backward: tensor(0.0063, device='cuda:0', grad_fn=<MseLossBackward>)
  33362. loss_train: 0.9422118063084781
  33363. step: 66
  33364. running loss: 0.014275936459219365
  33365.  
  33366. Train Steps: 66/90 Loss: 0.0143 torch.Size([8, 600, 800])
  33367. torch.Size([8, 8])
  33368. tensor([[0.6176, 0.3911, 0.8738, 0.4217, 0.3488, 0.4033, 0.6025, 0.4817],
  33369. [ nan, nan, 0.9088, 0.3783, 0.4562, 0.2617, 0.6741, 0.5575],
  33370. [0.6371, 0.4092, 0.8337, 0.5850, 0.3950, 0.5117, 0.6559, 0.5262],
  33371. [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500],
  33372. [0.6204, 0.4007, 0.7838, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167],
  33373. [0.6199, 0.4102, 0.8950, 0.4417, 0.4012, 0.5367, 0.6112, 0.5967],
  33374. [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633],
  33375. [0.6201, 0.3970, 0.8413, 0.4950, 0.4413, 0.5183, 0.6088, 0.5400]],
  33376. device='cuda:0', dtype=torch.float64)
  33377. predictions are: tensor([[ 0.5496, -0.4349, 1.7507, -0.4072, -0.6231, -0.4261, 0.2851, 0.0239],
  33378. [-2.2059, -2.2517, 1.9492, -0.6269, -0.1311, -1.0656, 0.8054, 0.3096],
  33379. [ 0.7202, -0.2978, 1.6103, 0.3138, -0.4298, 0.0925, 0.7170, 0.0483],
  33380. [ 0.4686, -0.4632, 1.3834, -0.9696, -0.6548, -0.4049, 0.5023, 0.1962],
  33381. [ 0.5204, -0.4374, 1.2730, -1.2612, -0.2085, -1.4333, 0.2908, 0.1288],
  33382. [ 0.7266, -0.2690, 1.8107, -0.2156, -0.3568, 0.3110, 0.4842, 0.3369],
  33383. [ 0.6644, -0.3100, 1.8463, 0.0909, -0.5299, -0.1275, 0.2842, 0.2648],
  33384. [ 0.6698, -0.3349, 1.6451, -0.0235, -0.1953, 0.1273, 0.3835, 0.2099]],
  33385. device='cuda:0', grad_fn=<AddmmBackward>)
  33386. landmarks are: tensor([[[ 5.6692e-01, -4.7937e-01, 1.7499e+00, -3.3826e-01, -6.7506e-01,
  33387. -4.2294e-01, 4.9700e-01, -6.1124e-02],
  33388. [-2.2859e+00, -2.2859e+00, 1.9115e+00, -5.3841e-01, -1.7852e-01,
  33389. -1.0773e+00, 8.2783e-01, 2.8902e-01],
  33390. [ 6.5695e-01, -3.9601e-01, 1.5651e+00, 4.1617e-01, -4.6143e-01,
  33391. 7.7444e-02, 7.4375e-01, 1.4474e-01],
  33392. [ 5.7633e-01, -4.1470e-01, 1.3226e+00, -1.0619e+00, -6.6351e-01,
  33393. -4.1524e-01, 5.3741e-01, 2.5450e-01],
  33394. [ 5.7956e-01, -4.3510e-01, 1.3342e+00, -1.3159e+00, -2.1894e-01,
  33395. -1.4853e+00, 4.0462e-01, 1.0054e-01],
  33396. [ 5.7719e-01, -3.9130e-01, 1.8480e+00, -2.4588e-01, -4.3256e-01,
  33397. 1.9292e-01, 5.3741e-01, 4.7005e-01],
  33398. [ 5.8793e-01, -3.5912e-01, 1.8018e+00, 1.2363e-01, -5.5958e-01,
  33399. -1.6120e-01, 3.4688e-01, 3.1609e-01],
  33400. [ 5.7812e-01, -4.5219e-01, 1.5998e+00, 4.6189e-04, -2.4781e-01,
  33401. 1.0824e-01, 5.2587e-01, 2.0831e-01]]], device='cuda:0')
  33402. loss_train_step before backward: tensor(0.0051, device='cuda:0', grad_fn=<MseLossBackward>)
  33403. loss_train_step after backward: tensor(0.0051, device='cuda:0', grad_fn=<MseLossBackward>)
  33404. loss_train: 0.9472914161160588
  33405. step: 67
  33406. running loss: 0.01413867785247849
  33407. Train Steps: 67/90 Loss: 0.0141 torch.Size([8, 600, 800])
  33408. torch.Size([8, 8])
  33409. tensor([[0.6200, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391],
  33410. [0.6212, 0.4033, 0.8938, 0.4167, 0.3813, 0.4267, 0.5613, 0.5583],
  33411. [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5413, 0.5433],
  33412. [0.6131, 0.4064, 0.8638, 0.5200, 0.4788, 0.4783, 0.5258, 0.5867],
  33413. [0.6222, 0.4169, 0.8638, 0.5650, 0.4313, 0.4783, 0.5637, 0.5633],
  33414. [0.6200, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913],
  33415. [0.6286, 0.4086, 0.8408, 0.2801, 0.4163, 0.2800, 0.6725, 0.5393],
  33416. [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351]],
  33417. device='cuda:0', dtype=torch.float64)
  33418. predictions are: tensor([[ 0.6238, -0.3564, 1.6074, -1.0028, -0.3612, -1.0364, 0.5066, 0.2033],
  33419. [ 0.4492, -0.5054, 1.8988, -0.3947, -0.5929, -0.2613, 0.4276, 0.2535],
  33420. [ 0.4352, -0.4504, 1.6065, -0.7117, -0.6445, -0.7353, 0.2741, 0.2579],
  33421. [ 0.5791, -0.3839, 1.6990, 0.0284, -0.1213, 0.0138, 0.2417, 0.3851],
  33422. [ 0.6664, -0.3126, 1.7192, 0.2874, -0.3526, 0.0089, 0.3909, 0.3248],
  33423. [ 0.5155, -0.5051, 1.7634, 0.0443, -0.3976, 0.1064, 0.4973, -0.0456],
  33424. [ 0.6025, -0.4032, 1.5351, -1.0168, -0.3299, -0.9217, 0.9194, 0.3174],
  33425. [ 0.5629, -0.4255, 1.3034, -1.1134, -0.3056, -1.3572, 0.3460, 0.2011]],
  33426. device='cuda:0', grad_fn=<AddmmBackward>)
  33427. landmarks are: tensor([[[ 0.5776, -0.3987, 1.6005, -0.9121, -0.3844, -1.2358, 0.4247,
  33428. 0.2043],
  33429. [ 0.5832, -0.4231, 1.8423, -0.3614, -0.5249, -0.3152, 0.3065,
  33430. 0.2930],
  33431. [ 0.5894, -0.3479, 1.7730, -0.6847, -0.5538, -0.7155, 0.2141,
  33432. 0.2237],
  33433. [ 0.5462, -0.4090, 1.7037, 0.1159, -0.0746, -0.0765, 0.1427,
  33434. 0.4239],
  33435. [ 0.5880, -0.3605, 1.7037, 0.3238, -0.2940, -0.0765, 0.3180,
  33436. 0.3161],
  33437. [ 0.5776, -0.4784, 1.7730, 0.1236, -0.4037, 0.0697, 0.5132,
  33438. -0.0168],
  33439. [ 0.6174, -0.3984, 1.5975, -0.9921, -0.3633, -0.9926, 0.8205,
  33440. 0.2050],
  33441. [ 0.5664, -0.4321, 1.2862, -1.0003, -0.2189, -1.4608, 0.3883,
  33442. 0.1855]]], device='cuda:0')
  33443. loss_train_step before backward: tensor(0.0051, device='cuda:0', grad_fn=<MseLossBackward>)
  33444. loss_train_step after backward: tensor(0.0051, device='cuda:0', grad_fn=<MseLossBackward>)
  33445. loss_train: 0.9523771712556481
  33446. step: 68
  33447. running loss: 0.014005546636112472
  33448. Train Steps: 68/90 Loss: 0.0140 torch.Size([8, 600, 800])
  33449. torch.Size([8, 8])
  33450. tensor([[0.6213, 0.4131, 0.8438, 0.3550, 0.3513, 0.4400, 0.5716, 0.5123],
  33451. [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767],
  33452. [0.6197, 0.4090, 0.7825, 0.2500, 0.4200, 0.2483, 0.5988, 0.5667],
  33453. [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5413, 0.5433],
  33454. [0.6339, 0.4118, 0.7988, 0.5800, 0.3912, 0.4583, 0.7343, 0.5760],
  33455. [0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343],
  33456. [0.6136, 0.4085, 0.6688, 0.2317, 0.3862, 0.2367, 0.5517, 0.5783],
  33457. [0.6250, 0.3961, 0.8672, 0.4929, 0.4199, 0.4972, 0.6312, 0.5301]],
  33458. device='cuda:0', dtype=torch.float64)
  33459. predictions are: tensor([[ 0.5940, -0.3932, 1.7485, -0.6328, -0.6519, -0.2000, 0.3192, 0.1487],
  33460. [ 0.5559, -0.3739, 1.8155, -0.1258, -0.5492, -0.6956, 0.2571, 0.4124],
  33461. [ 0.3322, -0.5662, 1.3762, -1.2519, -0.3556, -1.0895, 0.3842, 0.3540],
  33462. [ 0.5320, -0.3999, 1.7015, -0.6836, -0.5756, -0.6865, 0.1819, 0.2674],
  33463. [ 0.4998, -0.4898, 1.5446, 0.2641, -0.4719, -0.0665, 1.0350, 0.3207],
  33464. [ 0.6239, -0.4223, 1.8720, -1.0206, 0.0367, -1.1793, 0.8825, 0.1049],
  33465. [ 0.5348, -0.4411, 0.8810, -1.2536, -0.4768, -1.1120, 0.1583, 0.3431],
  33466. [ 0.6846, -0.3897, 1.8536, 0.0126, -0.2834, 0.1039, 0.6548, 0.1332]],
  33467. device='cuda:0', grad_fn=<AddmmBackward>)
  33468. landmarks are: tensor([[[ 0.5841, -0.3778, 1.6113, -0.6462, -0.6635, -0.2536, 0.3542,
  33469. 0.0802],
  33470. [ 0.5781, -0.3848, 1.7441, -0.0996, -0.5769, -0.8002, 0.2021,
  33471. 0.3778],
  33472. [ 0.5764, -0.3969, 1.3284, -1.1312, -0.3460, -1.1389, 0.4797,
  33473. 0.3315],
  33474. [ 0.5894, -0.3479, 1.7730, -0.6847, -0.5538, -0.7155, 0.2141,
  33475. 0.2237],
  33476. [ 0.6421, -0.3840, 1.4035, 0.3931, -0.4788, -0.1689, 1.1057,
  33477. 0.3745],
  33478. [ 0.6125, -0.4321, 1.8018, -0.9772, 0.0120, -1.2543, 0.9741,
  33479. 0.1821],
  33480. [ 0.5483, -0.3991, 0.8030, -1.2159, -0.5018, -1.1928, 0.2624,
  33481. 0.3852],
  33482. [ 0.6010, -0.4562, 1.7198, -0.0090, -0.3464, 0.0108, 0.6294,
  33483. 0.1627]]], device='cuda:0')
  33484. loss_train_step before backward: tensor(0.0058, device='cuda:0', grad_fn=<MseLossBackward>)
  33485. loss_train_step after backward: tensor(0.0058, device='cuda:0', grad_fn=<MseLossBackward>)
  33486. loss_train: 0.9581651668995619
  33487. step: 69
  33488. running loss: 0.013886451694196549
  33489. Train Steps: 69/90 Loss: 0.0139 torch.Size([8, 600, 800])
  33490. torch.Size([8, 8])
  33491. tensor([[0.6224, 0.4097, 0.7438, 0.2267, 0.3850, 0.2850, 0.5988, 0.5250],
  33492. [0.6145, 0.4007, 0.8775, 0.4533, 0.4562, 0.5533, 0.6088, 0.5533],
  33493. [0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148],
  33494. [ nan, nan, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999],
  33495. [0.6284, 0.4127, 0.8538, 0.5867, 0.4363, 0.5083, 0.6038, 0.5433],
  33496. [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938],
  33497. [0.6299, 0.4008, 0.8450, 0.5350, 0.4213, 0.5000, 0.6350, 0.5100],
  33498. [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896]],
  33499. device='cuda:0', dtype=torch.float64)
  33500. predictions are: tensor([[ 0.3494, -0.5370, 1.2136, -1.3041, -0.4782, -0.9355, 0.5158, 0.2729],
  33501. [ 0.8429, -0.2129, 1.8653, -0.2707, -0.1902, 0.2714, 0.5865, 0.2691],
  33502. [ 0.6540, -0.3671, 1.8191, 0.0418, -0.4627, -0.0966, 0.4293, 0.1704],
  33503. [-2.2697, -2.2630, 1.3518, -1.1985, -0.3079, -1.3828, 0.2759, 0.2044],
  33504. [ 0.6816, -0.3238, 1.6917, 0.3073, -0.2383, 0.0343, 0.5112, 0.2740],
  33505. [ 0.5939, -0.3918, 1.9143, -0.0183, -0.5544, -0.5055, 0.6066, 0.0947],
  33506. [ 0.7189, -0.3206, 1.6926, 0.0616, -0.3304, -0.0278, 0.7087, 0.2288],
  33507. [ 0.7295, -0.2626, 1.5060, -0.6643, -0.5875, -0.8475, 0.1134, 0.5077]],
  33508. device='cuda:0', grad_fn=<AddmmBackward>)
  33509. landmarks are: tensor([[[ 0.5891, -0.3937, 1.1494, -1.2390, -0.5076, -0.9695, 0.4797,
  33510. 0.1390],
  33511. [ 0.5526, -0.4351, 1.7672, -0.1920, -0.1785, 0.2699, 0.5259,
  33512. 0.2699],
  33513. [ 0.5773, -0.4316, 1.7399, 0.1287, -0.5153, -0.0817, 0.4313,
  33514. 0.0919],
  33515. [-2.2859, -2.2859, 1.3400, -1.0388, -0.3055, -1.4930, 0.1157,
  33516. 0.0231],
  33517. [ 0.6166, -0.3795, 1.6575, 0.4239, -0.2709, 0.0620, 0.5028,
  33518. 0.2237],
  33519. [ 0.6273, -0.4105, 1.8480, 0.1082, -0.5538, -0.5076, 0.6414,
  33520. -0.0049],
  33521. [ 0.6236, -0.4344, 1.6171, 0.1852, -0.3402, 0.0236, 0.6471,
  33522. 0.0697],
  33523. [ 0.5485, -0.3997, 1.4445, -0.4895, -0.6000, -0.8309, 0.1878,
  33524. 0.4374]]], device='cuda:0')
  33525. loss_train_step before backward: tensor(0.0097, device='cuda:0', grad_fn=<MseLossBackward>)
  33526. loss_train_step after backward: tensor(0.0097, device='cuda:0', grad_fn=<MseLossBackward>)
  33527. loss_train: 0.9678890649229288
  33528. step: 70
  33529. running loss: 0.013826986641756125
  33530.  
  33531. Train Steps: 70/90 Loss: 0.0138 torch.Size([8, 600, 800])
  33532. torch.Size([8, 8])
  33533. tensor([[0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383],
  33534. [0.6126, 0.3954, 0.8538, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350],
  33535. [0.6361, 0.4076, 0.8862, 0.5350, 0.3713, 0.4650, 0.6654, 0.5297],
  33536. [0.6161, 0.4076, 0.8900, 0.4667, 0.4125, 0.5917, 0.6262, 0.5367],
  33537. [0.6128, 0.4084, 0.8738, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397],
  33538. [ nan, nan, 0.7725, 0.2611, 0.3675, 0.2733, 0.5413, 0.5167],
  33539. [ nan, nan, 0.7612, 0.3250, 0.4037, 0.2533, 0.5438, 0.5767],
  33540. [0.6109, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567]],
  33541. device='cuda:0', dtype=torch.float64)
  33542. predictions are: tensor([[ 0.8417, -0.2317, 1.7029, 0.0056, -0.2438, -0.0576, 0.3418, 0.2024],
  33543. [ 0.8047, -0.2616, 1.6958, 0.0687, -0.2768, -0.1067, 0.3044, 0.2138],
  33544. [ 0.9906, -0.1865, 1.8446, 0.1755, -0.5234, -0.1499, 0.9876, 0.1798],
  33545. [ 0.9552, -0.1518, 1.8578, -0.1127, -0.2964, 0.4549, 0.7682, 0.1948],
  33546. [ 0.7645, -0.2472, 1.7689, -0.1310, -0.5927, -0.5891, 0.2263, 0.2930],
  33547. [-1.9544, -2.0844, 1.3214, -1.0800, -0.4260, -1.0365, 0.2664, 0.2658],
  33548. [-1.9802, -2.0483, 1.3216, -0.8993, -0.3252, -1.1174, 0.3076, 0.4552],
  33549. [ 0.7517, -0.2976, 1.0570, -1.4151, -0.4750, -1.1065, 0.4526, 0.2947]],
  33550. device='cuda:0', grad_fn=<AddmmBackward>)
  33551. landmarks are: tensor([[[ 5.3603e-01, -4.6490e-01, 1.6517e+00, 4.6189e-04, -2.8245e-01,
  33552. -6.8822e-02, 2.3086e-01, 2.0046e-01],
  33553. [ 5.4348e-01, -4.5974e-01, 1.6575e+00, 1.5858e-02, -3.2286e-01,
  33554. -1.1501e-01, 1.8767e-01, 1.8544e-01],
  33555. [ 6.5201e-01, -4.0323e-01, 1.8076e+00, 1.8522e-01, -5.7113e-01,
  33556. -1.3811e-01, 7.8762e-01, 1.6077e-01],
  33557. [ 5.5978e-01, -4.0323e-01, 1.8249e+00, -1.3041e-01, -3.8060e-01,
  33558. 4.4696e-01, 6.0670e-01, 1.9292e-01],
  33559. [ 5.4434e-01, -3.9938e-01, 1.7499e+00, -1.2271e-01, -6.1732e-01,
  33560. -5.7691e-01, 4.8756e-03, 2.0706e-01],
  33561. [-2.2859e+00, -2.2859e+00, 1.2820e+00, -1.0801e+00, -5.8845e-01,
  33562. -1.0234e+00, 2.1409e-01, 1.0054e-01],
  33563. [-2.2859e+00, -2.2859e+00, 1.2303e+00, -7.8476e-01, -4.2102e-01,
  33564. -1.1158e+00, 2.2564e-01, 3.7768e-01],
  33565. [ 5.3557e-01, -4.2171e-01, 1.0339e+00, -1.4776e+00, -5.0762e-01,
  33566. -1.1081e+00, 4.2194e-01, 2.8530e-01]]], device='cuda:0')
  33567. loss_train_step before backward: tensor(0.0211, device='cuda:0', grad_fn=<MseLossBackward>)
  33568. loss_train_step after backward: tensor(0.0211, device='cuda:0', grad_fn=<MseLossBackward>)
  33569. loss_train: 0.9890259634703398
  33570. step: 71
  33571. running loss: 0.013929943147469575
  33572. Train Steps: 71/90 Loss: 0.0139 torch.Size([8, 600, 800])
  33573. torch.Size([8, 8])
  33574. tensor([[0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904],
  33575. [ nan, nan, 0.8300, 0.3150, 0.3588, 0.3383, 0.5208, 0.5194],
  33576. [0.6192, 0.4128, 0.8513, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633],
  33577. [0.6200, 0.4039, 0.8880, 0.4799, 0.3625, 0.4285, 0.5866, 0.5148],
  33578. [0.6164, 0.4102, 0.8850, 0.4867, 0.4213, 0.5983, 0.5609, 0.5038],
  33579. [0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285],
  33580. [0.6224, 0.4061, 0.8988, 0.4300, 0.3838, 0.4750, 0.6112, 0.5483],
  33581. [ nan, nan, 0.9088, 0.3783, 0.4562, 0.2617, 0.6741, 0.5575]],
  33582. device='cuda:0', dtype=torch.float64)
  33583. predictions are: tensor([[ 0.7189, -0.2970, 1.5169, -1.1217, -0.1463, -1.2613, 0.6161, 0.1181],
  33584. [-1.8552, -1.9676, 1.3940, -0.8310, -0.5719, -0.8433, 0.2184, 0.2288],
  33585. [ 0.9510, -0.1599, 1.5640, 0.3152, -0.3457, 0.1025, 0.5300, 0.3758],
  33586. [ 0.7031, -0.3321, 1.7282, -0.0901, -0.5328, -0.3028, 0.4581, 0.1999],
  33587. [ 0.9493, -0.1386, 1.6505, -0.0565, -0.2638, 0.3398, 0.5135, 0.2446],
  33588. [ 0.6160, -0.3527, 1.2463, -0.4025, -0.5617, -0.2948, 0.1490, 0.2339],
  33589. [ 0.8022, -0.2647, 1.6993, -0.3240, -0.4886, -0.0780, 0.6073, 0.2791],
  33590. [-1.9800, -2.0272, 1.8216, -0.6188, -0.1633, -1.1645, 0.7517, 0.3891]],
  33591. device='cuda:0', grad_fn=<AddmmBackward>)
  33592. landmarks are: tensor([[[ 0.6126, -0.3871, 1.6229, -1.0773, -0.2132, -1.3698, 0.5829,
  33593. -0.0209],
  33594. [-2.2859, -2.2859, 1.5478, -0.8309, -0.6289, -0.7232, 0.1198,
  33595. 0.1133],
  33596. [ 0.5742, -0.3792, 1.6460, 0.3084, -0.3460, 0.1467, 0.4162,
  33597. 0.3161],
  33598. [ 0.5781, -0.4205, 1.8159, -0.0691, -0.6116, -0.3069, 0.4236,
  33599. 0.0919],
  33600. [ 0.5614, -0.3912, 1.8018, -0.0380, -0.3402, 0.4778, 0.3049,
  33601. 0.0412],
  33602. [ 0.5253, -0.4296, 1.3084, -0.3098, -0.6115, -0.2767, 0.0928,
  33603. 0.1552],
  33604. [ 0.5888, -0.4101, 1.8654, -0.2998, -0.5134, -0.0919, 0.5374,
  33605. 0.2468],
  33606. [-2.2859, -2.2859, 1.9115, -0.5384, -0.1785, -1.0773, 0.8278,
  33607. 0.2890]]], device='cuda:0')
  33608. loss_train_step before backward: tensor(0.0214, device='cuda:0', grad_fn=<MseLossBackward>)
  33609. loss_train_step after backward: tensor(0.0214, device='cuda:0', grad_fn=<MseLossBackward>)
  33610. loss_train: 1.010474806651473
  33611. step: 72
  33612. running loss: 0.014034372314603792
  33613. Train Steps: 72/90 Loss: 0.0140 torch.Size([8, 600, 800])
  33614. torch.Size([8, 8])
  33615. tensor([[0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364],
  33616. [0.6266, 0.4070, 0.8712, 0.5600, 0.3713, 0.4783, 0.5775, 0.6100],
  33617. [0.6266, 0.4101, 0.8350, 0.2333, 0.3950, 0.2950, 0.6264, 0.4921],
  33618. [0.6147, 0.4026, 0.6600, 0.2467, 0.4088, 0.2150, 0.5489, 0.5773],
  33619. [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773],
  33620. [0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148],
  33621. [0.6262, 0.4163, 0.8850, 0.5183, 0.3763, 0.4150, 0.6025, 0.5500],
  33622. [0.6263, 0.4233, 0.7924, 0.4626, 0.3788, 0.2883, 0.5573, 0.6047]],
  33623. device='cuda:0', dtype=torch.float64)
  33624. predictions are: tensor([[ 0.3368, -0.5989, 1.3968, -1.2855, -0.2685, -1.2013, 0.8332, 0.1484],
  33625. [ 0.6560, -0.3918, 1.6908, 0.3211, -0.5529, -0.0343, 0.5810, 0.3675],
  33626. [ 0.6294, -0.4085, 1.5747, -1.2638, -0.4456, -0.9089, 0.5935, 0.0504],
  33627. [ 0.3532, -0.5393, 0.8963, -1.2529, -0.4193, -1.2993, 0.1756, 0.4006],
  33628. [ 0.4367, -0.4870, 1.7705, -0.0473, -0.0324, 0.0032, 0.1528, 0.2683],
  33629. [ 0.6308, -0.4134, 1.8200, 0.1440, -0.5066, -0.0720, 0.4684, 0.0920],
  33630. [ 0.4497, -0.4605, 1.8136, 0.0541, -0.5289, -0.4077, 0.5499, 0.3112],
  33631. [ 0.4315, -0.4607, 1.4753, -0.1712, -0.5194, -0.8658, 0.3211, 0.5433]],
  33632. device='cuda:0', grad_fn=<AddmmBackward>)
  33633. landmarks are: tensor([[[ 0.6158, -0.3960, 1.4092, -1.2774, -0.2074, -1.1851, 0.8491,
  33634. 0.1917],
  33635. [ 0.6081, -0.4059, 1.7383, 0.3007, -0.5711, -0.0765, 0.3815,
  33636. 0.5316],
  33637. [ 0.6081, -0.3918, 1.5709, -1.2082, -0.4614, -0.9233, 0.6072,
  33638. -0.0129],
  33639. [ 0.5532, -0.4264, 0.7626, -1.1466, -0.3979, -1.2928, 0.2494,
  33640. 0.3808],
  33641. [ 0.5428, -0.4244, 1.7095, -0.0380, 0.0232, 0.0725, 0.0866,
  33642. 0.3806],
  33643. [ 0.5773, -0.4316, 1.7399, 0.1287, -0.5153, -0.0817, 0.4313,
  33644. 0.0919],
  33645. [ 0.6066, -0.3632, 1.8018, 0.1082, -0.5480, -0.3691, 0.4970,
  33646. 0.2545],
  33647. [ 0.6069, -0.3309, 1.3742, -0.1493, -0.5365, -0.9541, 0.2884,
  33648. 0.5071]]], device='cuda:0')
  33649. loss_train_step before backward: tensor(0.0073, device='cuda:0', grad_fn=<MseLossBackward>)
  33650. loss_train_step after backward: tensor(0.0073, device='cuda:0', grad_fn=<MseLossBackward>)
  33651. loss_train: 1.0177430603653193
  33652. step: 73
  33653. running loss: 0.01394168575842903
  33654. Train Steps: 73/90 Loss: 0.0139 torch.Size([8, 600, 800])
  33655. torch.Size([8, 8])
  33656. tensor([[0.6068, 0.3963, 0.8650, 0.4317, 0.4037, 0.5083, 0.5253, 0.4999],
  33657. [ nan, nan, 0.8363, 0.3317, 0.3563, 0.3367, 0.5329, 0.5142],
  33658. [0.6307, 0.4029, 0.8988, 0.4817, 0.3937, 0.3500, 0.7311, 0.5378],
  33659. [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378],
  33660. [ nan, nan, 0.8850, 0.3000, 0.5363, 0.2250, 0.7343, 0.5771],
  33661. [0.6043, 0.4022, 0.6887, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136],
  33662. [0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114],
  33663. [0.6125, 0.3974, 0.7725, 0.2517, 0.3538, 0.3317, 0.5887, 0.5500]],
  33664. device='cuda:0', dtype=torch.float64)
  33665. predictions are: tensor([[ 0.6748, -0.3631, 1.6707, -0.1981, -0.4835, 0.0423, 0.1712, 0.0683],
  33666. [-2.2875, -2.2457, 1.4929, -0.7364, -0.6461, -0.9071, 0.2137, 0.1640],
  33667. [ 0.6945, -0.3782, 1.8052, -0.0300, -0.4562, -0.6526, 0.8941, 0.2610],
  33668. [ 0.8142, -0.2674, 1.6115, 0.0434, -0.3279, 0.1068, 0.7172, 0.3199],
  33669. [-1.7700, -1.8789, 1.6584, -0.8913, 0.1248, -1.2807, 0.8153, 0.4383],
  33670. [ 0.5807, -0.4324, 0.8782, -1.2439, -0.5315, -1.1515, 0.2223, 0.1654],
  33671. [ 0.8077, -0.2522, 1.6139, 0.1851, -0.2085, 0.1892, 0.3838, 0.1822],
  33672. [ 0.6567, -0.3555, 1.2165, -1.0069, -0.6257, -0.7862, 0.3077, 0.2937]],
  33673. device='cuda:0', grad_fn=<AddmmBackward>)
  33674. landmarks are: tensor([[[ 0.5168, -0.4556, 1.7095, -0.2921, -0.4210, 0.0620, 0.1404,
  33675. 0.0231],
  33676. [-2.2859, -2.2859, 1.5767, -0.7540, -0.6404, -0.7309, 0.1753,
  33677. 0.0893],
  33678. [ 0.6273, -0.4249, 1.8654, -0.0611, -0.4672, -0.6693, 1.0910,
  33679. 0.1982],
  33680. [ 0.6141, -0.4345, 1.6864, -0.0303, -0.2882, 0.1544, 0.9521,
  33681. 0.1982],
  33682. [-2.2859, -2.2859, 1.8018, -0.9002, 0.1910, -1.2467, 1.1057,
  33683. 0.3799],
  33684. [ 0.5053, -0.4281, 0.8954, -1.3698, -0.5423, -1.1389, 0.2453,
  33685. 0.0862],
  33686. [ 0.5450, -0.4730, 1.7420, 0.1372, -0.1919, 0.2614, 0.4976,
  33687. 0.0764],
  33688. [ 0.5430, -0.4503, 1.2822, -1.1235, -0.6520, -0.7540, 0.4335,
  33689. 0.2545]]], device='cuda:0')
  33690. loss_train_step before backward: tensor(0.0173, device='cuda:0', grad_fn=<MseLossBackward>)
  33691. loss_train_step after backward: tensor(0.0173, device='cuda:0', grad_fn=<MseLossBackward>)
  33692. loss_train: 1.0350117646157742
  33693. step: 74
  33694. running loss: 0.013986645467780731
  33695.  
  33696. Train Steps: 74/90 Loss: 0.0140 torch.Size([8, 600, 800])
  33697. torch.Size([8, 8])
  33698. tensor([[0.6223, 0.4171, 0.8750, 0.5500, 0.4050, 0.4817, 0.5675, 0.5600],
  33699. [0.6293, 0.3982, 0.8700, 0.5300, 0.3763, 0.4717, 0.7050, 0.5297],
  33700. [0.6060, 0.3924, 0.8450, 0.5717, 0.4200, 0.5217, 0.5253, 0.4752],
  33701. [0.6085, 0.4008, 0.8588, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283],
  33702. [0.6130, 0.4072, 0.8550, 0.4567, 0.3638, 0.3667, 0.5290, 0.5892],
  33703. [0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835],
  33704. [0.6147, 0.4107, 0.8137, 0.3333, 0.3750, 0.2683, 0.5006, 0.5412],
  33705. [ nan, nan, 0.7612, 0.3250, 0.4037, 0.2533, 0.5438, 0.5767]],
  33706. device='cuda:0', dtype=torch.float64)
  33707. predictions are: tensor([[ 0.5489, -0.4094, 1.6848, 0.0984, -0.4474, -0.0212, 0.4780, 0.3581],
  33708. [ 0.3928, -0.5698, 1.6149, -0.0116, -0.5300, -0.1118, 0.9159, 0.1718],
  33709. [ 0.4654, -0.4955, 1.4773, 0.1833, -0.3214, 0.1166, 0.3754, 0.1369],
  33710. [ 0.3872, -0.5270, 1.5909, -0.0495, -0.0616, -0.0460, 0.1934, 0.1892],
  33711. [ 0.4057, -0.4826, 1.6038, -0.3406, -0.6131, -0.5951, 0.2500, 0.4324],
  33712. [ 0.6082, -0.4286, 1.7353, -0.7135, -0.3392, -1.2192, 0.5991, 0.0500],
  33713. [ 0.4191, -0.5111, 1.3351, -0.8959, -0.5430, -0.9761, 0.1950, 0.1611],
  33714. [-2.9487, -2.6707, 1.2990, -0.8877, -0.3762, -1.1304, 0.2738, 0.2526]],
  33715. device='cuda:0', grad_fn=<AddmmBackward>)
  33716. landmarks are: tensor([[[ 0.5883, -0.3594, 1.7557, 0.2545, -0.4152, -0.0611, 0.3353,
  33717. 0.3007],
  33718. [ 0.6207, -0.4466, 1.7326, 0.1621, -0.5480, -0.1073, 0.9704,
  33719. 0.1608],
  33720. [ 0.5132, -0.4736, 1.6171, 0.3546, -0.3460, 0.1236, 0.1404,
  33721. -0.0911],
  33722. [ 0.5245, -0.4347, 1.6806, 0.1159, 0.0046, 0.0129, 0.1044,
  33723. 0.1544],
  33724. [ 0.5454, -0.4053, 1.6633, -0.1766, -0.6058, -0.5923, 0.1577,
  33725. 0.4357],
  33726. [ 0.6117, -0.3877, 1.8249, -0.5230, -0.3517, -1.2851, 0.4200,
  33727. -0.0529],
  33728. [ 0.5532, -0.3888, 1.4727, -0.7463, -0.5538, -1.0465, 0.0265,
  33729. 0.2138],
  33730. [-2.2859, -2.2859, 1.2303, -0.7848, -0.4210, -1.1158, 0.2256,
  33731. 0.3777]]], device='cuda:0')
  33732. loss_train_step before backward: tensor(0.0204, device='cuda:0', grad_fn=<MseLossBackward>)
  33733. loss_train_step after backward: tensor(0.0204, device='cuda:0', grad_fn=<MseLossBackward>)
  33734. loss_train: 1.055397231131792
  33735. step: 75
  33736. running loss: 0.014071963081757228
  33737. Train Steps: 75/90 Loss: 0.0141 torch.Size([8, 600, 800])
  33738. torch.Size([8, 8])
  33739. tensor([[0.6179, 0.4040, 0.7412, 0.1850, 0.3825, 0.2783, 0.5837, 0.5600],
  33740. [0.6273, 0.4143, 0.8750, 0.5700, 0.3987, 0.4717, 0.6013, 0.5467],
  33741. [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466],
  33742. [0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5837, 0.5583],
  33743. [ nan, nan, 0.7648, 0.2722, 0.3962, 0.2183, 0.5060, 0.5422],
  33744. [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895],
  33745. [0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631],
  33746. [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817]],
  33747. device='cuda:0', dtype=torch.float64)
  33748. predictions are: tensor([[ 0.5642, -0.4334, 1.1553, -1.3390, -0.5370, -1.0926, 0.5606, 0.3162],
  33749. [ 0.4602, -0.4550, 1.6843, 0.3836, -0.4767, -0.2140, 0.6209, 0.2623],
  33750. [ 0.5543, -0.4040, 1.6614, 0.1765, -0.3040, 0.0755, 0.2942, 0.1887],
  33751. [ 0.5686, -0.4199, 1.7921, -0.1467, -0.1592, 0.0329, 0.5323, 0.2321],
  33752. [-2.2041, -2.1930, 1.2239, -0.9110, -0.5603, -1.3102, 0.1554, 0.1432],
  33753. [ 0.1707, -0.7323, 1.2666, -1.3800, -0.2301, -1.5770, 0.5435, 0.0338],
  33754. [ 0.3557, -0.5347, 1.6182, -0.3355, -0.6558, -0.1010, 0.4520, 0.2669],
  33755. [ 0.2443, -0.6425, 1.5739, 0.4655, -0.2944, -0.0588, 0.2487, 0.0846]],
  33756. device='cuda:0', grad_fn=<AddmmBackward>)
  33757. landmarks are: tensor([[[ 0.5680, -0.4201, 1.1379, -1.4314, -0.5192, -1.0003, 0.4104,
  33758. 0.3007],
  33759. [ 0.6115, -0.3724, 1.7557, 0.3469, -0.4441, -0.1073, 0.4912,
  33760. 0.2391],
  33761. [ 0.5178, -0.4480, 1.6748, 0.0620, -0.2767, 0.2083, 0.1067,
  33762. 0.2386],
  33763. [ 0.5223, -0.4336, 1.7557, -0.2074, -0.0573, 0.3084, 0.4104,
  33764. 0.2930],
  33765. [-2.2859, -2.2859, 1.2469, -1.0288, -0.4557, -1.2774, 0.0511,
  33766. 0.2183],
  33767. [ 0.5828, -0.4417, 1.2476, -1.3929, -0.1727, -1.5700, 0.4694,
  33768. -0.0248],
  33769. [ 0.5027, -0.4289, 1.5478, -0.4229, -0.6346, -0.0303, 0.3223,
  33770. 0.3148],
  33771. [ 0.5193, -0.4699, 1.5767, 0.4008, -0.2420, 0.0774, 0.1178,
  33772. -0.0610]]], device='cuda:0')
  33773. loss_train_step before backward: tensor(0.0134, device='cuda:0', grad_fn=<MseLossBackward>)
  33774. loss_train_step after backward: tensor(0.0134, device='cuda:0', grad_fn=<MseLossBackward>)
  33775. loss_train: 1.068770982325077
  33776. step: 76
  33777. running loss: 0.014062776083224699
  33778. Train Steps: 76/90 Loss: 0.0141 torch.Size([8, 600, 800])
  33779. torch.Size([8, 8])
  33780. tensor([[0.6264, 0.4049, 0.8988, 0.4633, 0.3813, 0.4983, 0.6326, 0.4843],
  33781. [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274],
  33782. [0.6126, 0.3954, 0.8538, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350],
  33783. [0.6161, 0.4024, 0.8838, 0.4583, 0.3688, 0.3733, 0.5311, 0.5344],
  33784. [0.6250, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6088, 0.5183],
  33785. [ nan, nan, 0.7097, 0.2346, 0.4250, 0.1850, 0.5175, 0.5583],
  33786. [0.6268, 0.4052, 0.8175, 0.2250, 0.4688, 0.1917, 0.6375, 0.5267],
  33787. [0.6270, 0.4267, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183]],
  33788. device='cuda:0', dtype=torch.float64)
  33789. predictions are: tensor([[ 5.2556e-01, -4.7396e-01, 1.8801e+00, -7.8138e-03, -6.2453e-01,
  33790. 3.7464e-02, 5.8580e-01, -1.2725e-02],
  33791. [ 4.0192e-01, -5.3662e-01, 1.7949e+00, 9.5810e-02, -6.1910e-01,
  33792. -1.8867e-01, 7.2414e-01, 2.0731e-01],
  33793. [ 3.6289e-01, -5.6181e-01, 1.6216e+00, 1.5098e-01, -3.6548e-01,
  33794. -4.9103e-02, 1.0999e-01, 1.4797e-01],
  33795. [ 3.2511e-01, -5.8615e-01, 1.7849e+00, -4.6862e-02, -6.3078e-01,
  33796. -4.6994e-01, 1.9428e-01, 1.4981e-01],
  33797. [ 5.6330e-01, -4.2058e-01, 1.7289e+00, -4.8876e-04, -2.2566e-01,
  33798. 3.5793e-01, 5.1060e-01, 1.5541e-01],
  33799. [-2.2766e+00, -2.2593e+00, 1.0513e+00, -1.0888e+00, -4.0269e-01,
  33800. -1.3363e+00, 1.0775e-01, 1.6716e-01],
  33801. [ 4.5438e-01, -5.3729e-01, 1.4152e+00, -1.1941e+00, -1.9299e-01,
  33802. -1.4076e+00, 7.0603e-01, 1.0110e-01],
  33803. [ 4.3415e-01, -4.6464e-01, 1.0296e+00, -7.9853e-01, -1.5969e-01,
  33804. -1.3360e+00, 3.0307e-01, 4.9511e-01]], device='cuda:0',
  33805. grad_fn=<AddmmBackward>)
  33806. landmarks are: tensor([[[ 0.6075, -0.4159, 1.8654, -0.1458, -0.5249, 0.0159, 0.6359,
  33807. -0.0490],
  33808. [ 0.6108, -0.4008, 1.9088, -0.0253, -0.5769, -0.3075, 0.7905,
  33809. 0.1499],
  33810. [ 0.5435, -0.4597, 1.6575, 0.0159, -0.3229, -0.1150, 0.1877,
  33811. 0.1854],
  33812. [ 0.5598, -0.4273, 1.7961, -0.1689, -0.5827, -0.5615, 0.1671,
  33813. 0.1824],
  33814. [ 0.6012, -0.3846, 1.7326, -0.0457, -0.1381, 0.2853, 0.5259,
  33815. 0.1082],
  33816. [-2.2859, -2.2859, 0.9922, -1.2021, -0.3229, -1.4314, 0.1044,
  33817. 0.2930],
  33818. [ 0.6092, -0.4143, 1.4901, -1.2467, -0.1208, -1.4006, 0.6587,
  33819. 0.1467],
  33820. [ 0.6101, -0.3152, 1.0166, -0.7540, -0.0226, -1.4468, 0.2823,
  33821. 0.5702]]], device='cuda:0')
  33822. loss_train_step before backward: tensor(0.0087, device='cuda:0', grad_fn=<MseLossBackward>)
  33823. loss_train_step after backward: tensor(0.0087, device='cuda:0', grad_fn=<MseLossBackward>)
  33824. loss_train: 1.0775065701454878
  33825. step: 77
  33826. running loss: 0.01399359182007127
  33827. Train Steps: 77/90 Loss: 0.0140 torch.Size([8, 600, 800])
  33828. torch.Size([8, 8])
  33829. tensor([[0.6266, 0.4070, 0.8712, 0.5600, 0.3713, 0.4783, 0.5775, 0.6100],
  33830. [0.6336, 0.4154, 0.8900, 0.2767, 0.4988, 0.2867, 0.7422, 0.5540],
  33831. [0.6117, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103],
  33832. [0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6362, 0.5200],
  33833. [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750],
  33834. [0.6176, 0.3911, 0.8738, 0.4217, 0.3488, 0.4033, 0.6025, 0.4817],
  33835. [0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058],
  33836. [0.6179, 0.4040, 0.7412, 0.1850, 0.3825, 0.2783, 0.5837, 0.5600]],
  33837. device='cuda:0', dtype=torch.float64)
  33838. predictions are: tensor([[ 1.8696e-01, -6.7659e-01, 1.6200e+00, 5.9537e-01, -5.5009e-01,
  33839. -5.2045e-02, 3.8511e-01, 3.9849e-01],
  33840. [ 5.6669e-01, -4.5132e-01, 1.8657e+00, -7.5506e-01, 5.8507e-02,
  33841. -1.1113e+00, 1.0879e+00, 2.1666e-01],
  33842. [ 2.5891e-01, -6.3093e-01, 8.7453e-01, -1.1247e+00, -5.1247e-01,
  33843. -1.2483e+00, 1.9881e-01, 7.4780e-02],
  33844. [ 3.3157e-01, -6.2260e-01, 1.7811e+00, -3.8282e-01, -5.5791e-01,
  33845. 1.4175e-01, 6.5250e-01, 1.2902e-01],
  33846. [ 1.8281e-01, -6.8666e-01, 1.7092e+00, 2.4061e-01, -2.2659e-01,
  33847. -6.1996e-02, 1.3360e-01, 2.7632e-01],
  33848. [ 5.4019e-01, -4.9346e-01, 1.7867e+00, -1.4107e-01, -6.5465e-01,
  33849. -4.7040e-01, 4.3645e-01, 1.0568e-03],
  33850. [ 3.5157e-01, -5.6563e-01, 1.6518e+00, -7.9254e-02, -5.3259e-01,
  33851. -8.8585e-02, 6.9966e-02, 4.8155e-02],
  33852. [ 6.5928e-01, -3.6591e-01, 1.1671e+00, -1.2855e+00, -5.0414e-01,
  33853. -1.1079e+00, 4.3803e-01, 2.8392e-01]], device='cuda:0',
  33854. grad_fn=<AddmmBackward>)
  33855. landmarks are: tensor([[[ 0.6081, -0.4059, 1.7383, 0.3007, -0.5711, -0.0765, 0.3815,
  33856. 0.5316],
  33857. [ 0.6405, -0.3671, 1.8249, -1.0080, 0.0178, -0.9618, 1.1422,
  33858. 0.2730],
  33859. [ 0.5393, -0.4299, 0.7453, -1.3775, -0.5596, -1.1081, 0.1527,
  33860. 0.0712],
  33861. [ 0.5841, -0.4199, 1.7961, -0.6693, -0.6231, 0.1082, 0.6529,
  33862. 0.1159],
  33863. [ 0.5525, -0.4463, 1.7326, 0.0313, -0.2536, -0.0688, 0.1968,
  33864. 0.3700],
  33865. [ 0.5669, -0.4794, 1.7499, -0.3383, -0.6751, -0.4229, 0.4970,
  33866. -0.0611],
  33867. [ 0.5245, -0.4361, 1.5940, -0.2921, -0.5480, -0.0919, 0.2432,
  33868. 0.0502],
  33869. [ 0.5680, -0.4201, 1.1379, -1.4314, -0.5192, -1.0003, 0.4104,
  33870. 0.3007]]], device='cuda:0')
  33871. loss_train_step before backward: tensor(0.0214, device='cuda:0', grad_fn=<MseLossBackward>)
  33872. loss_train_step after backward: tensor(0.0214, device='cuda:0', grad_fn=<MseLossBackward>)
  33873. loss_train: 1.0989203732460737
  33874. step: 78
  33875. running loss: 0.014088722733924022
  33876.  
  33877. Train Steps: 78/90 Loss: 0.0141 torch.Size([8, 600, 800])
  33878. torch.Size([8, 8])
  33879. tensor([[0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148],
  33880. [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391],
  33881. [0.6272, 0.4071, 0.8738, 0.5600, 0.3675, 0.3783, 0.5926, 0.4742],
  33882. [0.6243, 0.4128, 0.7762, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417],
  33883. [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5513, 0.5750],
  33884. [ nan, nan, 0.6900, 0.1917, 0.3937, 0.2367, 0.5240, 0.5246],
  33885. [0.6260, 0.4133, 0.8800, 0.5117, 0.3713, 0.3283, 0.6223, 0.4967],
  33886. [0.6179, 0.4082, 0.6688, 0.2667, 0.3588, 0.3317, 0.5750, 0.5783]],
  33887. device='cuda:0', dtype=torch.float64)
  33888. predictions are: tensor([[ 0.3864, -0.5742, 1.8245, 0.1772, -0.4627, 0.0362, 0.2757, 0.0727],
  33889. [ 0.3755, -0.5567, 1.5371, -1.0199, -0.3176, -1.1191, 0.5130, 0.1074],
  33890. [ 0.5378, -0.4717, 1.8535, 0.2957, -0.5370, -0.3660, 0.3919, -0.0246],
  33891. [ 0.4323, -0.5277, 1.4370, -1.0060, -0.4104, -0.7326, 0.5667, 0.1857],
  33892. [ 0.3792, -0.5396, 0.9493, -1.1634, -0.2327, -1.3323, 0.2315, 0.4083],
  33893. [-2.9494, -2.7514, 1.1699, -1.0930, -0.3473, -1.1785, 0.1991, 0.1744],
  33894. [ 0.4948, -0.4716, 1.8885, 0.0686, -0.4718, -0.5718, 0.5318, 0.0908],
  33895. [ 0.4863, -0.4793, 1.0017, -0.9466, -0.5978, -0.7258, 0.2886, 0.2970]],
  33896. device='cuda:0', grad_fn=<AddmmBackward>)
  33897. landmarks are: tensor([[[ 0.5773, -0.4316, 1.7399, 0.1287, -0.5153, -0.0817, 0.4313,
  33898. 0.0919],
  33899. [ 0.5785, -0.4259, 1.4228, -1.0261, -0.4190, -1.2189, 0.4763,
  33900. 0.2043],
  33901. [ 0.6113, -0.4057, 1.7499, 0.3007, -0.5885, -0.5384, 0.4513,
  33902. -0.0957],
  33903. [ 0.5977, -0.3792, 1.2995, -1.0311, -0.5192, -0.8386, 0.5836,
  33904. 0.2160],
  33905. [ 0.5663, -0.4396, 0.8261, -1.1312, -0.2940, -1.3929, 0.2603,
  33906. 0.3700],
  33907. [-2.2859, -2.2859, 0.9012, -1.4006, -0.4672, -1.1928, 0.1342,
  33908. 0.1373],
  33909. [ 0.6054, -0.3767, 1.7788, 0.0774, -0.5711, -0.7694, 0.5887,
  33910. 0.0081],
  33911. [ 0.5680, -0.4005, 0.8030, -1.0542, -0.6289, -0.7540, 0.3700,
  33912. 0.3854]]], device='cuda:0')
  33913. loss_train_step before backward: tensor(0.0227, device='cuda:0', grad_fn=<MseLossBackward>)
  33914. loss_train_step after backward: tensor(0.0227, device='cuda:0', grad_fn=<MseLossBackward>)
  33915. loss_train: 1.121607020497322
  33916. step: 79
  33917. running loss: 0.01419755722148509
  33918. Train Steps: 79/90 Loss: 0.0142 torch.Size([8, 600, 800])
  33919. torch.Size([8, 8])
  33920. tensor([[0.6236, 0.3966, 0.8850, 0.4967, 0.3638, 0.4017, 0.5850, 0.5183],
  33921. [0.6122, 0.3993, 0.8738, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927],
  33922. [0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5988, 0.5700],
  33923. [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332],
  33924. [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389],
  33925. [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317],
  33926. [0.6299, 0.4008, 0.8450, 0.5350, 0.4213, 0.5000, 0.6350, 0.5100],
  33927. [0.6276, 0.4235, 0.8888, 0.5333, 0.3800, 0.3117, 0.5427, 0.6164]],
  33928. device='cuda:0', dtype=torch.float64)
  33929. predictions are: tensor([[ 0.5590, -0.4708, 1.7701, -0.0477, -0.5697, -0.4099, 0.3951, 0.1113],
  33930. [ 0.2881, -0.6408, 1.7571, -0.1415, -0.1678, 0.0765, -0.0485, -0.0760],
  33931. [ 0.6762, -0.3469, 1.4456, -0.9191, -0.5233, -0.5599, 0.4077, 0.1622],
  33932. [ 0.5194, -0.5119, 1.8524, -0.1059, -0.5642, -0.4052, 0.7536, 0.1808],
  33933. [ 0.5069, -0.4797, 1.2314, -1.3000, -0.4628, -0.9837, 0.5290, 0.1718],
  33934. [ 0.5393, -0.4442, 1.7180, 0.1891, -0.5061, -0.5392, 0.3002, 0.1873],
  33935. [ 0.4009, -0.5789, 1.6085, 0.1635, -0.3419, 0.0440, 0.5460, 0.1089],
  33936. [ 0.1756, -0.6277, 1.6535, 0.0782, -0.4525, -0.8386, 0.2303, 0.5210]],
  33937. device='cuda:0', grad_fn=<AddmmBackward>)
  33938. landmarks are: tensor([[[ 0.5944, -0.4543, 1.8018, 0.0082, -0.6058, -0.4306, 0.4162,
  33939. 0.1082],
  33940. [ 0.5417, -0.4417, 1.7499, -0.1304, -0.1994, -0.0324, 0.0951,
  33941. -0.0099],
  33942. [ 0.5960, -0.3888, 1.4840, -1.0095, -0.6115, -0.6231, 0.4797,
  33943. 0.3469],
  33944. [ 0.6520, -0.4056, 1.9173, -0.0765, -0.5596, -0.4537, 0.7949,
  33945. 0.1768],
  33946. [ 0.5697, -0.4466, 1.1973, -1.1871, -0.4571, -0.9965, 0.5219,
  33947. 0.2032],
  33948. [ 0.5901, -0.4157, 1.7557, 0.1929, -0.5423, -0.5923, 0.3584,
  33949. 0.1698],
  33950. [ 0.6236, -0.4344, 1.6171, 0.1852, -0.3402, 0.0236, 0.6471,
  33951. 0.0697],
  33952. [ 0.6131, -0.3299, 1.8192, 0.1775, -0.5307, -0.8463, 0.2206,
  33953. 0.5612]]], device='cuda:0')
  33954. loss_train_step before backward: tensor(0.0109, device='cuda:0', grad_fn=<MseLossBackward>)
  33955. loss_train_step after backward: tensor(0.0109, device='cuda:0', grad_fn=<MseLossBackward>)
  33956. loss_train: 1.1325139282271266
  33957. step: 80
  33958. running loss: 0.014156424102839082
  33959. Train Steps: 80/90 Loss: 0.0142 torch.Size([8, 600, 800])
  33960. torch.Size([8, 8])
  33961. tensor([[ nan, nan, 0.8625, 0.2550, 0.5487, 0.2200, 0.7335, 0.5737],
  33962. [0.6182, 0.3982, 0.7541, 0.2379, 0.3959, 0.2792, 0.6079, 0.5389],
  33963. [0.6193, 0.4108, 0.7425, 0.2350, 0.3887, 0.2750, 0.5900, 0.5717],
  33964. [0.6184, 0.4079, 0.8350, 0.3700, 0.3675, 0.2883, 0.5312, 0.5783],
  33965. [0.6254, 0.3993, 0.8988, 0.4767, 0.3987, 0.5517, 0.6955, 0.5285],
  33966. [0.6317, 0.4038, 0.8287, 0.5900, 0.3800, 0.4717, 0.6295, 0.4986],
  33967. [0.6353, 0.4128, 0.8488, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540],
  33968. [0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567]],
  33969. device='cuda:0', dtype=torch.float64)
  33970. predictions are: tensor([[-2.6894, -2.5739, 1.5702, -1.0534, 0.0456, -1.2679, 0.7645, 0.2454],
  33971. [ 0.4485, -0.5052, 1.2072, -1.1342, -0.5686, -0.9861, 0.4241, 0.1608],
  33972. [ 0.4587, -0.4829, 1.0750, -1.1103, -0.5859, -0.9948, 0.3502, 0.2427],
  33973. [ 0.5041, -0.4563, 1.5614, -0.5170, -0.6090, -0.8942, -0.1033, 0.2127],
  33974. [ 0.5302, -0.4635, 1.8460, -0.0276, -0.5491, 0.2861, 0.7146, 0.0353],
  33975. [ 0.5678, -0.4470, 1.5932, 0.4860, -0.6324, -0.0811, 0.4204, -0.0671],
  33976. [ 0.2759, -0.6126, 1.5074, -1.1428, 0.0767, -1.4070, 0.6163, 0.1728],
  33977. [ 0.5419, -0.3975, 1.6601, -0.1506, -0.1774, 0.0067, 0.1483, 0.1847]],
  33978. device='cuda:0', grad_fn=<AddmmBackward>)
  33979. landmarks are: tensor([[[-2.2859, -2.2859, 1.6979, -1.1081, 0.2487, -1.2697, 1.1020,
  33980. 0.3638],
  33981. [ 0.5697, -0.4466, 1.1973, -1.1871, -0.4571, -0.9965, 0.5219,
  33982. 0.2032],
  33983. [ 0.5746, -0.3882, 1.1436, -1.2005, -0.4903, -1.0157, 0.4393,
  33984. 0.3546],
  33985. [ 0.5704, -0.4019, 1.5709, -0.5769, -0.5885, -0.9541, 0.1679,
  33986. 0.3854],
  33987. [ 0.6026, -0.4417, 1.8654, -0.0842, -0.4441, 0.2622, 0.9265,
  33988. 0.1554],
  33989. [ 0.6320, -0.4206, 1.5420, 0.4393, -0.5307, -0.1073, 0.6216,
  33990. 0.0171],
  33991. [ 0.6487, -0.3792, 1.6344, -1.0850, 0.2659, -1.5397, 0.8059,
  33992. 0.2730],
  33993. [ 0.5538, -0.4139, 1.7557, -0.1843, -0.0459, 0.1242, 0.4219,
  33994. 0.2853]]], device='cuda:0')
  33995. loss_train_step before backward: tensor(0.0195, device='cuda:0', grad_fn=<MseLossBackward>)
  33996. loss_train_step after backward: tensor(0.0195, device='cuda:0', grad_fn=<MseLossBackward>)
  33997. loss_train: 1.1520524276420474
  33998. step: 81
  33999. running loss: 0.014222869477062314
  34000. Train Steps: 81/90 Loss: 0.0142 torch.Size([8, 600, 800])
  34001. torch.Size([8, 8])
  34002. tensor([[0.6248, 0.4185, 0.8500, 0.5767, 0.4463, 0.4550, 0.5613, 0.5917],
  34003. [0.6189, 0.4049, 0.8888, 0.4417, 0.4213, 0.5200, 0.5988, 0.5633],
  34004. [0.6043, 0.4022, 0.6887, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136],
  34005. [0.6038, 0.3946, 0.8413, 0.4883, 0.3563, 0.4550, 0.5266, 0.4693],
  34006. [ nan, nan, 0.7612, 0.3250, 0.4037, 0.2533, 0.5438, 0.5767],
  34007. [0.6271, 0.4081, 0.9081, 0.4894, 0.3700, 0.4283, 0.6661, 0.5274],
  34008. [0.6182, 0.4058, 0.8738, 0.4350, 0.3563, 0.3400, 0.5290, 0.5822],
  34009. [0.6164, 0.4066, 0.6625, 0.3033, 0.3775, 0.2967, 0.5725, 0.5833]],
  34010. device='cuda:0', dtype=torch.float64)
  34011. predictions are: tensor([[ 0.6217, -0.3808, 1.7538, 0.3747, -0.1461, -0.2232, 0.3396, 0.3322],
  34012. [ 0.5353, -0.4553, 1.8654, -0.2540, -0.3222, 0.1267, 0.5113, 0.1193],
  34013. [ 0.5424, -0.4752, 0.9778, -1.3349, -0.4621, -1.2054, 0.3810, -0.0704],
  34014. [ 0.5623, -0.4164, 1.7162, -0.0753, -0.5480, -0.1832, 0.1149, -0.0326],
  34015. [-2.3358, -2.3530, 1.2751, -0.9237, -0.3565, -1.1791, 0.2657, 0.3107],
  34016. [ 0.6190, -0.3997, 1.9346, -0.0340, -0.5371, -0.2918, 0.8412, 0.0862],
  34017. [ 0.6145, -0.3911, 1.7789, -0.3152, -0.5916, -0.7133, 0.2112, 0.2105],
  34018. [ 0.3885, -0.5344, 0.9275, -0.9589, -0.5050, -1.0208, 0.3161, 0.3429]],
  34019. device='cuda:0', grad_fn=<AddmmBackward>)
  34020. landmarks are: tensor([[[ 0.5999, -0.3530, 1.6402, 0.3777, -0.2247, -0.1843, 0.3065,
  34021. 0.4470],
  34022. [ 0.5726, -0.4159, 1.8192, -0.2459, -0.3402, 0.1159, 0.4797,
  34023. 0.3161],
  34024. [ 0.5053, -0.4281, 0.8954, -1.3698, -0.5423, -1.1389, 0.2453,
  34025. 0.0862],
  34026. [ 0.5030, -0.4631, 1.5998, -0.0303, -0.6404, -0.1843, 0.1465,
  34027. -0.1181],
  34028. [-2.2859, -2.2859, 1.2303, -0.7848, -0.4210, -1.1158, 0.2256,
  34029. 0.3777],
  34030. [ 0.6108, -0.4008, 1.9088, -0.0253, -0.5769, -0.3075, 0.7905,
  34031. 0.1499],
  34032. [ 0.5697, -0.4113, 1.7499, -0.2767, -0.6404, -0.7155, 0.1576,
  34033. 0.4032],
  34034. [ 0.5614, -0.4080, 0.7741, -0.8848, -0.5423, -0.9156, 0.3584,
  34035. 0.4085]]], device='cuda:0')
  34036. loss_train_step before backward: tensor(0.0059, device='cuda:0', grad_fn=<MseLossBackward>)
  34037. loss_train_step after backward: tensor(0.0059, device='cuda:0', grad_fn=<MseLossBackward>)
  34038. loss_train: 1.157949673011899
  34039. step: 82
  34040. running loss: 0.014121337475754865
  34041.  
  34042. Train Steps: 82/90 Loss: 0.0141 torch.Size([8, 600, 800])
  34043. torch.Size([8, 8])
  34044. tensor([[0.6300, 0.4133, 0.8538, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413],
  34045. [0.6251, 0.4163, 0.8662, 0.4467, 0.3625, 0.3567, 0.6038, 0.5533],
  34046. [0.6185, 0.4129, 0.8900, 0.4567, 0.3937, 0.5417, 0.5734, 0.5110],
  34047. [0.6200, 0.3999, 0.8653, 0.5207, 0.4100, 0.5125, 0.5975, 0.5103],
  34048. [0.6129, 0.4114, 0.6950, 0.2467, 0.4050, 0.2133, 0.5363, 0.5550],
  34049. [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5787, 0.5550],
  34050. [0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947],
  34051. [0.6126, 0.4039, 0.8237, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138]],
  34052. device='cuda:0', dtype=torch.float64)
  34053. predictions are: tensor([[ 0.4949, -0.4959, 1.6806, -1.2023, 0.2291, -1.2822, 1.0642, 0.2773],
  34054. [ 0.5310, -0.4471, 1.7362, -0.1250, -0.6350, -0.7626, 0.4045, 0.2777],
  34055. [ 0.6785, -0.3380, 1.8746, -0.0475, -0.4372, 0.1743, 0.2674, 0.1442],
  34056. [ 0.6515, -0.4176, 1.7337, 0.2529, -0.4218, 0.0880, 0.3510, 0.0264],
  34057. [ 0.6312, -0.3546, 0.8784, -0.9879, -0.4545, -1.2742, 0.0937, 0.2359],
  34058. [ 0.6393, -0.3567, 1.4234, -0.8862, -0.6905, -0.6463, 0.3072, 0.2282],
  34059. [ 0.4639, -0.5364, 1.3209, -1.2894, -0.3382, -1.3093, 0.6672, -0.0192],
  34060. [ 0.6300, -0.3829, 1.5335, -0.3926, -0.6023, -0.6566, 0.4111, 0.4879]],
  34061. device='cuda:0', grad_fn=<AddmmBackward>)
  34062. landmarks are: tensor([[[ 6.2401e-01, -3.7675e-01, 1.6575e+00, -1.2851e+00, 2.9492e-01,
  34063. -1.2467e+00, 1.1276e+00, 2.1421e-01],
  34064. [ 6.0162e-01, -3.6328e-01, 1.7152e+00, -2.2279e-01, -6.1155e-01,
  34065. -6.3849e-01, 5.0277e-01, 2.6990e-01],
  34066. [ 5.7113e-01, -3.7875e-01, 1.8249e+00, -1.7660e-01, -4.6721e-01,
  34067. 2.1601e-01, 3.6246e-01, 7.4222e-02],
  34068. [ 5.7777e-01, -4.3888e-01, 1.7107e+00, 1.1921e-01, -3.9207e-01,
  34069. 8.1507e-02, 4.7413e-01, 7.1077e-02],
  34070. [ 5.4480e-01, -3.8591e-01, 9.2425e-01, -1.1466e+00, -4.1524e-01,
  34071. -1.3005e+00, 1.9099e-01, 2.7760e-01],
  34072. [ 5.6472e-01, -4.1286e-01, 1.4901e+00, -1.0619e+00, -6.4619e-01,
  34073. -5.8460e-01, 3.8730e-01, 2.7760e-01],
  34074. [ 6.0739e-01, -4.0631e-01, 1.3630e+00, -1.5238e+00, -3.1132e-01,
  34075. -1.2390e+00, 6.1950e-01, -9.2270e-04],
  34076. [ 5.4376e-01, -4.2055e-01, 1.5189e+00, -4.5373e-01, -6.1155e-01,
  34077. -6.2309e-01, 4.3649e-01, 5.4914e-01]]], device='cuda:0')
  34078. loss_train_step before backward: tensor(0.0066, device='cuda:0', grad_fn=<MseLossBackward>)
  34079. loss_train_step after backward: tensor(0.0066, device='cuda:0', grad_fn=<MseLossBackward>)
  34080. loss_train: 1.164504498243332
  34081. step: 83
  34082. running loss: 0.014030174677630505
  34083. Train Steps: 83/90 Loss: 0.0140 torch.Size([8, 600, 800])
  34084. torch.Size([8, 8])
  34085. tensor([[0.6213, 0.4001, 0.7712, 0.2117, 0.4338, 0.1800, 0.5850, 0.5183],
  34086. [0.6219, 0.4097, 0.8738, 0.3400, 0.3563, 0.4117, 0.5975, 0.5683],
  34087. [0.6200, 0.4112, 0.8862, 0.4100, 0.3638, 0.4917, 0.6088, 0.6050],
  34088. [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367],
  34089. [0.6216, 0.4167, 0.8588, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667],
  34090. [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767],
  34091. [0.6339, 0.4102, 0.9088, 0.4767, 0.3925, 0.5283, 0.7509, 0.5390],
  34092. [0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090]],
  34093. device='cuda:0', dtype=torch.float64)
  34094. predictions are: tensor([[ 0.4710, -0.5196, 1.1014, -1.3862, -0.2148, -1.5333, 0.3553, 0.0757],
  34095. [ 0.7050, -0.3146, 1.6446, -0.7035, -0.6196, -0.5095, 0.3162, 0.2193],
  34096. [ 0.7285, -0.2890, 1.6976, -0.3794, -0.5090, -0.0404, 0.3566, 0.4185],
  34097. [ 0.3456, -0.5866, 1.2864, -0.9612, -0.5184, -0.8896, 0.4405, 0.2379],
  34098. [ 0.8533, -0.2240, 1.6473, 0.3931, -0.4096, 0.0653, 0.3822, 0.2566],
  34099. [ 0.5720, -0.3857, 1.6606, -0.0632, -0.5481, -0.8728, 0.1911, 0.3357],
  34100. [ 0.6318, -0.4071, 1.8293, -0.1566, -0.4036, 0.1316, 1.0711, 0.2480],
  34101. [ 0.7790, -0.3147, 1.7108, -0.5726, -0.3795, -1.0664, 0.6091, 0.0483]],
  34102. device='cuda:0', grad_fn=<AddmmBackward>)
  34103. landmarks are: tensor([[[ 0.5838, -0.4379, 1.2764, -1.3082, -0.2824, -1.4545, 0.4162,
  34104. 0.1082],
  34105. [ 0.5867, -0.3937, 1.7499, -0.7155, -0.6404, -0.3844, 0.4739,
  34106. 0.3392],
  34107. [ 0.5777, -0.3868, 1.8076, -0.3921, -0.6058, -0.0149, 0.5259,
  34108. 0.5085],
  34109. [ 0.5697, -0.4442, 1.3529, -0.9515, -0.5774, -0.7801, 0.5253,
  34110. 0.1931],
  34111. [ 0.5853, -0.3614, 1.6806, 0.2930, -0.4499, 0.1005, 0.3815,
  34112. 0.3315],
  34113. [ 0.5781, -0.3848, 1.7441, -0.0996, -0.5769, -0.8002, 0.2021,
  34114. 0.3778],
  34115. [ 0.6421, -0.3912, 1.9115, -0.0842, -0.4730, 0.1544, 1.1824,
  34116. 0.2035],
  34117. [ 0.6165, -0.4249, 1.7961, -0.5384, -0.4499, -0.9695, 0.6401,
  34118. 0.0652]]], device='cuda:0')
  34119. loss_train_step before backward: tensor(0.0089, device='cuda:0', grad_fn=<MseLossBackward>)
  34120. loss_train_step after backward: tensor(0.0089, device='cuda:0', grad_fn=<MseLossBackward>)
  34121. loss_train: 1.173423076979816
  34122. step: 84
  34123. running loss: 0.013969322344997809
  34124. Train Steps: 84/90 Loss: 0.0140 torch.Size([8, 600, 800])
  34125. torch.Size([8, 8])
  34126. tensor([[0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117],
  34127. [0.6229, 0.4107, 0.8137, 0.2883, 0.4750, 0.1717, 0.5813, 0.5400],
  34128. [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5413, 0.5433],
  34129. [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837],
  34130. [0.6109, 0.4003, 0.8650, 0.4883, 0.4775, 0.4867, 0.5175, 0.5683],
  34131. [0.6277, 0.4103, 0.8087, 0.5717, 0.4188, 0.4750, 0.5663, 0.6083],
  34132. [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100],
  34133. [ nan, nan, 0.8213, 0.2700, 0.3775, 0.2817, 0.5425, 0.5533]],
  34134. device='cuda:0', dtype=torch.float64)
  34135. predictions are: tensor([[ 0.7204, -0.3026, 1.6254, 0.1875, -0.4491, -0.1383, 0.5085, 0.3773],
  34136. [ 0.5447, -0.3915, 1.5013, -1.1210, -0.1454, -1.4991, 0.5017, 0.1182],
  34137. [ 0.5424, -0.3848, 1.6505, -0.8521, -0.5953, -0.8380, 0.3130, 0.1086],
  34138. [ 0.5663, -0.3752, 0.8812, -1.0786, -0.5915, -1.0182, 0.2309, 0.2956],
  34139. [ 0.5830, -0.4065, 1.7906, -0.2180, -0.1265, -0.1026, 0.2333, 0.1712],
  34140. [ 0.7589, -0.2776, 1.6148, 0.1922, -0.3882, -0.1449, 0.5068, 0.4028],
  34141. [ 0.5904, -0.4397, 1.7383, 0.0546, -0.4931, -0.0932, 0.7224, 0.0257],
  34142. [-2.2089, -2.2824, 1.3509, -1.0788, -0.5869, -1.0564, 0.3734, 0.2143]],
  34143. device='cuda:0', grad_fn=<AddmmBackward>)
  34144. landmarks are: tensor([[[ 0.6118, -0.3983, 1.5824, 0.3469, -0.4268, -0.0688, 0.3469,
  34145. 0.5393],
  34146. [ 0.5911, -0.3888, 1.4727, -0.9541, -0.0919, -1.4930, 0.3988,
  34147. 0.2083],
  34148. [ 0.5894, -0.3479, 1.7730, -0.6847, -0.5538, -0.7155, 0.2141,
  34149. 0.2237],
  34150. [ 0.5629, -0.3871, 0.7243, -0.9581, -0.5827, -0.9849, 0.1288,
  34151. 0.4103],
  34152. [ 0.5359, -0.4370, 1.7095, -0.0303, -0.0804, -0.0380, 0.1044,
  34153. 0.3392],
  34154. [ 0.6134, -0.3910, 1.4497, 0.3546, -0.3517, -0.0919, 0.3296,
  34155. 0.5239],
  34156. [ 0.6125, -0.4273, 1.6402, 0.2006, -0.4788, -0.0303, 0.6182,
  34157. 0.0697],
  34158. [-2.2859, -2.2859, 1.5074, -1.0388, -0.5423, -0.9849, 0.2199,
  34159. 0.2699]]], device='cuda:0')
  34160. loss_train_step before backward: tensor(0.0103, device='cuda:0', grad_fn=<MseLossBackward>)
  34161. loss_train_step after backward: tensor(0.0103, device='cuda:0', grad_fn=<MseLossBackward>)
  34162. loss_train: 1.1836915975436568
  34163. step: 85
  34164. running loss: 0.01392578350051361
  34165. Train Steps: 85/90 Loss: 0.0139 torch.Size([8, 600, 800])
  34166. torch.Size([8, 8])
  34167. tensor([[0.6146, 0.4124, 0.8800, 0.4867, 0.3800, 0.4500, 0.5106, 0.5524],
  34168. [0.6199, 0.4065, 0.7598, 0.2385, 0.4317, 0.1981, 0.5933, 0.5221],
  34169. [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6038, 0.4833],
  34170. [0.6233, 0.4091, 0.8100, 0.2950, 0.3563, 0.3883, 0.6013, 0.5200],
  34171. [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617],
  34172. [0.6361, 0.4165, 0.9038, 0.5017, 0.3750, 0.3783, 0.6448, 0.5309],
  34173. [0.6230, 0.4152, 0.7588, 0.2283, 0.4012, 0.2883, 0.6200, 0.5767],
  34174. [ nan, nan, 0.7525, 0.2291, 0.3838, 0.3017, 0.6050, 0.5667]],
  34175. device='cuda:0', dtype=torch.float64)
  34176. predictions are: tensor([[ 0.7068, -0.2859, 1.7985, 0.0704, -0.5089, -0.2907, 0.1986, 0.2836],
  34177. [ 0.8525, -0.2179, 1.2044, -1.2195, -0.2814, -1.3455, 0.5076, 0.1458],
  34178. [ 0.8303, -0.2774, 1.7756, -0.0766, -0.5239, -0.1445, 0.4094, 0.0691],
  34179. [ 0.5732, -0.3968, 1.4232, -0.9013, -0.6039, -0.4955, 0.5431, 0.2011],
  34180. [ 0.6532, -0.3466, 1.7894, -0.0737, -0.2789, -0.0805, 0.4266, 0.3231],
  34181. [ 0.7064, -0.3133, 1.8737, 0.0259, -0.4378, -0.6395, 0.7381, 0.2126],
  34182. [ 0.5834, -0.3917, 1.1698, -1.2030, -0.4331, -1.0781, 0.5621, 0.4097],
  34183. [-1.9059, -2.0811, 1.1619, -1.2046, -0.4429, -1.0495, 0.4665, 0.3691]],
  34184. device='cuda:0', grad_fn=<AddmmBackward>)
  34185. landmarks are: tensor([[[ 0.5529, -0.3811, 1.7788, -0.0380, -0.5307, -0.2074, 0.0727,
  34186. 0.2657],
  34187. [ 0.5774, -0.4082, 1.2235, -1.1844, -0.2919, -1.3709, 0.4544,
  34188. 0.1256],
  34189. [ 0.5726, -0.4795, 1.7788, -0.0149, -0.5711, -0.0688, 0.5028,
  34190. -0.0534],
  34191. [ 0.5932, -0.3962, 1.4554, -0.9233, -0.6404, -0.4922, 0.4912,
  34192. 0.1159],
  34193. [ 0.5637, -0.4143, 1.7519, -0.0787, -0.3055, -0.0149, 0.3758,
  34194. 0.3084],
  34195. [ 0.6520, -0.3623, 1.8885, 0.0313, -0.5538, -0.5384, 0.6926,
  34196. 0.1661],
  34197. [ 0.5915, -0.3682, 1.2187, -1.2313, -0.4326, -0.9541, 0.5778,
  34198. 0.3777],
  34199. [-2.2859, -2.2859, 1.1898, -1.2278, -0.5134, -0.8925, 0.5085,
  34200. 0.3315]]], device='cuda:0')
  34201. loss_train_step before backward: tensor(0.0100, device='cuda:0', grad_fn=<MseLossBackward>)
  34202. loss_train_step after backward: tensor(0.0100, device='cuda:0', grad_fn=<MseLossBackward>)
  34203. loss_train: 1.1936637787148356
  34204. step: 86
  34205. running loss: 0.013879811380405066
  34206.  
  34207. Train Steps: 86/90 Loss: 0.0139 torch.Size([8, 600, 800])
  34208. torch.Size([8, 8])
  34209. tensor([[0.6112, 0.4029, 0.8638, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567],
  34210. [0.6138, 0.4054, 0.8750, 0.4750, 0.4363, 0.5017, 0.5086, 0.5822],
  34211. [0.6202, 0.4066, 0.8398, 0.2648, 0.3925, 0.2627, 0.5845, 0.5124],
  34212. [0.6286, 0.4086, 0.8408, 0.2801, 0.4163, 0.2800, 0.6725, 0.5393],
  34213. [0.6199, 0.4015, 0.8716, 0.5228, 0.3833, 0.4772, 0.5883, 0.5148],
  34214. [0.6222, 0.3937, 0.8350, 0.5617, 0.4138, 0.4600, 0.5800, 0.5233],
  34215. [0.6245, 0.4100, 0.7762, 0.2583, 0.4963, 0.1517, 0.5875, 0.5417],
  34216. [ nan, nan, 0.8300, 0.3150, 0.3588, 0.3383, 0.5208, 0.5194]],
  34217. device='cuda:0', dtype=torch.float64)
  34218. predictions are: tensor([[ 0.6049, -0.3581, 1.7083, -0.1912, -0.1407, -0.0747, 0.4234, 0.3232],
  34219. [ 0.5529, -0.3694, 1.6717, -0.1886, -0.3450, -0.0744, 0.1994, 0.3777],
  34220. [ 0.5310, -0.4016, 1.4163, -1.1305, -0.4936, -1.0623, 0.4344, 0.1493],
  34221. [ 0.5927, -0.3553, 1.4686, -1.0565, -0.3416, -1.0536, 0.8740, 0.3268],
  34222. [ 0.6938, -0.3412, 1.7493, 0.0628, -0.5975, -0.1540, 0.5167, 0.1286],
  34223. [ 0.7393, -0.3116, 1.5618, 0.2102, -0.4959, -0.1859, 0.4538, 0.2352],
  34224. [ 0.6530, -0.2842, 1.1949, -1.1057, -0.1295, -1.5413, 0.4090, 0.3264],
  34225. [-2.0142, -2.1302, 1.3487, -0.9828, -0.6667, -0.8325, 0.3096, 0.2276]],
  34226. device='cuda:0', grad_fn=<AddmmBackward>)
  34227. landmarks are: tensor([[[ 0.5371, -0.4252, 1.7037, -0.0688, -0.0342, 0.0620, 0.3758,
  34228. 0.2853],
  34229. [ 0.5491, -0.4132, 1.7557, -0.0919, -0.2709, 0.0313, 0.0635,
  34230. 0.4032],
  34231. [ 0.5790, -0.4079, 1.5929, -1.0630, -0.4729, -1.0725, 0.4137,
  34232. 0.0807],
  34233. [ 0.6174, -0.3984, 1.5975, -0.9921, -0.3633, -0.9926, 0.8205,
  34234. 0.2050],
  34235. [ 0.5773, -0.4316, 1.7399, 0.1287, -0.5153, -0.0817, 0.4313,
  34236. 0.0919],
  34237. [ 0.5880, -0.4676, 1.5709, 0.3084, -0.3748, -0.1612, 0.3931,
  34238. 0.1313],
  34239. [ 0.5985, -0.3921, 1.2995, -1.0927, 0.0062, -1.5854, 0.4277,
  34240. 0.2160],
  34241. [-2.2859, -2.2859, 1.5478, -0.8309, -0.6289, -0.7232, 0.1198,
  34242. 0.1133]]], device='cuda:0')
  34243. loss_train_step before backward: tensor(0.0096, device='cuda:0', grad_fn=<MseLossBackward>)
  34244. loss_train_step after backward: tensor(0.0096, device='cuda:0', grad_fn=<MseLossBackward>)
  34245. loss_train: 1.2032914850860834
  34246. step: 87
  34247. running loss: 0.013830936610184866
  34248. Train Steps: 87/90 Loss: 0.0138 torch.Size([8, 600, 800])
  34249. torch.Size([8, 8])
  34250. tensor([[0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283],
  34251. [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395],
  34252. [0.6068, 0.3963, 0.8650, 0.4317, 0.4037, 0.5083, 0.5253, 0.4999],
  34253. [0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266],
  34254. [0.6216, 0.4099, 0.7225, 0.2033, 0.4188, 0.2217, 0.5975, 0.5283],
  34255. [0.6204, 0.4007, 0.7838, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167],
  34256. [0.6136, 0.4117, 0.8700, 0.5167, 0.4188, 0.5083, 0.5147, 0.5495],
  34257. [ nan, nan, 0.6900, 0.1917, 0.3937, 0.2367, 0.5240, 0.5246]],
  34258. device='cuda:0', dtype=torch.float64)
  34259. predictions are: tensor([[ 0.6805, -0.3348, 1.9624, -0.4664, -0.5938, -0.4975, 0.7282, 0.2224],
  34260. [ 0.6928, -0.3152, 1.1401, -1.1865, -0.4487, -1.0811, 0.5481, 0.2631],
  34261. [ 0.6312, -0.3960, 1.7950, -0.2478, -0.4948, 0.0599, 0.3920, 0.0744],
  34262. [ 0.6732, -0.2913, 1.1461, -1.0722, -0.3414, -1.0949, 0.5678, 0.6630],
  34263. [ 0.7776, -0.2525, 1.1160, -1.1318, -0.3865, -1.2380, 0.4634, 0.2548],
  34264. [ 0.6188, -0.3683, 1.3264, -1.0727, -0.2328, -1.4757, 0.4626, 0.1861],
  34265. [ 0.6825, -0.3208, 1.7779, 0.1908, -0.3806, 0.0440, 0.3096, 0.2934],
  34266. [-1.7795, -1.9782, 1.0829, -1.3242, -0.4301, -1.2074, 0.2961, 0.3201]],
  34267. device='cuda:0', grad_fn=<AddmmBackward>)
  34268. landmarks are: tensor([[[ 0.6073, -0.4058, 1.8885, -0.4999, -0.5942, -0.4768, 0.6413,
  34269. 0.1544],
  34270. [ 0.5713, -0.4538, 1.1028, -1.3659, -0.3831, -1.1273, 0.5340,
  34271. 0.2058],
  34272. [ 0.5168, -0.4556, 1.7095, -0.2921, -0.4210, 0.0620, 0.1404,
  34273. 0.0231],
  34274. [ 0.5746, -0.4021, 1.0801, -1.1312, -0.3229, -1.1081, 0.4803,
  34275. 0.6084],
  34276. [ 0.5850, -0.3925, 1.0513, -1.3467, -0.3517, -1.2620, 0.4739,
  34277. 0.1544],
  34278. [ 0.5796, -0.4351, 1.3342, -1.3159, -0.2189, -1.4853, 0.4046,
  34279. 0.1005],
  34280. [ 0.5482, -0.3841, 1.7326, 0.1005, -0.3517, 0.0620, 0.0912,
  34281. 0.2522],
  34282. [-2.2859, -2.2859, 0.9012, -1.4006, -0.4672, -1.1928, 0.1342,
  34283. 0.1373]]], device='cuda:0')
  34284. loss_train_step before backward: tensor(0.0152, device='cuda:0', grad_fn=<MseLossBackward>)
  34285. loss_train_step after backward: tensor(0.0152, device='cuda:0', grad_fn=<MseLossBackward>)
  34286. loss_train: 1.2185192555189133
  34287. step: 88
  34288. running loss: 0.013846809721805832
  34289. Train Steps: 88/90 Loss: 0.0138 torch.Size([8, 600, 800])
  34290. torch.Size([8, 8])
  34291. tensor([[0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240],
  34292. [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817],
  34293. [0.6305, 0.3983, 0.8950, 0.4833, 0.3688, 0.4683, 0.6375, 0.5117],
  34294. [0.6346, 0.4144, 0.9088, 0.4667, 0.3850, 0.4333, 0.7121, 0.5899],
  34295. [0.6200, 0.4118, 0.8287, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799],
  34296. [0.6175, 0.4093, 0.8800, 0.4433, 0.4075, 0.4367, 0.5128, 0.4901],
  34297. [0.6214, 0.4112, 0.7838, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083],
  34298. [0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266]],
  34299. device='cuda:0', dtype=torch.float64)
  34300. predictions are: tensor([[ 0.6163, -0.3500, 1.1294, -1.2303, -0.3170, -1.3979, 0.2945, 0.2448],
  34301. [ 0.6485, -0.4020, 1.5849, 0.4244, -0.2580, 0.0700, 0.1429, 0.1287],
  34302. [ 0.7184, -0.3741, 1.8447, -0.0733, -0.5955, -0.0847, 0.7442, 0.2117],
  34303. [ 0.6278, -0.3934, 1.9591, -0.1562, -0.4049, -0.2477, 1.0511, 0.4199],
  34304. [ 0.6740, -0.3122, 1.5357, -0.4706, -0.5488, -0.9307, 0.1926, 0.4752],
  34305. [ 0.6191, -0.4002, 1.8453, -0.2578, -0.4682, -0.1923, 0.1952, 0.1291],
  34306. [ 0.6972, -0.3326, 1.3606, -1.3241, -0.5866, -0.8800, 0.4947, 0.1693],
  34307. [ 0.6533, -0.3122, 1.0827, -1.2065, -0.3026, -1.0755, 0.5253, 0.7117]],
  34308. device='cuda:0', grad_fn=<AddmmBackward>)
  34309. landmarks are: tensor([[[ 0.5676, -0.4112, 1.1898, -1.2467, -0.2940, -1.4622, 0.2103,
  34310. 0.1343],
  34311. [ 0.5193, -0.4699, 1.5767, 0.4008, -0.2420, 0.0774, 0.1178,
  34312. -0.0610],
  34313. [ 0.6262, -0.4461, 1.8480, -0.0534, -0.5827, -0.1227, 0.6587,
  34314. 0.0774],
  34315. [ 0.6454, -0.3719, 1.9115, -0.1304, -0.5076, -0.2844, 1.0033,
  34316. 0.4386],
  34317. [ 0.5779, -0.3840, 1.5420, -0.4306, -0.5423, -0.9772, 0.2041,
  34318. 0.3928],
  34319. [ 0.5663, -0.3955, 1.7788, -0.2382, -0.4037, -0.2690, 0.0828,
  34320. -0.0220],
  34321. [ 0.5845, -0.3864, 1.3342, -1.3082, -0.6000, -0.8386, 0.3353,
  34322. 0.0620],
  34323. [ 0.5746, -0.4021, 1.0801, -1.1312, -0.3229, -1.1081, 0.4803,
  34324. 0.6084]]], device='cuda:0')
  34325. loss_train_step before backward: tensor(0.0050, device='cuda:0', grad_fn=<MseLossBackward>)
  34326. loss_train_step after backward: tensor(0.0050, device='cuda:0', grad_fn=<MseLossBackward>)
  34327. loss_train: 1.223516573663801
  34328. step: 89
  34329. running loss: 0.01374737723217754
  34330. Train Steps: 89/90 Loss: 0.0137 torch.Size([8, 600, 800])
  34331. torch.Size([8, 8])
  34332. tensor([[0.6200, 0.4059, 0.8700, 0.4900, 0.4163, 0.5000, 0.6162, 0.5467],
  34333. [0.6228, 0.4004, 0.8750, 0.5250, 0.3825, 0.5233, 0.6362, 0.5000],
  34334. [0.6198, 0.4115, 0.7762, 0.2717, 0.3713, 0.3200, 0.5837, 0.5683],
  34335. [0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804],
  34336. [0.6151, 0.4085, 0.8750, 0.4367, 0.3887, 0.4367, 0.5066, 0.5846],
  34337. [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398],
  34338. [0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285],
  34339. [0.6271, 0.4040, 0.9138, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413]],
  34340. device='cuda:0', dtype=torch.float64)
  34341. predictions are: tensor([[ 0.6415, -0.3877, 1.6808, -0.0775, -0.2826, 0.0688, 0.3160, 0.3622],
  34342. [ 0.6123, -0.4358, 1.6953, 0.1156, -0.4587, 0.1190, 0.5347, 0.1479],
  34343. [ 0.6549, -0.3544, 1.2989, -1.1406, -0.5752, -0.8567, 0.2800, 0.4329],
  34344. [ 0.7401, -0.3168, 1.7433, -0.5885, -0.5935, -0.5729, 0.5201, 0.0110],
  34345. [ 0.6757, -0.2947, 1.5974, -0.2606, -0.4664, -0.2819, -0.0857, 0.5438],
  34346. [ 0.6677, -0.3903, 1.5811, 0.2361, -0.5221, -0.2246, 0.3248, 0.2999],
  34347. [ 0.6235, -0.3914, 1.6939, -0.7000, -0.5345, -0.1516, 0.8339, 0.3037],
  34348. [ 0.6724, -0.3755, 1.9923, -0.5858, -0.0867, -1.1977, 0.9983, 0.3048]],
  34349. device='cuda:0', grad_fn=<AddmmBackward>)
  34350. landmarks are: tensor([[[ 0.5776, -0.4109, 1.7326, -0.0226, -0.3633, 0.0236, 0.5605,
  34351. 0.2391],
  34352. [ 0.5908, -0.4366, 1.7557, 0.1390, -0.5192, 0.1313, 0.6529,
  34353. 0.0236],
  34354. [ 0.5768, -0.3852, 1.2995, -1.0311, -0.5711, -0.8079, 0.4104,
  34355. 0.3392],
  34356. [ 0.6071, -0.4119, 1.7788, -0.5153, -0.6000, -0.5692, 0.6586,
  34357. -0.0670],
  34358. [ 0.5553, -0.3992, 1.7557, -0.2690, -0.4903, -0.2690, 0.0542,
  34359. 0.4145],
  34360. [ 0.5786, -0.4463, 1.6655, 0.2216, -0.5115, -0.2675, 0.4236,
  34361. 0.2075],
  34362. [ 0.6108, -0.4273, 1.8711, -0.6616, -0.5769, -0.1997, 0.9156,
  34363. 0.1554],
  34364. [ 0.6108, -0.4201, 1.9346, -0.5538, -0.1497, -1.0773, 1.0545,
  34365. 0.2142]]], device='cuda:0')
  34366. loss_train_step before backward: tensor(0.0074, device='cuda:0', grad_fn=<MseLossBackward>)
  34367. loss_train_step after backward: tensor(0.0074, device='cuda:0', grad_fn=<MseLossBackward>)
  34368. loss_train: 1.230874984525144
  34369. step: 90
  34370. running loss: 0.013676388716946045
  34371.  
  34372. Valid Steps: 10/10 Loss: nan 37
  34373. --------------------------------------------------
  34374. Epoch: 9 Train Loss: 0.0137 Valid Loss: nan
  34375. --------------------------------------------------
  34376. size of train loader is: 90
  34377. torch.Size([8, 600, 800])
  34378. torch.Size([8, 8])
  34379. tensor([[0.6321, 0.4048, 0.8738, 0.5617, 0.3875, 0.4417, 0.6361, 0.4927],
  34380. [0.6259, 0.4156, 0.8812, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960],
  34381. [0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689],
  34382. [0.6087, 0.3976, 0.8337, 0.3867, 0.3713, 0.3117, 0.5938, 0.5300],
  34383. [0.6200, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183],
  34384. [0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862],
  34385. [0.6201, 0.4004, 0.8786, 0.5043, 0.3833, 0.5138, 0.5997, 0.5092],
  34386. [0.6198, 0.4164, 0.8700, 0.5067, 0.4625, 0.5650, 0.5464, 0.5197]],
  34387. device='cuda:0', dtype=torch.float64)
  34388. predictions are: tensor([[ 0.6679, -0.4081, 1.7148, 0.2545, -0.5189, -0.2133, 0.6253, 0.0836],
  34389. [ 0.6459, -0.3418, 1.8009, -0.9554, -0.0123, -1.4325, 0.6081, 0.2512],
  34390. [ 0.5468, -0.3994, 1.2339, -0.7346, -0.6692, -0.4572, 0.1664, 0.4841],
  34391. [ 0.6728, -0.3519, 1.5303, -0.6012, -0.5453, -0.7852, 0.4589, 0.3779],
  34392. [ 0.6847, -0.3457, 1.3959, -1.0181, -0.4633, -0.9420, 0.4814, 0.2562],
  34393. [ 0.7260, -0.3154, 1.8198, -0.4252, -0.6058, -0.1752, 0.6467, 0.0982],
  34394. [ 0.5705, -0.4656, 1.7108, 0.0119, -0.4905, 0.1314, 0.3809, 0.2373],
  34395. [ 0.5818, -0.3801, 1.7161, -0.0282, -0.1464, 0.2088, 0.2626, 0.3285]],
  34396. device='cuda:0', grad_fn=<AddmmBackward>)
  34397. landmarks are: tensor([[[ 0.6335, -0.4162, 1.7499, 0.3084, -0.4961, -0.2459, 0.6524,
  34398. -0.0102],
  34399. [ 0.6052, -0.3663, 1.7845, -0.8156, -0.0804, -1.4237, 0.5866,
  34400. 0.0051],
  34401. [ 0.5327, -0.4381, 1.2880, -0.6308, -0.6866, -0.5307, 0.2658,
  34402. 0.3417],
  34403. [ 0.5255, -0.4495, 1.5651, -0.4999, -0.5711, -0.8463, 0.4566,
  34404. 0.1621],
  34405. [ 0.5781, -0.4129, 1.4142, -0.9057, -0.5115, -0.9937, 0.4621,
  34406. 0.1080],
  34407. [ 0.6072, -0.4075, 1.8942, -0.3537, -0.5423, -0.1612, 0.6277,
  34408. -0.0400],
  34409. [ 0.5783, -0.4363, 1.7724, 0.0432, -0.5153, 0.0871, 0.4840,
  34410. 0.0663],
  34411. [ 0.5770, -0.3624, 1.7326, 0.0543, -0.1497, 0.3238, 0.2378,
  34412. 0.1146]]], device='cuda:0')
  34413. loss_train_step before backward: tensor(0.0075, device='cuda:0', grad_fn=<MseLossBackward>)
  34414. loss_train_step after backward: tensor(0.0075, device='cuda:0', grad_fn=<MseLossBackward>)
  34415. loss_train: 0.0074615185149014
  34416. step: 1
  34417. running loss: 0.0074615185149014
  34418. Train Steps: 1/90 Loss: 0.0075 torch.Size([8, 600, 800])
  34419. torch.Size([8, 8])
  34420. tensor([[0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5988, 0.5667],
  34421. [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320],
  34422. [0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038],
  34423. [0.6339, 0.4118, 0.7988, 0.5800, 0.3912, 0.4583, 0.7343, 0.5760],
  34424. [0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407],
  34425. [0.6280, 0.4101, 0.9050, 0.4533, 0.3775, 0.3217, 0.6338, 0.4915],
  34426. [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329],
  34427. [0.6263, 0.4065, 0.9038, 0.4317, 0.3588, 0.4550, 0.6325, 0.5250]],
  34428. device='cuda:0', dtype=torch.float64)
  34429. predictions are: tensor([[ 0.6624, -0.3220, 1.6149, -0.7544, -0.4754, -0.6492, 0.2969, 0.4028],
  34430. [ 0.6560, -0.3936, 1.6232, 0.0781, -0.4228, 0.0890, 0.8605, 0.1984],
  34431. [ 0.6372, -0.3622, 1.6846, -0.1787, -0.1467, 0.0819, 0.0969, 0.0911],
  34432. [ 0.5155, -0.4592, 1.3896, 0.2184, -0.4286, -0.1787, 0.9320, 0.3922],
  34433. [ 0.6200, -0.3272, 1.5840, -0.5143, -0.5560, -0.8308, -0.1218, 0.2780],
  34434. [ 0.5761, -0.4320, 1.9143, -0.2647, -0.5046, -0.7932, 0.4572, 0.0386],
  34435. [ 0.6207, -0.3765, 1.6158, -0.8149, -0.6073, -0.5826, 0.3833, 0.2445],
  34436. [ 0.4286, -0.5189, 1.9060, -0.4003, -0.5482, -0.1925, 0.5965, 0.2424]],
  34437. device='cuda:0', grad_fn=<AddmmBackward>)
  34438. landmarks are: tensor([[[ 0.5783, -0.3933, 1.6748, -0.6154, -0.5769, -0.6462, 0.4797,
  34439. 0.3315],
  34440. [ 0.6273, -0.4393, 1.6402, 0.1313, -0.5076, 0.0467, 1.1532,
  34441. 0.1715],
  34442. [ 0.5249, -0.4473, 1.7326, -0.0919, -0.2016, 0.1544, 0.1733,
  34443. 0.0412],
  34444. [ 0.6421, -0.3840, 1.4035, 0.3931, -0.4788, -0.1689, 1.1057,
  34445. 0.3745],
  34446. [ 0.5470, -0.3952, 1.6377, -0.4253, -0.6289, -0.8079, 0.0249,
  34447. 0.2116],
  34448. [ 0.6148, -0.3918, 1.8942, -0.1920, -0.5423, -0.8002, 0.6414,
  34449. -0.0156],
  34450. [ 0.5680, -0.4393, 1.5920, -0.6672, -0.6453, -0.5457, 0.5149,
  34451. 0.1753],
  34452. [ 0.6070, -0.4085, 1.8885, -0.2921, -0.6289, -0.1843, 0.6356,
  34453. 0.1390]]], device='cuda:0')
  34454. loss_train_step before backward: tensor(0.0085, device='cuda:0', grad_fn=<MseLossBackward>)
  34455. loss_train_step after backward: tensor(0.0085, device='cuda:0', grad_fn=<MseLossBackward>)
  34456. loss_train: 0.015985853504389524
  34457. step: 2
  34458. running loss: 0.007992926752194762
  34459. Train Steps: 2/90 Loss: 0.0080 torch.Size([8, 600, 800])
  34460. torch.Size([8, 8])
  34461. tensor([[0.6274, 0.4087, 0.8375, 0.5700, 0.4025, 0.4800, 0.5700, 0.6117],
  34462. [0.6185, 0.4067, 0.8838, 0.4450, 0.4037, 0.4733, 0.5213, 0.5142],
  34463. [0.6214, 0.4112, 0.7838, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083],
  34464. [0.6166, 0.4008, 0.8563, 0.5667, 0.4388, 0.4933, 0.5575, 0.5567],
  34465. [0.6250, 0.4103, 0.8950, 0.4400, 0.3912, 0.5650, 0.6050, 0.5133],
  34466. [0.6076, 0.3953, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954],
  34467. [0.6170, 0.4102, 0.7468, 0.3695, 0.3463, 0.3767, 0.5238, 0.5823],
  34468. [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167]],
  34469. device='cuda:0', dtype=torch.float64)
  34470. predictions are: tensor([[ 0.6501, -0.3832, 1.6167, 0.3688, -0.4298, -0.0988, 0.5013, 0.4481],
  34471. [ 0.5538, -0.4239, 1.8637, -0.2437, -0.4463, -0.1293, 0.2182, 0.0533],
  34472. [ 0.6812, -0.3732, 1.5055, -1.3821, -0.6032, -0.9587, 0.6223, -0.0342],
  34473. [ 0.5355, -0.4735, 1.7399, 0.3028, -0.2351, 0.0205, 0.4293, 0.2091],
  34474. [ 0.6038, -0.3897, 1.9341, -0.3007, -0.4164, 0.3271, 0.8311, 0.0847],
  34475. [ 0.4763, -0.4982, 1.6687, -0.5504, -0.7019, -0.4592, 0.3585, -0.0793],
  34476. [ 0.5035, -0.3993, 1.2370, -0.6837, -0.6868, -0.5294, 0.2725, 0.3857],
  34477. [ 0.6356, -0.3197, 1.2649, -0.7366, -0.0193, -1.4324, 0.3899, 0.5164]],
  34478. device='cuda:0', grad_fn=<AddmmBackward>)
  34479. landmarks are: tensor([[[ 0.6118, -0.3983, 1.5824, 0.3469, -0.4268, -0.0688, 0.3469,
  34480. 0.5393],
  34481. [ 0.5708, -0.4075, 1.7961, -0.2305, -0.4210, -0.0996, 0.1219,
  34482. 0.0893],
  34483. [ 0.5845, -0.3864, 1.3342, -1.3082, -0.6000, -0.8386, 0.3353,
  34484. 0.0620],
  34485. [ 0.5620, -0.4346, 1.6691, 0.3315, -0.2594, -0.0072, 0.2891,
  34486. 0.2853],
  34487. [ 0.6010, -0.3909, 1.8480, -0.2536, -0.4788, 0.3238, 0.5085,
  34488. 0.0851],
  34489. [ 0.5206, -0.4603, 1.4670, -0.4768, -0.6577, -0.3998, 0.1836,
  34490. 0.0021],
  34491. [ 0.5639, -0.3911, 1.1634, -0.5794, -0.6866, -0.5461, 0.1334,
  34492. 0.4036],
  34493. [ 0.6161, -0.3075, 1.1678, -0.6500, 0.0813, -1.4006, 0.2545,
  34494. 0.5624]]], device='cuda:0')
  34495. loss_train_step before backward: tensor(0.0084, device='cuda:0', grad_fn=<MseLossBackward>)
  34496. loss_train_step after backward: tensor(0.0084, device='cuda:0', grad_fn=<MseLossBackward>)
  34497. loss_train: 0.02438948256894946
  34498. step: 3
  34499. running loss: 0.008129827522983154
  34500. Train Steps: 3/90 Loss: 0.0081 torch.Size([8, 600, 800])
  34501. torch.Size([8, 8])
  34502. tensor([[0.6250, 0.4110, 0.7238, 0.2067, 0.4263, 0.1883, 0.5625, 0.5633],
  34503. [0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136],
  34504. [0.6221, 0.4107, 0.7788, 0.3033, 0.3950, 0.2817, 0.6075, 0.5517],
  34505. [0.6250, 0.4054, 0.8770, 0.4723, 0.4662, 0.5367, 0.6162, 0.5433],
  34506. [0.6262, 0.4052, 0.8888, 0.4700, 0.3675, 0.5117, 0.6350, 0.5233],
  34507. [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609],
  34508. [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927],
  34509. [ nan, nan, 0.7335, 0.2569, 0.3788, 0.2667, 0.5066, 0.5578]],
  34510. device='cuda:0', dtype=torch.float64)
  34511. predictions are: tensor([[ 5.5326e-01, -4.2534e-01, 1.0864e+00, -1.1242e+00, -2.1720e-01,
  34512. -1.4227e+00, 3.4412e-01, 1.9398e-01],
  34513. [ 4.9047e-01, -4.4277e-01, 1.7040e+00, -3.5306e-01, -6.0177e-01,
  34514. -3.8281e-01, 2.0634e-01, 6.6394e-03],
  34515. [ 4.8487e-01, -4.6687e-01, 1.4244e+00, -8.3061e-01, -4.0475e-01,
  34516. -9.4819e-01, 4.2932e-01, 2.4966e-01],
  34517. [ 6.4183e-01, -3.5516e-01, 1.7091e+00, -1.8088e-02, -1.4074e-01,
  34518. 2.4657e-01, 5.6285e-01, 1.6769e-01],
  34519. [ 6.6839e-01, -3.6778e-01, 1.8622e+00, -2.1339e-03, -5.3337e-01,
  34520. 1.3466e-01, 7.2328e-01, 5.1827e-02],
  34521. [ 5.6919e-01, -3.9543e-01, 1.7194e+00, -7.2178e-01, -3.5065e-01,
  34522. -8.3430e-01, 7.7734e-01, 2.2063e-01],
  34523. [ 6.5032e-01, -3.9691e-01, 1.7636e+00, -2.2513e-01, -6.1440e-01,
  34524. -2.3528e-01, 4.7430e-01, 3.8070e-02],
  34525. [-2.1769e+00, -2.2064e+00, 1.1524e+00, -1.1870e+00, -5.2893e-01,
  34526. -9.4108e-01, 8.8347e-02, 2.0739e-01]], device='cuda:0',
  34527. grad_fn=<AddmmBackward>)
  34528. landmarks are: tensor([[[ 0.6010, -0.3875, 1.0570, -1.3313, -0.3171, -1.4160, 0.3122,
  34529. 0.3161],
  34530. [ 0.5235, -0.4273, 1.7499, -0.4306, -0.5827, -0.4614, 0.1651,
  34531. 0.0862],
  34532. [ 0.5875, -0.3888, 1.3111, -0.8848, -0.4614, -0.9849, 0.5201,
  34533. 0.2622],
  34534. [ 0.6009, -0.4135, 1.7651, -0.1043, -0.1323, 0.1929, 0.5605,
  34535. 0.2237],
  34536. [ 0.6063, -0.4142, 1.8192, -0.1150, -0.5885, 0.0774, 0.6471,
  34537. 0.1313],
  34538. [ 0.6026, -0.4032, 1.7326, -0.7771, -0.3691, -0.8617, 0.9704,
  34539. 0.3050],
  34540. [ 0.6010, -0.4345, 1.8480, -0.3537, -0.6462, -0.2613, 0.6524,
  34541. -0.0099],
  34542. [-2.2859, -2.2859, 1.1020, -1.0994, -0.5365, -1.0542, 0.0542,
  34543. 0.2905]]], device='cuda:0')
  34544. loss_train_step before backward: tensor(0.0057, device='cuda:0', grad_fn=<MseLossBackward>)
  34545. loss_train_step after backward: tensor(0.0057, device='cuda:0', grad_fn=<MseLossBackward>)
  34546. loss_train: 0.030109287705272436
  34547. step: 4
  34548. running loss: 0.007527321926318109
  34549.  
  34550. Train Steps: 4/90 Loss: 0.0075 torch.Size([8, 600, 800])
  34551. torch.Size([8, 8])
  34552. tensor([[0.6250, 0.4116, 0.8700, 0.4850, 0.4650, 0.5567, 0.6088, 0.5183],
  34553. [0.6259, 0.4156, 0.8812, 0.3183, 0.4775, 0.1867, 0.6219, 0.4960],
  34554. [0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733],
  34555. [0.6175, 0.3957, 0.8700, 0.4817, 0.4662, 0.5133, 0.5800, 0.5517],
  34556. [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600],
  34557. [0.6151, 0.4058, 0.7068, 0.2680, 0.3400, 0.4083, 0.5775, 0.5733],
  34558. [0.6097, 0.4000, 0.7325, 0.2667, 0.3450, 0.3517, 0.5284, 0.5045],
  34559. [0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433]],
  34560. device='cuda:0', dtype=torch.float64)
  34561. predictions are: tensor([[ 0.5593, -0.4216, 1.8539, 0.0233, -0.2211, 0.3257, 0.6745, 0.1137],
  34562. [ 0.5730, -0.4056, 1.9641, -0.6987, -0.1147, -1.4240, 0.6490, -0.0203],
  34563. [ 0.4865, -0.4275, 1.9484, 0.1791, -0.6681, -0.2434, 0.3987, 0.2926],
  34564. [ 0.6164, -0.4252, 1.7836, 0.0594, -0.1791, 0.0937, 0.4037, 0.1212],
  34565. [ 0.5586, -0.4110, 1.1464, -1.0049, -0.6523, -0.8031, 0.6228, 0.1723],
  34566. [ 0.4328, -0.5101, 1.1184, -0.9284, -0.7571, -0.4165, 0.4390, 0.2188],
  34567. [ 0.4403, -0.5205, 1.1551, -0.9474, -0.7202, -0.6720, 0.2133, -0.0217],
  34568. [ 0.5382, -0.4574, 1.7547, 0.2606, -0.0796, 0.0177, 0.1632, 0.2114]],
  34569. device='cuda:0', grad_fn=<AddmmBackward>)
  34570. landmarks are: tensor([[[ 0.6012, -0.3846, 1.7326, -0.0457, -0.1381, 0.2853, 0.5259,
  34571. 0.1082],
  34572. [ 0.6052, -0.3663, 1.7845, -0.8156, -0.0804, -1.4237, 0.5866,
  34573. 0.0051],
  34574. [ 0.6121, -0.3138, 1.8423, 0.0082, -0.6462, -0.3075, 0.3469,
  34575. 0.3623],
  34576. [ 0.5662, -0.4581, 1.7326, -0.0611, -0.1323, 0.0851, 0.3931,
  34577. 0.2622],
  34578. [ 0.5466, -0.4080, 1.0668, -1.1764, -0.6289, -0.7617, 0.4855,
  34579. 0.3007],
  34580. [ 0.5552, -0.4113, 0.9790, -1.0480, -0.7155, -0.3998, 0.3815,
  34581. 0.3623],
  34582. [ 0.5303, -0.4384, 1.0975, -1.0542, -0.6924, -0.6616, 0.1548,
  34583. 0.0442],
  34584. [ 0.5295, -0.4373, 1.6553, 0.1011, 0.0380, 0.0671, 0.0813,
  34585. 0.2237]]], device='cuda:0')
  34586. loss_train_step before backward: tensor(0.0076, device='cuda:0', grad_fn=<MseLossBackward>)
  34587. loss_train_step after backward: tensor(0.0076, device='cuda:0', grad_fn=<MseLossBackward>)
  34588. loss_train: 0.037714792881160975
  34589. step: 5
  34590. running loss: 0.0075429585762321946
  34591. Train Steps: 5/90 Loss: 0.0075 torch.Size([8, 600, 800])
  34592. torch.Size([8, 8])
  34593. tensor([[0.6250, 0.4013, 0.8525, 0.5417, 0.4037, 0.5117, 0.6325, 0.5017],
  34594. [0.6098, 0.3991, 0.8638, 0.4717, 0.4263, 0.4967, 0.5212, 0.5650],
  34595. [0.6197, 0.4050, 0.7527, 0.2000, 0.4042, 0.2249, 0.5895, 0.4995],
  34596. [0.6283, 0.4283, 0.7477, 0.3542, 0.5125, 0.1917, 0.5500, 0.6167],
  34597. [0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343],
  34598. [0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6138, 0.5450],
  34599. [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356],
  34600. [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973]],
  34601. device='cuda:0', dtype=torch.float64)
  34602. predictions are: tensor([[ 0.4132, -0.5643, 1.6594, 0.1873, -0.5100, 0.1605, 0.4882, 0.0102],
  34603. [ 0.3890, -0.5216, 1.6669, -0.0909, -0.3992, 0.1360, 0.0391, 0.1829],
  34604. [ 0.6699, -0.3643, 1.1973, -1.2907, -0.4870, -1.1349, 0.3494, -0.0741],
  34605. [ 0.5747, -0.3525, 1.1663, -0.6527, -0.0847, -1.2640, 0.1935, 0.5250],
  34606. [ 0.5285, -0.4959, 1.8223, -0.5301, -0.4036, -0.7936, 0.8845, 0.0212],
  34607. [ 0.5436, -0.4472, 1.8121, -0.1635, -0.4170, 0.2699, 0.4882, 0.1760],
  34608. [ 0.5868, -0.4045, 1.4909, -1.0393, -0.3746, -1.0620, 0.5821, 0.1467],
  34609. [ 0.5726, -0.4658, 1.8135, 0.1574, -0.6902, -0.0282, 0.5791, -0.0738]],
  34610. device='cuda:0', grad_fn=<AddmmBackward>)
  34611. landmarks are: tensor([[[ 0.6010, -0.4321, 1.6517, 0.2160, -0.4210, 0.0774, 0.6356,
  34612. 0.0313],
  34613. [ 0.5308, -0.4425, 1.7037, -0.1073, -0.3171, 0.0082, 0.1217,
  34614. 0.3238],
  34615. [ 0.5762, -0.4153, 1.1908, -1.3622, -0.4190, -1.2471, 0.4368,
  34616. 0.0213],
  34617. [ 0.6161, -0.3075, 1.1678, -0.6500, 0.0813, -1.4006, 0.2545,
  34618. 0.5624],
  34619. [ 0.6125, -0.4369, 1.9173, -0.5384, -0.2594, -0.8386, 0.9741,
  34620. 0.1821],
  34621. [ 0.5960, -0.4102, 1.8018, -0.1612, -0.3344, 0.1159, 0.5490,
  34622. 0.2314],
  34623. [ 0.6174, -0.4225, 1.4975, -1.1709, -0.3174, -1.1806, 0.6539,
  34624. 0.1879],
  34625. [ 0.6240, -0.4321, 1.8423, 0.1852, -0.5885, -0.1612, 0.6962,
  34626. 0.0111]]], device='cuda:0')
  34627. loss_train_step before backward: tensor(0.0076, device='cuda:0', grad_fn=<MseLossBackward>)
  34628. loss_train_step after backward: tensor(0.0076, device='cuda:0', grad_fn=<MseLossBackward>)
  34629. loss_train: 0.04535488644614816
  34630. step: 6
  34631. running loss: 0.0075591477410246926
  34632. Train Steps: 6/90 Loss: 0.0076 torch.Size([8, 600, 800])
  34633. torch.Size([8, 8])
  34634. tensor([[0.6207, 0.4081, 0.7662, 0.2067, 0.3962, 0.3200, 0.6312, 0.5300],
  34635. [0.6161, 0.4024, 0.8838, 0.4583, 0.3688, 0.3733, 0.5311, 0.5344],
  34636. [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550],
  34637. [0.6072, 0.4029, 0.7037, 0.2150, 0.3912, 0.2267, 0.5516, 0.5507],
  34638. [0.6228, 0.4004, 0.8750, 0.5250, 0.3825, 0.5233, 0.6362, 0.5000],
  34639. [0.6164, 0.4076, 0.8838, 0.4117, 0.3713, 0.5550, 0.6238, 0.5350],
  34640. [0.6128, 0.4084, 0.8738, 0.4683, 0.3613, 0.3700, 0.4960, 0.5397],
  34641. [0.6118, 0.4052, 0.8463, 0.3917, 0.3538, 0.3450, 0.5053, 0.5593]],
  34642. device='cuda:0', dtype=torch.float64)
  34643. predictions are: tensor([[ 0.5229, -0.4542, 1.2582, -1.1586, -0.4608, -0.8265, 0.7056, 0.1246],
  34644. [ 0.5421, -0.4597, 1.8666, -0.0591, -0.5739, -0.4832, 0.3031, 0.0289],
  34645. [ 0.4805, -0.4632, 1.7951, -0.0926, 0.0650, 0.0324, 0.5053, 0.2165],
  34646. [ 0.5597, -0.4463, 0.9369, -1.2581, -0.4177, -1.2128, 0.3655, 0.1525],
  34647. [ 0.5571, -0.4922, 1.7682, 0.2352, -0.4721, 0.1582, 0.8246, -0.0903],
  34648. [ 0.5490, -0.4201, 1.7635, -0.2910, -0.4697, 0.3081, 0.7349, 0.2057],
  34649. [ 0.5234, -0.4357, 1.7965, -0.0241, -0.6572, -0.5234, 0.0793, 0.1230],
  34650. [ 0.4955, -0.4340, 1.6476, -0.4130, -0.6548, -0.6279, -0.0118, 0.1236]],
  34651. device='cuda:0', grad_fn=<AddmmBackward>)
  34652. landmarks are: tensor([[[ 0.5809, -0.4011, 1.2533, -1.3313, -0.4557, -0.8079, 0.6298,
  34653. 0.1621],
  34654. [ 0.5598, -0.4273, 1.7961, -0.1689, -0.5827, -0.5615, 0.1671,
  34655. 0.1824],
  34656. [ 0.5351, -0.4321, 1.7326, -0.1381, 0.0463, 0.1049, 0.4162,
  34657. 0.2776],
  34658. [ 0.5185, -0.4252, 0.9647, -1.2928, -0.4788, -1.2390, 0.2617,
  34659. 0.2576],
  34660. [ 0.5908, -0.4366, 1.7557, 0.1390, -0.5192, 0.1313, 0.6529,
  34661. 0.0236],
  34662. [ 0.5614, -0.4032, 1.7961, -0.3844, -0.5711, 0.2776, 0.5952,
  34663. 0.1852],
  34664. [ 0.5443, -0.3994, 1.7499, -0.1227, -0.6173, -0.5769, 0.0049,
  34665. 0.2071],
  34666. [ 0.5399, -0.4142, 1.6229, -0.4768, -0.6520, -0.6924, 0.0481,
  34667. 0.2972]]], device='cuda:0')
  34668. loss_train_step before backward: tensor(0.0054, device='cuda:0', grad_fn=<MseLossBackward>)
  34669. loss_train_step after backward: tensor(0.0054, device='cuda:0', grad_fn=<MseLossBackward>)
  34670. loss_train: 0.050726964604109526
  34671. step: 7
  34672. running loss: 0.007246709229158503
  34673. Train Steps: 7/90 Loss: 0.0072 torch.Size([8, 600, 800])
  34674. torch.Size([8, 8])
  34675. tensor([[0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869],
  34676. [0.6102, 0.4020, 0.8638, 0.3717, 0.3625, 0.5017, 0.6038, 0.5500],
  34677. [0.6248, 0.4032, 0.7738, 0.1900, 0.4813, 0.1400, 0.5941, 0.4904],
  34678. [0.6277, 0.4118, 0.8988, 0.3833, 0.3950, 0.2650, 0.6290, 0.4938],
  34679. [0.6102, 0.4005, 0.8688, 0.5100, 0.4813, 0.5400, 0.5404, 0.5064],
  34680. [0.6200, 0.4071, 0.7338, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517],
  34681. [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841],
  34682. [0.6226, 0.4185, 0.8850, 0.5500, 0.3800, 0.4250, 0.5625, 0.5617]],
  34683. device='cuda:0', dtype=torch.float64)
  34684. predictions are: tensor([[ 0.5618, -0.3972, 1.6424, -0.7693, -0.2796, -1.2178, 0.3273, -0.0264],
  34685. [ 0.5328, -0.4261, 1.5671, -0.5516, -0.6503, 0.1277, 0.5681, 0.2026],
  34686. [ 0.5060, -0.4781, 1.2327, -1.2211, -0.0598, -1.4889, 0.3843, 0.0457],
  34687. [ 0.5032, -0.4759, 1.8289, -0.4374, -0.5492, -0.8765, 0.5361, -0.0238],
  34688. [ 0.4846, -0.4673, 1.5984, 0.0356, -0.1474, 0.2934, 0.0812, -0.0116],
  34689. [ 0.4216, -0.5137, 1.0399, -1.3717, -0.4076, -0.9645, 0.4648, 0.2284],
  34690. [ 0.5114, -0.4945, 1.7288, -0.2810, -0.4495, -0.4657, 1.0433, 0.2603],
  34691. [ 0.4677, -0.4175, 1.6918, 0.2672, -0.5914, -0.1755, 0.2433, 0.2968]],
  34692. device='cuda:0', grad_fn=<AddmmBackward>)
  34693. landmarks are: tensor([[[ 0.6119, -0.3927, 1.6979, -0.7925, -0.2536, -1.3698, 0.4282,
  34694. -0.0368],
  34695. [ 0.5324, -0.4293, 1.7037, -0.5692, -0.6115, 0.0313, 0.5028,
  34696. 0.2545],
  34697. [ 0.5999, -0.4236, 1.2880, -1.4083, -0.0630, -1.6393, 0.4584,
  34698. -0.0208],
  34699. [ 0.6135, -0.3841, 1.8654, -0.5153, -0.4614, -1.0619, 0.6195,
  34700. -0.0049],
  34701. [ 0.5328, -0.4361, 1.7268, 0.0697, -0.0630, 0.2083, 0.2103,
  34702. 0.0532],
  34703. [ 0.5779, -0.4054, 1.1032, -1.4006, -0.3460, -1.1543, 0.5547,
  34704. 0.2622],
  34705. [ 0.6438, -0.3936, 2.0039, -0.2690, -0.2940, -0.6231, 1.0618,
  34706. 0.4119],
  34707. [ 0.5897, -0.3527, 1.8018, 0.2545, -0.5307, -0.3229, 0.3122,
  34708. 0.3084]]], device='cuda:0')
  34709. loss_train_step before backward: tensor(0.0097, device='cuda:0', grad_fn=<MseLossBackward>)
  34710. loss_train_step after backward: tensor(0.0097, device='cuda:0', grad_fn=<MseLossBackward>)
  34711. loss_train: 0.06042189570143819
  34712. step: 8
  34713. running loss: 0.007552736962679774
  34714.  
  34715. Train Steps: 8/90 Loss: 0.0076 torch.Size([8, 600, 800])
  34716. torch.Size([8, 8])
  34717. tensor([[0.6060, 0.3924, 0.8450, 0.5717, 0.4200, 0.5217, 0.5253, 0.4752],
  34718. [0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183],
  34719. [0.6241, 0.4143, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550],
  34720. [0.6198, 0.4105, 0.7950, 0.3267, 0.3675, 0.3767, 0.6038, 0.6167],
  34721. [0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742],
  34722. [0.6293, 0.4024, 0.8750, 0.5000, 0.4012, 0.5733, 0.7121, 0.5633],
  34723. [0.6250, 0.4008, 0.8950, 0.4183, 0.3550, 0.4383, 0.6361, 0.4927],
  34724. [0.6307, 0.4029, 0.8988, 0.4817, 0.3937, 0.3500, 0.7311, 0.5378]],
  34725. device='cuda:0', dtype=torch.float64)
  34726. predictions are: tensor([[ 0.5586, -0.4238, 1.5115, 0.3208, -0.2791, 0.0410, 0.1167, -0.0283],
  34727. [ 0.5033, -0.4404, 1.2076, -0.9702, -0.4188, -0.7943, 0.3816, 0.4679],
  34728. [ 0.4406, -0.4708, 1.8064, -0.1864, -0.3350, 0.1449, 0.4002, 0.2132],
  34729. [ 0.3537, -0.5212, 1.3906, -0.8135, -0.5936, -0.4937, 0.3015, 0.4278],
  34730. [ 0.6171, -0.4108, 1.7262, -0.7019, -0.3729, -1.1624, 0.4567, -0.1241],
  34731. [ 0.6642, -0.3580, 1.7266, 0.0246, -0.3493, 0.3263, 0.7752, 0.1696],
  34732. [ 0.5559, -0.4502, 1.7533, -0.3565, -0.6303, -0.3345, 0.4073, 0.0063],
  34733. [ 0.5353, -0.4811, 1.8886, -0.0844, -0.4030, -0.6989, 0.9185, 0.0296]],
  34734. device='cuda:0', grad_fn=<AddmmBackward>)
  34735. landmarks are: tensor([[[ 0.5132, -0.4736, 1.6171, 0.3546, -0.3460, 0.1236, 0.1404,
  34736. -0.0911],
  34737. [ 0.5759, -0.3948, 1.2072, -0.9849, -0.4672, -0.8156, 0.4912,
  34738. 0.5701],
  34739. [ 0.5968, -0.3725, 1.8423, -0.1381, -0.4037, 0.1852, 0.6009,
  34740. 0.2776],
  34741. [ 0.5768, -0.3899, 1.3861, -0.7771, -0.5885, -0.5461, 0.5028,
  34742. 0.5624],
  34743. [ 0.6134, -0.4218, 1.7268, -0.6154, -0.4730, -1.0850, 0.5463,
  34744. -0.0957],
  34745. [ 0.6207, -0.4273, 1.7557, 0.0236, -0.4326, 0.3623, 1.0033,
  34746. 0.3157],
  34747. [ 0.6010, -0.4345, 1.8480, -0.3537, -0.6462, -0.2613, 0.6524,
  34748. -0.0099],
  34749. [ 0.6273, -0.4249, 1.8654, -0.0611, -0.4672, -0.6693, 1.0910,
  34750. 0.1982]]], device='cuda:0')
  34751. loss_train_step before backward: tensor(0.0087, device='cuda:0', grad_fn=<MseLossBackward>)
  34752. loss_train_step after backward: tensor(0.0087, device='cuda:0', grad_fn=<MseLossBackward>)
  34753. loss_train: 0.06914769345894456
  34754. step: 9
  34755. running loss: 0.00768307705099384
  34756. Train Steps: 9/90 Loss: 0.0077 torch.Size([8, 600, 800])
  34757. torch.Size([8, 8])
  34758. tensor([[0.6329, 0.4196, 0.9238, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748],
  34759. [0.6179, 0.4082, 0.6688, 0.2667, 0.3588, 0.3317, 0.5750, 0.5783],
  34760. [0.6277, 0.4057, 0.8300, 0.2650, 0.4363, 0.1850, 0.6140, 0.4823],
  34761. [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600],
  34762. [0.6173, 0.4114, 0.7325, 0.2500, 0.4213, 0.1917, 0.5338, 0.5700],
  34763. [0.6325, 0.4165, 0.9000, 0.4617, 0.3813, 0.4900, 0.7485, 0.5447],
  34764. [0.6080, 0.4010, 0.8750, 0.4500, 0.4825, 0.5617, 0.5837, 0.5583],
  34765. [0.6201, 0.4116, 0.8725, 0.4733, 0.3700, 0.3217, 0.5386, 0.5767]],
  34766. device='cuda:0', dtype=torch.float64)
  34767. predictions are: tensor([[ 0.6758, -0.3886, 2.0155, -0.1198, -0.3333, -0.9112, 1.0076, 0.2153],
  34768. [ 0.4710, -0.4752, 0.8565, -1.0481, -0.6728, -0.7240, 0.1994, 0.2647],
  34769. [ 0.6471, -0.4175, 1.6222, -1.1371, -0.1707, -1.3694, 0.5561, -0.1339],
  34770. [ 0.5243, -0.4422, 1.0698, -1.1698, -0.5872, -0.7628, 0.4584, 0.1919],
  34771. [ 0.2242, -0.6373, 1.0507, -1.1499, -0.3268, -1.2658, 0.1022, 0.2798],
  34772. [ 0.6209, -0.4149, 1.9634, -0.0922, -0.5583, 0.1483, 1.0144, 0.0745],
  34773. [ 0.4827, -0.4694, 1.8218, -0.1977, -0.0812, 0.2834, 0.3023, 0.1277],
  34774. [ 0.4675, -0.4487, 1.7748, -0.0849, -0.5758, -0.7073, 0.1284, 0.2613]],
  34775. device='cuda:0', grad_fn=<AddmmBackward>)
  34776. landmarks are: tensor([[[ 0.6372, -0.3479, 1.9808, -0.1689, -0.3171, -0.9310, 1.1057,
  34777. 0.3692],
  34778. [ 0.5680, -0.4005, 0.8030, -1.0542, -0.6289, -0.7540, 0.3700,
  34779. 0.3854],
  34780. [ 0.6132, -0.4122, 1.5478, -1.0619, -0.2709, -1.4314, 0.5500,
  34781. -0.0583],
  34782. [ 0.5466, -0.4080, 1.0668, -1.1764, -0.6289, -0.7617, 0.4855,
  34783. 0.3007],
  34784. [ 0.5652, -0.3858, 1.0975, -1.1312, -0.3402, -1.4006, 0.1794,
  34785. 0.3469],
  34786. [ 0.6355, -0.3623, 1.8711, -0.1535, -0.5249, -0.0226, 1.1715,
  34787. 0.2302],
  34788. [ 0.5223, -0.4336, 1.7557, -0.2074, -0.0573, 0.3084, 0.4104,
  34789. 0.2930],
  34790. [ 0.5781, -0.3848, 1.7441, -0.0996, -0.5769, -0.8002, 0.2021,
  34791. 0.3778]]], device='cuda:0')
  34792. loss_train_step before backward: tensor(0.0086, device='cuda:0', grad_fn=<MseLossBackward>)
  34793. loss_train_step after backward: tensor(0.0086, device='cuda:0', grad_fn=<MseLossBackward>)
  34794. loss_train: 0.07775980466976762
  34795. step: 10
  34796. running loss: 0.007775980466976762
  34797. Train Steps: 10/90 Loss: 0.0078 torch.Size([8, 600, 800])
  34798. torch.Size([8, 8])
  34799. tensor([[0.6263, 0.4029, 0.8900, 0.4933, 0.4075, 0.5183, 0.6406, 0.4758],
  34800. [0.6205, 0.4081, 0.8950, 0.4017, 0.3788, 0.4700, 0.5963, 0.5667],
  34801. [ nan, nan, 0.6900, 0.1917, 0.3937, 0.2367, 0.5240, 0.5246],
  34802. [0.6192, 0.4128, 0.8513, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633],
  34803. [0.6201, 0.4027, 0.8029, 0.2728, 0.4042, 0.2310, 0.5980, 0.5391],
  34804. [ nan, nan, 0.8525, 0.2217, 0.5413, 0.2367, 0.7367, 0.5482],
  34805. [0.6311, 0.4008, 0.7935, 0.5746, 0.3900, 0.5033, 0.6955, 0.5366],
  34806. [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402]],
  34807. device='cuda:0', dtype=torch.float64)
  34808. predictions are: tensor([[ 0.7300, -0.3389, 1.8162, -0.1014, -0.4492, 0.1221, 0.6110, -0.0020],
  34809. [ 0.7533, -0.2762, 1.7339, -0.4205, -0.6105, -0.0945, 0.4151, 0.1860],
  34810. [-1.9275, -2.0202, 1.1070, -1.3464, -0.4541, -1.1237, 0.0258, 0.1614],
  34811. [ 0.7344, -0.2898, 1.6910, 0.2567, -0.4216, 0.0965, 0.2483, 0.2432],
  34812. [ 0.7589, -0.2931, 1.4342, -1.0748, -0.3835, -1.2520, 0.3794, 0.0544],
  34813. [-1.2442, -1.5647, 1.6967, -1.2537, 0.2367, -1.2305, 0.9762, 0.3556],
  34814. [ 0.7263, -0.3257, 1.4122, 0.2276, -0.4983, -0.0615, 0.8091, 0.2047],
  34815. [ 0.6517, -0.3751, 1.6237, 0.1518, -0.4258, -0.1553, 0.3233, 0.1533]],
  34816. device='cuda:0', grad_fn=<AddmmBackward>)
  34817. landmarks are: tensor([[[ 0.6072, -0.4250, 1.8249, -0.0072, -0.4037, 0.1082, 0.6730,
  34818. -0.0881],
  34819. [ 0.5799, -0.4012, 1.8480, -0.4306, -0.5365, -0.1150, 0.4681,
  34820. 0.3315],
  34821. [-2.2859, -2.2859, 0.9012, -1.4006, -0.4672, -1.1928, 0.1342,
  34822. 0.1373],
  34823. [ 0.5742, -0.3792, 1.6460, 0.3084, -0.3460, 0.1467, 0.4162,
  34824. 0.3161],
  34825. [ 0.5785, -0.4259, 1.4228, -1.0261, -0.4190, -1.2189, 0.4763,
  34826. 0.2043],
  34827. [-2.2859, -2.2859, 1.6517, -1.2620, 0.2141, -1.1928, 1.1166,
  34828. 0.2463],
  34829. [ 0.6289, -0.4345, 1.3794, 0.3679, -0.4845, 0.0390, 0.9265,
  34830. 0.1928],
  34831. [ 0.5780, -0.4565, 1.6221, 0.2532, -0.3728, -0.1718, 0.4357,
  34832. 0.2091]]], device='cuda:0')
  34833. loss_train_step before backward: tensor(0.0361, device='cuda:0', grad_fn=<MseLossBackward>)
  34834. loss_train_step after backward: tensor(0.0361, device='cuda:0', grad_fn=<MseLossBackward>)
  34835. loss_train: 0.11386870825663209
  34836. step: 11
  34837. running loss: 0.010351700750602917
  34838. Train Steps: 11/90 Loss: 0.0104 torch.Size([8, 600, 800])
  34839. torch.Size([8, 8])
  34840. tensor([[0.6216, 0.4167, 0.8588, 0.5583, 0.3975, 0.5167, 0.5775, 0.5667],
  34841. [0.6125, 0.3974, 0.7725, 0.2517, 0.3538, 0.3317, 0.5887, 0.5500],
  34842. [0.6143, 0.4040, 0.8237, 0.3333, 0.4075, 0.2117, 0.5137, 0.4973],
  34843. [0.6307, 0.4045, 0.8025, 0.5833, 0.3775, 0.4867, 0.6892, 0.5459],
  34844. [0.6200, 0.3993, 0.8639, 0.3687, 0.3658, 0.3139, 0.6002, 0.5374],
  34845. [0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461],
  34846. [0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170],
  34847. [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483]],
  34848. device='cuda:0', dtype=torch.float64)
  34849. predictions are: tensor([[ 0.6214, -0.3683, 1.7116, 0.2403, -0.3745, 0.1278, 0.5127, 0.3513],
  34850. [ 0.5240, -0.4460, 1.2222, -1.2010, -0.5856, -0.7551, 0.4109, 0.2644],
  34851. [ 0.3942, -0.5166, 1.4681, -0.8702, -0.3608, -1.2539, 0.1577, 0.0390],
  34852. [ 0.6655, -0.3826, 1.4986, 0.1916, -0.4446, -0.0653, 1.0156, 0.2396],
  34853. [ 0.5855, -0.4402, 1.7162, -0.6379, -0.5470, -0.8691, 0.5754, 0.1821],
  34854. [ 0.3844, -0.5187, 1.5678, -0.6282, -0.6653, -0.4034, 0.3104, 0.2106],
  34855. [ 0.7308, -0.3358, 1.6167, -1.1079, 0.0411, -1.5191, 0.8376, 0.1383],
  34856. [ 0.5378, -0.4333, 1.7431, -0.5032, -0.4683, 0.0062, 0.4150, 0.2511]],
  34857. device='cuda:0', grad_fn=<AddmmBackward>)
  34858. landmarks are: tensor([[[ 0.5853, -0.3614, 1.6806, 0.2930, -0.4499, 0.1005, 0.3815,
  34859. 0.3315],
  34860. [ 0.5430, -0.4503, 1.2822, -1.1235, -0.6520, -0.7540, 0.4335,
  34861. 0.2545],
  34862. [ 0.5515, -0.4201, 1.5189, -0.7463, -0.4037, -1.3082, 0.0869,
  34863. 0.0111],
  34864. [ 0.6273, -0.4177, 1.4208, 0.4085, -0.5423, -0.0380, 0.8973,
  34865. 0.2356],
  34866. [ 0.5777, -0.4416, 1.7044, -0.5827, -0.5962, -0.8361, 0.4862,
  34867. 0.1963],
  34868. [ 0.5124, -0.4483, 1.5420, -0.6231, -0.7155, -0.4152, 0.2679,
  34869. 0.2365],
  34870. [ 0.6471, -0.3840, 1.5767, -1.0311, -0.0457, -1.5007, 0.6889,
  34871. 0.1020],
  34872. [ 0.5546, -0.4620, 1.7788, -0.4229, -0.5192, -0.0226, 0.4277,
  34873. 0.2468]]], device='cuda:0')
  34874. loss_train_step before backward: tensor(0.0045, device='cuda:0', grad_fn=<MseLossBackward>)
  34875. loss_train_step after backward: tensor(0.0045, device='cuda:0', grad_fn=<MseLossBackward>)
  34876. loss_train: 0.11839128797873855
  34877. step: 12
  34878. running loss: 0.009865940664894879
  34879.  
  34880. Train Steps: 12/90 Loss: 0.0099 torch.Size([8, 600, 800])
  34881. torch.Size([8, 8])
  34882. tensor([[0.6207, 0.4081, 0.7662, 0.2067, 0.3962, 0.3200, 0.6312, 0.5300],
  34883. [0.6122, 0.4048, 0.7506, 0.3014, 0.3475, 0.3333, 0.5487, 0.5749],
  34884. [0.6263, 0.4057, 0.8800, 0.3833, 0.3650, 0.3717, 0.6375, 0.4804],
  34885. [0.6282, 0.4092, 0.8000, 0.2183, 0.4500, 0.2383, 0.6787, 0.5364],
  34886. [0.6124, 0.4075, 0.7696, 0.4153, 0.3475, 0.3767, 0.5157, 0.5427],
  34887. [0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466],
  34888. [0.6262, 0.4052, 0.8888, 0.4700, 0.3675, 0.5117, 0.6350, 0.5233],
  34889. [ nan, nan, 0.8938, 0.2850, 0.4662, 0.3117, 0.7406, 0.5528]],
  34890. device='cuda:0', dtype=torch.float64)
  34891. predictions are: tensor([[ 0.5405, -0.4378, 1.2390, -1.2238, -0.4669, -0.9086, 0.6034, 0.2385],
  34892. [ 0.4922, -0.4575, 1.1102, -0.8623, -0.6171, -0.7878, 0.2250, 0.3683],
  34893. [ 0.6639, -0.4080, 1.8109, -0.4827, -0.5623, -0.6579, 0.6498, -0.0299],
  34894. [ 0.6907, -0.3644, 1.3608, -1.1843, -0.1819, -1.3539, 0.8125, 0.1886],
  34895. [ 0.5084, -0.4234, 1.3220, -0.4061, -0.6198, -0.5928, 0.0843, 0.2962],
  34896. [ 0.6199, -0.3647, 1.6480, 0.1301, -0.1844, 0.1268, 0.2038, 0.2411],
  34897. [ 0.6979, -0.3626, 1.8685, -0.0838, -0.4998, 0.0246, 0.6699, 0.1938],
  34898. [-1.8352, -1.9717, 1.9515, -0.8713, -0.0920, -0.9134, 1.1173, 0.3375]],
  34899. device='cuda:0', grad_fn=<AddmmBackward>)
  34900. landmarks are: tensor([[[ 0.5809, -0.4011, 1.2533, -1.3313, -0.4557, -0.8079, 0.6298,
  34901. 0.1621],
  34902. [ 0.5419, -0.4160, 1.1810, -0.8939, -0.6808, -0.7463, 0.2485,
  34903. 0.3694],
  34904. [ 0.6071, -0.4119, 1.7788, -0.5153, -0.6000, -0.5692, 0.6586,
  34905. -0.0670],
  34906. [ 0.6158, -0.3960, 1.4092, -1.2774, -0.2074, -1.1851, 0.8491,
  34907. 0.1917],
  34908. [ 0.5427, -0.4035, 1.2688, -0.3675, -0.6808, -0.5461, 0.0959,
  34909. 0.2206],
  34910. [ 0.5178, -0.4480, 1.6748, 0.0620, -0.2767, 0.2083, 0.1067,
  34911. 0.2386],
  34912. [ 0.6063, -0.4142, 1.8192, -0.1150, -0.5885, 0.0774, 0.6471,
  34913. 0.1313],
  34914. [-2.2859, -2.2859, 1.8423, -0.9695, -0.1323, -0.8463, 1.1349,
  34915. 0.2676]]], device='cuda:0')
  34916. loss_train_step before backward: tensor(0.0085, device='cuda:0', grad_fn=<MseLossBackward>)
  34917. loss_train_step after backward: tensor(0.0085, device='cuda:0', grad_fn=<MseLossBackward>)
  34918. loss_train: 0.1268529868684709
  34919. step: 13
  34920. running loss: 0.009757922066805454
  34921. Train Steps: 13/90 Loss: 0.0098 torch.Size([8, 600, 800])
  34922. torch.Size([8, 8])
  34923. tensor([[0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967],
  34924. [0.6198, 0.3997, 0.8582, 0.5361, 0.4117, 0.5016, 0.5942, 0.5134],
  34925. [0.6189, 0.4049, 0.8888, 0.4417, 0.4213, 0.5200, 0.5988, 0.5633],
  34926. [ nan, nan, 0.6688, 0.2513, 0.4113, 0.2117, 0.5193, 0.5933],
  34927. [ nan, nan, 0.6412, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700],
  34928. [0.6265, 0.4251, 0.7113, 0.3550, 0.4375, 0.2117, 0.5587, 0.6118],
  34929. [0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400],
  34930. [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611]],
  34931. device='cuda:0', dtype=torch.float64)
  34932. predictions are: tensor([[ 0.6785, -0.3535, 1.6476, -1.0522, -0.1186, -1.3418, 0.6660, 0.0418],
  34933. [ 0.7046, -0.3856, 1.7158, 0.1641, -0.4697, 0.1161, 0.5764, 0.0492],
  34934. [ 0.5759, -0.4075, 1.8268, -0.2217, -0.4446, 0.2211, 0.6357, 0.2442],
  34935. [-1.3988, -1.6600, 0.8677, -1.2617, -0.4867, -1.2016, 0.2127, 0.3967],
  34936. [-2.0198, -2.1018, 0.9475, -1.3378, -0.4070, -1.3136, 0.2460, 0.3122],
  34937. [ 0.7884, -0.2667, 1.1408, -0.7161, -0.3460, -1.2052, 0.4208, 0.4894],
  34938. [ 0.5331, -0.4345, 1.5298, -1.1118, -0.2301, -1.2077, 0.6807, 0.1989],
  34939. [ 0.6563, -0.3672, 1.7630, -0.0805, -0.3497, -0.0191, 0.3838, 0.3103]],
  34940. device='cuda:0', grad_fn=<AddmmBackward>)
  34941. landmarks are: tensor([[[ 6.0479e-01, -3.7229e-01, 1.6517e+00, -1.0773e+00, 4.6189e-04,
  34942. -1.5161e+00, 5.8660e-01, 8.0947e-03],
  34943. [ 5.7673e-01, -4.3957e-01, 1.6782e+00, 1.9046e-01, -3.8437e-01,
  34944. 3.0841e-02, 4.5876e-01, 8.5521e-02],
  34945. [ 5.7258e-01, -4.1594e-01, 1.8192e+00, -2.4588e-01, -3.4018e-01,
  34946. 1.1594e-01, 4.7968e-01, 3.1609e-01],
  34947. [-2.2859e+00, -2.2859e+00, 8.0331e-01, -1.1250e+00, -3.8637e-01,
  34948. -1.3082e+00, 1.1262e-01, 4.5430e-01],
  34949. [-2.2859e+00, -2.2859e+00, 6.7598e-01, -1.4083e+00, -3.2864e-01,
  34950. -1.4160e+00, 2.4873e-01, 3.4688e-01],
  34951. [ 6.0774e-01, -3.2256e-01, 9.9931e-01, -6.4619e-01, -2.6513e-01,
  34952. -1.3082e+00, 2.9460e-01, 5.4012e-01],
  34953. [ 5.9579e-01, -3.8176e-01, 1.5536e+00, -1.1081e+00, -2.0739e-01,
  34954. -1.3390e+00, 5.6628e-01, 2.0831e-01],
  34955. [ 5.4908e-01, -4.2902e-01, 1.7788e+00, -1.0731e-01, -2.6513e-01,
  34956. -1.0731e-01, 2.5553e-01, 3.0567e-01]]], device='cuda:0')
  34957. loss_train_step before backward: tensor(0.0278, device='cuda:0', grad_fn=<MseLossBackward>)
  34958. loss_train_step after backward: tensor(0.0278, device='cuda:0', grad_fn=<MseLossBackward>)
  34959. loss_train: 0.15469072526320815
  34960. step: 14
  34961. running loss: 0.011049337518800582
  34962. Train Steps: 14/90 Loss: 0.0110 torch.Size([8, 600, 800])
  34963. torch.Size([8, 8])
  34964. tensor([[0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767],
  34965. [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611],
  34966. [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133],
  34967. [0.6095, 0.3970, 0.8688, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940],
  34968. [0.6276, 0.4235, 0.8888, 0.5333, 0.3800, 0.3117, 0.5427, 0.6164],
  34969. [ nan, nan, 0.7225, 0.2167, 0.3987, 0.2283, 0.5427, 0.5181],
  34970. [0.6161, 0.4099, 0.8738, 0.4383, 0.3788, 0.5483, 0.5605, 0.5019],
  34971. [0.6282, 0.4029, 0.8988, 0.3000, 0.4250, 0.4183, 0.7042, 0.5575]],
  34972. device='cuda:0', dtype=torch.float64)
  34973. predictions are: tensor([[ 0.4819, -0.5002, 1.7234, -0.0345, -0.4618, -0.2502, 0.4140, 0.3121],
  34974. [ 0.5653, -0.4157, 1.7060, -0.1241, -0.2971, -0.1908, 0.3426, 0.3553],
  34975. [ 0.4675, -0.4928, 1.6902, 0.0889, -0.2267, 0.0156, 0.6133, 0.1752],
  34976. [ 0.4411, -0.5210, 1.7248, -0.1622, -0.1141, -0.0964, 0.2784, 0.0968],
  34977. [ 0.8619, -0.1851, 1.7110, 0.1010, -0.5076, -0.9692, 0.5351, 0.5752],
  34978. [-2.5459, -2.4300, 0.9579, -1.4209, -0.5373, -1.2719, 0.2063, 0.1480],
  34979. [ 0.5651, -0.4017, 1.7222, -0.3233, -0.5675, 0.1211, 0.5236, 0.2215],
  34980. [ 0.6173, -0.4305, 1.7402, -0.8352, -0.4560, -0.4488, 1.1476, 0.3046]],
  34981. device='cuda:0', grad_fn=<AddmmBackward>)
  34982. landmarks are: tensor([[[ 0.5610, -0.4381, 1.7730, 0.0390, -0.4326, -0.1458, 0.1794,
  34983. 0.3777],
  34984. [ 0.5491, -0.4290, 1.7788, -0.1073, -0.2651, -0.1073, 0.2555,
  34985. 0.3057],
  34986. [ 0.5786, -0.4141, 1.7037, 0.1544, -0.1862, 0.0736, 0.4393,
  34987. 0.0851],
  34988. [ 0.5292, -0.4520, 1.7268, -0.0842, -0.0413, -0.0324, 0.1116,
  34989. -0.0039],
  34990. [ 0.6131, -0.3299, 1.8192, 0.1775, -0.5307, -0.8463, 0.2206,
  34991. 0.5612],
  34992. [-2.2859, -2.2859, 1.0513, -1.2851, -0.4441, -1.2313, 0.2206,
  34993. 0.1073],
  34994. [ 0.5597, -0.3928, 1.7499, -0.2613, -0.5365, 0.2468, 0.3028,
  34995. 0.0321],
  34996. [ 0.6158, -0.4249, 1.8654, -0.9002, -0.3229, -0.3537, 0.9667,
  34997. 0.2890]]], device='cuda:0')
  34998. loss_train_step before backward: tensor(0.0121, device='cuda:0', grad_fn=<MseLossBackward>)
  34999. loss_train_step after backward: tensor(0.0121, device='cuda:0', grad_fn=<MseLossBackward>)
  35000. loss_train: 0.1667864709161222
  35001. step: 15
  35002. running loss: 0.011119098061074813
  35003. Train Steps: 15/90 Loss: 0.0111 torch.Size([8, 600, 800])
  35004. torch.Size([8, 8])
  35005. tensor([[0.6339, 0.4149, 0.8800, 0.5000, 0.3900, 0.5283, 0.7541, 0.5424],
  35006. [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133],
  35007. [0.6275, 0.4024, 0.8500, 0.5383, 0.3912, 0.4883, 0.6288, 0.5100],
  35008. [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617],
  35009. [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261],
  35010. [0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240],
  35011. [0.6104, 0.4029, 0.8738, 0.4900, 0.4088, 0.4533, 0.5070, 0.5510],
  35012. [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377]],
  35013. device='cuda:0', dtype=torch.float64)
  35014. predictions are: tensor([[ 0.4449, -0.5116, 1.8714, -0.0182, -0.5449, 0.1283, 1.1254, 0.3308],
  35015. [ 0.4398, -0.5067, 1.7343, 0.0700, -0.2354, 0.0319, 0.5108, 0.2732],
  35016. [ 0.5798, -0.4334, 1.6550, 0.1778, -0.5138, -0.1207, 0.6395, 0.2326],
  35017. [ 0.6115, -0.3747, 1.8011, -0.1351, -0.3285, -0.0546, 0.4288, 0.4305],
  35018. [ 0.4821, -0.4128, 1.7874, -0.1781, -0.4174, -0.1471, 0.2563, 0.2960],
  35019. [ 0.7083, -0.3171, 1.1143, -1.4232, -0.4142, -1.5667, 0.2975, 0.2217],
  35020. [ 0.5611, -0.3972, 1.7412, -0.0379, -0.4422, -0.2419, 0.1391, 0.3804],
  35021. [ 0.5256, -0.4907, 1.7250, 0.1376, -0.4755, -0.2529, 0.5504, 0.2964]],
  35022. device='cuda:0', grad_fn=<AddmmBackward>)
  35023. landmarks are: tensor([[[ 0.6421, -0.3695, 1.7788, 0.0236, -0.4845, 0.1544, 1.1971,
  35024. 0.2196],
  35025. [ 0.5786, -0.4141, 1.7037, 0.1544, -0.1862, 0.0736, 0.4393,
  35026. 0.0851],
  35027. [ 0.6125, -0.4273, 1.6402, 0.2006, -0.4788, -0.0303, 0.6182,
  35028. 0.0697],
  35029. [ 0.5637, -0.4143, 1.7519, -0.0787, -0.3055, -0.0149, 0.3758,
  35030. 0.3084],
  35031. [ 0.5746, -0.3623, 1.7961, -0.1150, -0.3691, -0.0380, 0.2208,
  35032. 0.1439],
  35033. [ 0.5676, -0.4112, 1.1898, -1.2467, -0.2940, -1.4622, 0.2103,
  35034. 0.1343],
  35035. [ 0.5334, -0.4252, 1.7499, -0.0226, -0.3979, -0.1920, 0.0558,
  35036. 0.2589],
  35037. [ 0.5775, -0.4607, 1.6741, 0.1962, -0.4036, -0.1212, 0.4588,
  35038. 0.1979]]], device='cuda:0')
  35039. loss_train_step before backward: tensor(0.0075, device='cuda:0', grad_fn=<MseLossBackward>)
  35040. loss_train_step after backward: tensor(0.0075, device='cuda:0', grad_fn=<MseLossBackward>)
  35041. loss_train: 0.17424756661057472
  35042. step: 16
  35043. running loss: 0.01089047291316092
  35044.  
  35045. Train Steps: 16/90 Loss: 0.0109 torch.Size([8, 600, 800])
  35046. torch.Size([8, 8])
  35047. tensor([[0.6350, 0.4118, 0.8363, 0.2717, 0.4850, 0.1700, 0.6441, 0.5170],
  35048. [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350],
  35049. [0.6139, 0.4019, 0.7137, 0.2150, 0.4375, 0.1533, 0.5293, 0.5006],
  35050. [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317],
  35051. [0.6201, 0.4029, 0.8776, 0.4602, 0.3683, 0.4203, 0.5938, 0.5450],
  35052. [0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461],
  35053. [0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204],
  35054. [0.6300, 0.4133, 0.8538, 0.2167, 0.5587, 0.2250, 0.7390, 0.5413]],
  35055. device='cuda:0', dtype=torch.float64)
  35056. predictions are: tensor([[ 0.6731, -0.3697, 1.5592, -1.0559, -0.0977, -1.3919, 0.7334, 0.1976],
  35057. [ 0.6478, -0.3031, 1.7829, -0.3219, -0.6307, -0.3326, 0.2386, 0.3660],
  35058. [ 0.4876, -0.4768, 0.8883, -1.4431, -0.3269, -1.5276, 0.1864, 0.1209],
  35059. [ 0.5578, -0.4057, 1.6874, 0.1956, -0.5940, -0.4984, 0.3570, 0.3694],
  35060. [ 0.5158, -0.4404, 1.7622, -0.1544, -0.6287, -0.2941, 0.4888, 0.3767],
  35061. [ 0.3452, -0.5299, 1.5267, -0.5571, -0.7532, -0.3578, 0.2696, 0.2688],
  35062. [ 0.6597, -0.3832, 1.6093, 0.2269, -0.3988, 0.2980, 0.7815, 0.2671],
  35063. [ 0.0781, -0.7422, 1.6755, -1.3120, 0.2650, -1.0328, 1.1781, 0.4177]],
  35064. device='cuda:0', grad_fn=<AddmmBackward>)
  35065. landmarks are: tensor([[[ 0.6471, -0.3840, 1.5767, -1.0311, -0.0457, -1.5007, 0.6889,
  35066. 0.1020],
  35067. [ 0.6072, -0.3239, 1.8423, -0.3537, -0.4961, -0.3921, 0.2083,
  35068. 0.1852],
  35069. [ 0.5496, -0.4295, 1.0109, -1.2928, -0.2651, -1.5777, 0.1589,
  35070. 0.0261],
  35071. [ 0.5901, -0.4157, 1.7557, 0.1929, -0.5423, -0.5923, 0.3584,
  35072. 0.1698],
  35073. [ 0.5785, -0.4252, 1.7676, -0.1602, -0.5845, -0.3446, 0.4566,
  35074. 0.2314],
  35075. [ 0.5124, -0.4483, 1.5420, -0.6231, -0.7155, -0.4152, 0.2679,
  35076. 0.2365],
  35077. [ 0.6042, -0.4273, 1.7198, 0.2184, -0.3478, 0.1149, 0.8062,
  35078. 0.1176],
  35079. [ 0.6240, -0.3768, 1.6575, -1.2851, 0.2949, -1.2467, 1.1276,
  35080. 0.2142]]], device='cuda:0')
  35081. loss_train_step before backward: tensor(0.0138, device='cuda:0', grad_fn=<MseLossBackward>)
  35082. loss_train_step after backward: tensor(0.0138, device='cuda:0', grad_fn=<MseLossBackward>)
  35083. loss_train: 0.18802320212125778
  35084. step: 17
  35085. running loss: 0.011060188360073987
  35086. Train Steps: 17/90 Loss: 0.0111 torch.Size([8, 600, 800])
  35087. torch.Size([8, 8])
  35088. tensor([[0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400],
  35089. [0.6073, 0.3932, 0.8363, 0.5817, 0.4425, 0.5117, 0.5204, 0.4817],
  35090. [0.6168, 0.4111, 0.6517, 0.2875, 0.3688, 0.2817, 0.5228, 0.5837],
  35091. [0.6329, 0.4196, 0.9238, 0.4583, 0.4263, 0.2933, 0.7343, 0.5748],
  35092. [0.6125, 0.3999, 0.8750, 0.4883, 0.4750, 0.4700, 0.5533, 0.5617],
  35093. [0.6115, 0.4005, 0.8838, 0.3867, 0.3763, 0.4700, 0.5800, 0.5550],
  35094. [0.6229, 0.4066, 0.8513, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350],
  35095. [0.6200, 0.4059, 0.8700, 0.4900, 0.4163, 0.5000, 0.6162, 0.5467]],
  35096. device='cuda:0', dtype=torch.float64)
  35097. predictions are: tensor([[ 0.4197, -0.4912, 1.5191, -1.2403, -0.1832, -1.3755, 0.6581, 0.2036],
  35098. [ 0.4366, -0.5245, 1.5805, 0.2763, -0.2760, 0.0055, 0.1530, 0.0448],
  35099. [ 0.6062, -0.3579, 0.8087, -1.0960, -0.6512, -1.0815, 0.1575, 0.4478],
  35100. [ 0.6468, -0.3977, 1.9815, -0.2081, -0.4123, -1.0205, 1.1302, 0.3880],
  35101. [ 0.5060, -0.4523, 1.7256, -0.1588, -0.0872, -0.1265, 0.2759, 0.3836],
  35102. [ 0.4599, -0.4719, 1.7737, -0.5584, -0.6429, -0.1395, 0.4024, 0.2736],
  35103. [ 0.3906, -0.5403, 1.6324, 0.2416, -0.2583, -0.0143, 0.4181, 0.2703],
  35104. [ 0.4343, -0.5111, 1.7783, -0.1365, -0.3835, 0.0125, 0.5447, 0.2704]],
  35105. device='cuda:0', grad_fn=<AddmmBackward>)
  35106. landmarks are: tensor([[[ 0.5958, -0.3818, 1.5536, -1.1081, -0.2074, -1.3390, 0.5663,
  35107. 0.2083],
  35108. [ 0.5193, -0.4699, 1.5767, 0.4008, -0.2420, 0.0774, 0.1178,
  35109. -0.0610],
  35110. [ 0.5629, -0.3871, 0.7243, -0.9581, -0.5827, -0.9849, 0.1288,
  35111. 0.4103],
  35112. [ 0.6372, -0.3479, 1.9808, -0.1689, -0.3171, -0.9310, 1.1057,
  35113. 0.3692],
  35114. [ 0.5432, -0.4388, 1.7557, -0.0303, -0.0919, -0.1150, 0.2699,
  35115. 0.3087],
  35116. [ 0.5384, -0.4361, 1.7961, -0.4999, -0.5480, -0.1150, 0.3931,
  35117. 0.2776],
  35118. [ 0.5911, -0.4080, 1.6460, 0.3546, -0.2074, 0.0467, 0.4970,
  35119. 0.1852],
  35120. [ 0.5776, -0.4109, 1.7326, -0.0226, -0.3633, 0.0236, 0.5605,
  35121. 0.2391]]], device='cuda:0')
  35122. loss_train_step before backward: tensor(0.0057, device='cuda:0', grad_fn=<MseLossBackward>)
  35123. loss_train_step after backward: tensor(0.0057, device='cuda:0', grad_fn=<MseLossBackward>)
  35124. loss_train: 0.19375777710229158
  35125. step: 18
  35126. running loss: 0.010764320950127311
  35127. Train Steps: 18/90 Loss: 0.0108 torch.Size([8, 600, 800])
  35128. torch.Size([8, 8])
  35129. tensor([[0.6102, 0.3999, 0.8750, 0.5133, 0.3825, 0.4750, 0.5637, 0.5083],
  35130. [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317],
  35131. [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383],
  35132. [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550],
  35133. [0.6289, 0.4019, 0.8113, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332],
  35134. [0.6201, 0.4064, 0.8688, 0.5050, 0.4225, 0.5100, 0.6138, 0.5500],
  35135. [0.6127, 0.4084, 0.8700, 0.4467, 0.3987, 0.4317, 0.5013, 0.5471],
  35136. [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082]],
  35137. device='cuda:0', dtype=torch.float64)
  35138. predictions are: tensor([[ 0.5465, -0.4084, 1.7556, -0.0485, -0.4730, -0.2105, 0.2575, 0.1068],
  35139. [ 0.5799, -0.3872, 1.7264, 0.1061, -0.5587, -0.7446, 0.4119, 0.3354],
  35140. [ 0.6407, -0.3606, 1.6930, -0.0809, -0.2809, -0.1557, 0.2516, 0.2782],
  35141. [ 0.3428, -0.5433, 1.7698, -0.2409, 0.0305, -0.0332, 0.5215, 0.3632],
  35142. [ 0.5239, -0.4432, 1.5371, 0.1497, -0.5197, -0.1187, 0.9927, 0.2922],
  35143. [ 0.6542, -0.3643, 1.7341, -0.0491, -0.3532, -0.0352, 0.5050, 0.3007],
  35144. [ 0.4386, -0.4446, 1.6842, -0.2906, -0.4989, -0.4463, 0.0770, 0.3000],
  35145. [ 0.5291, -0.4532, 1.7692, 0.0302, -0.3224, 0.2420, 0.4699, 0.1400]],
  35146. device='cuda:0', grad_fn=<AddmmBackward>)
  35147. landmarks are: tensor([[[ 5.3262e-01, -4.3895e-01, 1.7557e+00, 8.5142e-02, -5.1917e-01,
  35148. -9.1917e-02, 3.1801e-01, 6.2048e-02],
  35149. [ 5.9013e-01, -4.1570e-01, 1.7557e+00, 1.9292e-01, -5.4226e-01,
  35150. -5.9230e-01, 3.5843e-01, 1.6982e-01],
  35151. [ 5.3603e-01, -4.6490e-01, 1.6517e+00, 4.6189e-04, -2.8245e-01,
  35152. -6.8822e-02, 2.3086e-01, 2.0046e-01],
  35153. [ 5.3508e-01, -4.3212e-01, 1.7326e+00, -1.3811e-01, 4.6329e-02,
  35154. 1.0491e-01, 4.1617e-01, 2.7760e-01],
  35155. [ 6.1907e-01, -4.2971e-01, 1.4612e+00, 2.3911e-01, -4.9607e-01,
  35156. 3.1255e-02, 1.1166e+00, 1.7680e-01],
  35157. [ 5.7841e-01, -4.0878e-01, 1.7268e+00, 4.6651e-02, -3.3441e-01,
  35158. 6.9746e-02, 5.4896e-01, 2.5450e-01],
  35159. [ 5.4405e-01, -3.9969e-01, 1.7326e+00, -2.2279e-01, -4.4411e-01,
  35160. -2.9207e-01, 2.9551e-02, 2.4088e-01],
  35161. [ 5.4496e-01, -4.7064e-01, 1.7643e+00, 7.2204e-02, -3.7076e-01,
  35162. 3.2001e-01, 4.8543e-01, 6.1219e-02]]], device='cuda:0')
  35163. loss_train_step before backward: tensor(0.0066, device='cuda:0', grad_fn=<MseLossBackward>)
  35164. loss_train_step after backward: tensor(0.0066, device='cuda:0', grad_fn=<MseLossBackward>)
  35165. loss_train: 0.20038292463868856
  35166. step: 19
  35167. running loss: 0.010546469717825713
  35168. Train Steps: 19/90 Loss: 0.0105 torch.Size([8, 600, 800])
  35169. torch.Size([8, 8])
  35170. tensor([[0.6179, 0.3993, 0.8925, 0.4789, 0.3879, 0.4900, 0.6041, 0.5279],
  35171. [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483],
  35172. [0.6109, 0.4015, 0.7668, 0.3639, 0.3513, 0.3667, 0.5200, 0.5641],
  35173. [0.6189, 0.4049, 0.8888, 0.4417, 0.4213, 0.5200, 0.5988, 0.5633],
  35174. [0.6125, 0.3999, 0.8750, 0.4883, 0.4750, 0.4700, 0.5533, 0.5617],
  35175. [0.6095, 0.3970, 0.8688, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940],
  35176. [0.6239, 0.4123, 0.8313, 0.2550, 0.4500, 0.2050, 0.6175, 0.5400],
  35177. [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550]],
  35178. device='cuda:0', dtype=torch.float64)
  35179. predictions are: tensor([[ 6.1788e-01, -3.8531e-01, 1.8123e+00, -8.3692e-03, -5.4274e-01,
  35180. -1.6935e-01, 6.3533e-01, 1.3633e-01],
  35181. [ 6.6055e-01, -3.6350e-01, 1.0613e+00, -1.3483e+00, -4.5378e-01,
  35182. -1.2605e+00, 5.3864e-01, 2.7648e-01],
  35183. [ 5.9610e-01, -3.5126e-01, 1.2928e+00, -6.1835e-01, -7.0665e-01,
  35184. -7.1798e-01, 2.0525e-01, 3.0496e-01],
  35185. [ 4.8049e-01, -4.5639e-01, 1.7886e+00, -1.5145e-01, -3.8253e-01,
  35186. 7.3162e-02, 5.6609e-01, 2.5449e-01],
  35187. [ 5.1796e-01, -4.4070e-01, 1.7000e+00, 3.3282e-02, -6.6376e-02,
  35188. -1.6498e-01, 2.8864e-01, 3.2839e-01],
  35189. [ 3.5361e-01, -5.8107e-01, 1.7022e+00, -2.0324e-02, -9.5997e-02,
  35190. -7.1067e-02, 2.6798e-01, 6.7780e-02],
  35191. [ 5.0432e-01, -4.3011e-01, 1.5125e+00, -1.0313e+00, -1.7659e-01,
  35192. -1.4298e+00, 6.5754e-01, 1.9049e-01],
  35193. [ 3.6905e-01, -5.3715e-01, 1.7403e+00, -5.2373e-02, -1.7687e-02,
  35194. -1.7288e-03, 5.1245e-01, 2.9648e-01]], device='cuda:0',
  35195. grad_fn=<AddmmBackward>)
  35196. landmarks are: tensor([[[ 0.5680, -0.4417, 1.8365, -0.0740, -0.4941, -0.0227, 0.5045,
  35197. 0.1525],
  35198. [ 0.5784, -0.4085, 1.0859, -1.3929, -0.4037, -1.1158, 0.5605,
  35199. 0.2468],
  35200. [ 0.5357, -0.4315, 1.2557, -0.6051, -0.6635, -0.5923, 0.1159,
  35201. 0.3198],
  35202. [ 0.5726, -0.4159, 1.8192, -0.2459, -0.3402, 0.1159, 0.4797,
  35203. 0.3161],
  35204. [ 0.5432, -0.4388, 1.7557, -0.0303, -0.0919, -0.1150, 0.2699,
  35205. 0.3087],
  35206. [ 0.5292, -0.4520, 1.7268, -0.0842, -0.0413, -0.0324, 0.1116,
  35207. -0.0039],
  35208. [ 0.5958, -0.3818, 1.5536, -1.1081, -0.2074, -1.3390, 0.5663,
  35209. 0.2083],
  35210. [ 0.5351, -0.4321, 1.7326, -0.1381, 0.0463, 0.1049, 0.4162,
  35211. 0.2776]]], device='cuda:0')
  35212. loss_train_step before backward: tensor(0.0057, device='cuda:0', grad_fn=<MseLossBackward>)
  35213. loss_train_step after backward: tensor(0.0057, device='cuda:0', grad_fn=<MseLossBackward>)
  35214. loss_train: 0.20609361585229635
  35215. step: 20
  35216. running loss: 0.010304680792614818
  35217.  
  35218. Train Steps: 20/90 Loss: 0.0103 torch.Size([8, 600, 800])
  35219. torch.Size([8, 8])
  35220. tensor([[0.6200, 0.3993, 0.8519, 0.4923, 0.3962, 0.4717, 0.6013, 0.5433],
  35221. [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217],
  35222. [0.6131, 0.4037, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680],
  35223. [0.6176, 0.3911, 0.8738, 0.4217, 0.3488, 0.4033, 0.6025, 0.4817],
  35224. [0.6200, 0.4086, 0.8414, 0.2974, 0.4117, 0.2274, 0.5869, 0.5391],
  35225. [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5687, 0.5567],
  35226. [0.6127, 0.4118, 0.8650, 0.5083, 0.4088, 0.5367, 0.5300, 0.5456],
  35227. [0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133]],
  35228. device='cuda:0', dtype=torch.float64)
  35229. predictions are: tensor([[ 0.4543, -0.5039, 1.7159, 0.1398, -0.3645, -0.0678, 0.5161, 0.2738],
  35230. [ 0.4250, -0.5397, 1.8740, 0.0648, -0.3627, 0.3088, 0.6891, 0.1285],
  35231. [ 0.5432, -0.4129, 0.9103, -1.0536, -0.5007, -1.0081, 0.1384, 0.3272],
  35232. [ 0.5651, -0.4690, 1.7934, -0.2742, -0.5696, -0.4388, 0.4848, 0.0175],
  35233. [ 0.6463, -0.3446, 1.6692, -0.8134, -0.2191, -1.0957, 0.4650, 0.1368],
  35234. [ 0.3467, -0.5055, 1.0952, -0.9362, -0.3407, -1.0388, 0.3832, 0.4081],
  35235. [ 0.6227, -0.3434, 1.7308, 0.1826, -0.2582, 0.2336, 0.2736, 0.2280],
  35236. [ 0.6184, -0.4156, 1.5748, -1.0515, -0.0119, -1.4673, 0.5948, 0.1301]],
  35237. device='cuda:0', grad_fn=<AddmmBackward>)
  35238. landmarks are: tensor([[[ 0.5777, -0.4415, 1.6491, -0.0122, -0.4557, -0.1073, 0.4912,
  35239. 0.2237],
  35240. [ 0.6047, -0.4209, 1.7557, -0.0303, -0.4845, 0.2545, 0.6587,
  35241. 0.1236],
  35242. [ 0.5459, -0.4215, 0.9043, -0.9838, -0.5827, -1.0388, 0.1236,
  35243. 0.3378],
  35244. [ 0.5669, -0.4794, 1.7499, -0.3383, -0.6751, -0.4229, 0.4970,
  35245. -0.0611],
  35246. [ 0.5776, -0.3987, 1.6005, -0.9121, -0.3844, -1.2358, 0.4247,
  35247. 0.2043],
  35248. [ 0.5824, -0.3625, 1.0405, -0.9325, -0.4268, -1.1389, 0.3411,
  35249. 0.2853],
  35250. [ 0.5443, -0.3840, 1.7095, 0.0620, -0.3979, 0.1929, 0.1622,
  35251. 0.2341],
  35252. [ 0.6085, -0.4084, 1.5536, -1.1466, -0.0746, -1.4853, 0.6298,
  35253. 0.0851]]], device='cuda:0')
  35254. loss_train_step before backward: tensor(0.0070, device='cuda:0', grad_fn=<MseLossBackward>)
  35255. loss_train_step after backward: tensor(0.0070, device='cuda:0', grad_fn=<MseLossBackward>)
  35256. loss_train: 0.21307452209293842
  35257. step: 21
  35258. running loss: 0.010146405813949448
  35259. Train Steps: 21/90 Loss: 0.0101 torch.Size([8, 600, 800])
  35260. torch.Size([8, 8])
  35261. tensor([[0.6196, 0.4090, 0.8450, 0.3900, 0.3575, 0.3933, 0.5975, 0.5583],
  35262. [0.6353, 0.4128, 0.9138, 0.3533, 0.4688, 0.3250, 0.7145, 0.5991],
  35263. [0.6198, 0.4114, 0.8263, 0.3283, 0.3550, 0.3583, 0.5813, 0.5667],
  35264. [0.6339, 0.4149, 0.8800, 0.5000, 0.3900, 0.5283, 0.7541, 0.5424],
  35265. [0.6236, 0.4084, 0.7738, 0.2133, 0.3663, 0.3233, 0.5813, 0.5567],
  35266. [0.6048, 0.3928, 0.8538, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719],
  35267. [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960],
  35268. [0.6125, 0.4076, 0.8488, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505]],
  35269. device='cuda:0', dtype=torch.float64)
  35270. predictions are: tensor([[ 0.6394, -0.3515, 1.6325, -0.3776, -0.5088, -0.5482, 0.3714, 0.3221],
  35271. [-0.5533, -1.1389, 1.7858, -0.4646, 0.0421, -0.8907, 1.0837, 0.3898],
  35272. [ 0.6300, -0.3319, 1.5087, -0.6143, -0.5620, -0.6781, 0.2920, 0.3042],
  35273. [ 0.6850, -0.3482, 1.7831, 0.1664, -0.3374, 0.1902, 1.0232, 0.1071],
  35274. [ 0.5287, -0.4297, 1.2409, -1.1871, -0.4289, -0.8006, 0.4612, 0.2733],
  35275. [ 0.4822, -0.4647, 1.5526, 0.3464, -0.3086, 0.0092, 0.1336, -0.0019],
  35276. [ 0.5267, -0.4623, 1.7165, -0.0318, -0.0668, 0.0993, 0.1245, -0.0606],
  35277. [ 0.4854, -0.4350, 1.5649, -0.3733, -0.5061, -0.6786, -0.0152, 0.1631]],
  35278. device='cuda:0', grad_fn=<AddmmBackward>)
  35279. landmarks are: tensor([[[ 0.5761, -0.3966, 1.6171, -0.4845, -0.6346, -0.4691, 0.4739,
  35280. 0.2930],
  35281. [ 0.6487, -0.3792, 1.9346, -0.6539, -0.1208, -0.7848, 1.0143,
  35282. 0.4814],
  35283. [ 0.5768, -0.3857, 1.5305, -0.7694, -0.6462, -0.6308, 0.3988,
  35284. 0.3315],
  35285. [ 0.6421, -0.3695, 1.7788, 0.0236, -0.4845, 0.1544, 1.1971,
  35286. 0.2196],
  35287. [ 0.5946, -0.3995, 1.2880, -1.3005, -0.5942, -0.7925, 0.3988,
  35288. 0.2853],
  35289. [ 0.5079, -0.4714, 1.6575, 0.2237, -0.4961, 0.0774, 0.1465,
  35290. -0.1061],
  35291. [ 0.5205, -0.4577, 1.7326, -0.1304, -0.1862, 0.0447, 0.1301,
  35292. 0.0051],
  35293. [ 0.5433, -0.4032, 1.6344, -0.4922, -0.5769, -0.5846, 0.0357,
  35294. 0.2567]]], device='cuda:0')
  35295. loss_train_step before backward: tensor(0.0396, device='cuda:0', grad_fn=<MseLossBackward>)
  35296. loss_train_step after backward: tensor(0.0396, device='cuda:0', grad_fn=<MseLossBackward>)
  35297. loss_train: 0.2526354994624853
  35298. step: 22
  35299. running loss: 0.011483431793749332
  35300. Train Steps: 22/90 Loss: 0.0115 torch.Size([8, 600, 800])
  35301. torch.Size([8, 8])
  35302. tensor([[0.6179, 0.4008, 0.8600, 0.4015, 0.3932, 0.2515, 0.5711, 0.5438],
  35303. [0.6090, 0.4010, 0.7838, 0.3483, 0.3538, 0.3783, 0.5462, 0.5077],
  35304. [0.6265, 0.4088, 0.8025, 0.1850, 0.4163, 0.2500, 0.6290, 0.4947],
  35305. [0.6286, 0.4274, 0.8500, 0.4500, 0.4525, 0.2583, 0.5440, 0.6209],
  35306. [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672],
  35307. [0.6263, 0.4039, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804],
  35308. [0.6081, 0.3950, 0.8538, 0.4667, 0.3850, 0.4917, 0.5342, 0.4954],
  35309. [ nan, nan, 0.7097, 0.2346, 0.4250, 0.1850, 0.5175, 0.5583]],
  35310. device='cuda:0', dtype=torch.float64)
  35311. predictions are: tensor([[ 0.6787, -0.3309, 1.5545, -0.5017, -0.2383, -1.0141, 0.2557, 0.2843],
  35312. [ 0.6255, -0.3809, 1.3916, -0.6963, -0.5258, -0.5401, 0.1449, 0.1210],
  35313. [ 0.5222, -0.4661, 1.4427, -1.3368, -0.2490, -0.9906, 0.6824, 0.1045],
  35314. [ 0.5180, -0.3779, 1.5996, -0.1218, -0.0797, -0.9529, 0.3193, 0.5528],
  35315. [ 0.6336, -0.4351, 1.7194, 0.2918, -0.4420, -0.1731, 0.6663, 0.0219],
  35316. [ 0.5221, -0.4751, 1.8630, -0.2368, -0.4454, -0.0110, 0.6517, -0.0719],
  35317. [ 0.5000, -0.4733, 1.6916, -0.1716, -0.3797, 0.0467, 0.2077, -0.0549],
  35318. [-2.3529, -2.2758, 0.7838, -1.2784, -0.2598, -1.3029, 0.2252, 0.2022]],
  35319. device='cuda:0', grad_fn=<AddmmBackward>)
  35320. landmarks are: tensor([[[ 5.6801e-01, -4.3453e-01, 1.6864e+00, -4.3153e-01, -4.6981e-01,
  35321. -1.1241e+00, 3.5183e-01, 2.2607e-01],
  35322. [ 5.2702e-01, -4.3356e-01, 1.3342e+00, -6.7698e-01, -6.5196e-01,
  35323. -5.3841e-01, 2.3702e-01, 5.9193e-02],
  35324. [ 6.0785e-01, -3.9761e-01, 1.4208e+00, -1.4314e+00, -3.6328e-01,
  35325. -1.1312e+00, 6.1950e-01, -9.2270e-04],
  35326. [ 6.1742e-01, -3.1175e-01, 1.6402e+00, -2.0739e-01, -1.9584e-01,
  35327. -1.0927e+00, 2.2674e-01, 5.8220e-01],
  35328. [ 6.1276e-01, -4.3749e-01, 1.7788e+00, 2.6990e-01, -6.3464e-01,
  35329. -2.5358e-01, 5.4635e-01, -1.2778e-01],
  35330. [ 6.0716e-01, -4.2055e-01, 1.8711e+00, -2.5358e-01, -6.1155e-01,
  35331. -1.3041e-01, 6.8119e-01, -6.7050e-02],
  35332. [ 5.2269e-01, -4.6151e-01, 1.6575e+00, -1.3041e-01, -5.0762e-01,
  35333. -1.4935e-02, 1.8150e-01, 2.0831e-03],
  35334. [-2.2859e+00, -2.2859e+00, 9.9216e-01, -1.2021e+00, -3.2286e-01,
  35335. -1.4314e+00, 1.0439e-01, 2.9299e-01]]], device='cuda:0')
  35336. loss_train_step before backward: tensor(0.0088, device='cuda:0', grad_fn=<MseLossBackward>)
  35337. loss_train_step after backward: tensor(0.0088, device='cuda:0', grad_fn=<MseLossBackward>)
  35338. loss_train: 0.2614791188389063
  35339. step: 23
  35340. running loss: 0.011368657340822012
  35341. Train Steps: 23/90 Loss: 0.0114 torch.Size([8, 600, 800])
  35342. torch.Size([8, 8])
  35343. tensor([[0.6079, 0.3964, 0.7420, 0.2958, 0.3563, 0.2917, 0.5351, 0.4980],
  35344. [0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869],
  35345. [0.6138, 0.4101, 0.8800, 0.5083, 0.4637, 0.5950, 0.5587, 0.5077],
  35346. [0.6154, 0.4112, 0.7037, 0.2317, 0.4238, 0.1833, 0.5350, 0.5600],
  35347. [0.6300, 0.4102, 0.9088, 0.4433, 0.4088, 0.3067, 0.6820, 0.5540],
  35348. [0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585],
  35349. [0.6219, 0.4114, 0.8175, 0.2817, 0.3925, 0.2783, 0.5900, 0.5350],
  35350. [0.6189, 0.3911, 0.8800, 0.4917, 0.3713, 0.4800, 0.6038, 0.4833]],
  35351. device='cuda:0', dtype=torch.float64)
  35352. predictions are: tensor([[ 0.5986, -0.4155, 1.1743, -0.8504, -0.5632, -0.9299, 0.2082, -0.0190],
  35353. [ 0.5951, -0.3816, 1.5139, -1.1340, -0.0489, -1.3246, 0.5283, -0.0248],
  35354. [ 0.5897, -0.3812, 1.7208, 0.1187, -0.1267, 0.3652, 0.2454, 0.0767],
  35355. [ 0.5437, -0.4196, 0.9207, -1.1098, -0.2944, -1.4431, 0.1695, 0.3735],
  35356. [-0.3162, -1.0101, 1.7977, -0.1153, -0.3829, -0.8892, 0.7867, 0.1933],
  35357. [ 0.5377, -0.4063, 1.7138, -0.2150, -0.2550, 0.3595, 0.3530, 0.1509],
  35358. [ 0.4393, -0.4756, 1.4093, -0.8884, -0.4290, -0.8770, 0.4530, 0.2676],
  35359. [ 0.7966, -0.3260, 1.7673, 0.0466, -0.4888, -0.0788, 0.4975, -0.0127]],
  35360. device='cuda:0', grad_fn=<AddmmBackward>)
  35361. landmarks are: tensor([[[ 0.5219, -0.4550, 1.1415, -0.9196, -0.6404, -0.9387, 0.1856,
  35362. 0.0141],
  35363. [ 0.6127, -0.3944, 1.5189, -1.2467, -0.1323, -1.4622, 0.5646,
  35364. -0.0369],
  35365. [ 0.5491, -0.3918, 1.7788, 0.0620, -0.1439, 0.4624, 0.2946,
  35366. 0.0592],
  35367. [ 0.5565, -0.3864, 0.9647, -1.2159, -0.3286, -1.4391, 0.1852,
  35368. 0.3007],
  35369. [ 0.6240, -0.3912, 1.9115, -0.2382, -0.3979, -0.8694, 0.8644,
  35370. 0.2730],
  35371. [ 0.5115, -0.4332, 1.7557, -0.3152, -0.2536, 0.3931, 0.4139,
  35372. 0.2936],
  35373. [ 0.5868, -0.3858, 1.4901, -0.9849, -0.4730, -1.0003, 0.4393,
  35374. 0.1852],
  35375. [ 0.5726, -0.4795, 1.7788, -0.0149, -0.5711, -0.0688, 0.5028,
  35376. -0.0534]]], device='cuda:0')
  35377. loss_train_step before backward: tensor(0.0251, device='cuda:0', grad_fn=<MseLossBackward>)
  35378. loss_train_step after backward: tensor(0.0251, device='cuda:0', grad_fn=<MseLossBackward>)
  35379. loss_train: 0.28657217137515545
  35380. step: 24
  35381. running loss: 0.011940507140631476
  35382.  
  35383. Train Steps: 24/90 Loss: 0.0119 torch.Size([8, 600, 800])
  35384. torch.Size([8, 8])
  35385. tensor([[0.6339, 0.4118, 0.7988, 0.5800, 0.3912, 0.4583, 0.7343, 0.5760],
  35386. [ nan, nan, 0.6469, 0.1943, 0.4025, 0.2000, 0.5125, 0.5533],
  35387. [0.6314, 0.4107, 0.8750, 0.5100, 0.3788, 0.4900, 0.7121, 0.5864],
  35388. [0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879],
  35389. [0.6204, 0.4007, 0.7838, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167],
  35390. [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383],
  35391. [0.6277, 0.4103, 0.8087, 0.5717, 0.4188, 0.4750, 0.5663, 0.6083],
  35392. [0.6275, 0.4050, 0.9038, 0.3767, 0.3838, 0.3533, 0.7074, 0.5575]],
  35393. device='cuda:0', dtype=torch.float64)
  35394. predictions are: tensor([[ 0.5811, -0.4193, 1.4924, 0.1970, -0.4542, -0.1588, 0.7031, 0.1884],
  35395. [-2.8906, -2.6697, 0.7015, -1.2611, -0.3438, -1.3725, 0.1311, 0.1136],
  35396. [ 0.6681, -0.3650, 1.8313, 0.0553, -0.4647, 0.0585, 0.6894, 0.1858],
  35397. [ 0.6057, -0.3989, 1.8592, -0.2476, -0.4531, 0.3330, 0.2408, -0.1263],
  35398. [ 0.4641, -0.5044, 1.3320, -1.1807, -0.1842, -1.3894, 0.2375, -0.0075],
  35399. [ 0.5326, -0.4550, 1.2745, -1.0956, -0.1449, -1.3885, 0.2040, 0.1259],
  35400. [ 0.6367, -0.3457, 1.5546, 0.2843, -0.3334, -0.0293, 0.0620, 0.3324],
  35401. [ 0.6460, -0.3894, 1.9278, -0.4953, -0.4726, -0.5593, 0.6979, 0.1154]],
  35402. device='cuda:0', grad_fn=<AddmmBackward>)
  35403. landmarks are: tensor([[[ 0.6421, -0.3840, 1.4035, 0.3931, -0.4788, -0.1689, 1.1057,
  35404. 0.3745],
  35405. [-2.2859, -2.2859, 0.7023, -1.3883, -0.4268, -1.3621, 0.0813,
  35406. 0.2699],
  35407. [ 0.6306, -0.3888, 1.7557, 0.0697, -0.5365, -0.0226, 1.0033,
  35408. 0.4226],
  35409. [ 0.5676, -0.4327, 1.8252, -0.2095, -0.4886, 0.3084, 0.4727,
  35410. -0.0322],
  35411. [ 0.5796, -0.4351, 1.3342, -1.3159, -0.2189, -1.4853, 0.4046,
  35412. 0.1005],
  35413. [ 0.5911, -0.3984, 1.1956, -1.0850, -0.0804, -1.5392, 0.4393,
  35414. 0.2006],
  35415. [ 0.6134, -0.3910, 1.4497, 0.3546, -0.3517, -0.0919, 0.3296,
  35416. 0.5239],
  35417. [ 0.6125, -0.4153, 1.8885, -0.5461, -0.5134, -0.6539, 0.9814,
  35418. 0.2890]]], device='cuda:0')
  35419. loss_train_step before backward: tensor(0.0235, device='cuda:0', grad_fn=<MseLossBackward>)
  35420. loss_train_step after backward: tensor(0.0235, device='cuda:0', grad_fn=<MseLossBackward>)
  35421. loss_train: 0.3100247588008642
  35422. step: 25
  35423. running loss: 0.012400990352034569
  35424. Train Steps: 25/90 Loss: 0.0124 torch.Size([8, 600, 800])
  35425. torch.Size([8, 8])
  35426. tensor([[ nan, nan, 0.7412, 0.2200, 0.4450, 0.1517, 0.5312, 0.4983],
  35427. [0.6162, 0.3949, 0.8838, 0.4517, 0.4250, 0.5183, 0.5837, 0.5500],
  35428. [0.6310, 0.4017, 0.8563, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006],
  35429. [0.6254, 0.4076, 0.8700, 0.3267, 0.4150, 0.3083, 0.7050, 0.5609],
  35430. [0.6200, 0.4071, 0.7338, 0.1917, 0.4200, 0.2450, 0.6150, 0.5517],
  35431. [0.6038, 0.3946, 0.8413, 0.4883, 0.3563, 0.4550, 0.5266, 0.4693],
  35432. [ nan, nan, 0.8037, 0.2483, 0.3975, 0.2517, 0.5575, 0.5600],
  35433. [0.6185, 0.4067, 0.8838, 0.4450, 0.4037, 0.4733, 0.5213, 0.5142]],
  35434. device='cuda:0', dtype=torch.float64)
  35435. predictions are: tensor([[-1.1408, -1.5361, 1.0224, -1.2326, -0.2584, -1.5380, 0.1721, 0.0526],
  35436. [ 0.7885, -0.2863, 1.8197, -0.1257, -0.3346, 0.1764, 0.3208, 0.1103],
  35437. [ 0.9252, -0.2396, 1.6979, 0.4724, -0.5313, -0.0838, 0.6209, -0.0663],
  35438. [ 0.8023, -0.2682, 1.7999, -0.7008, -0.3828, -0.8174, 0.7808, 0.2176],
  35439. [ 0.7457, -0.3124, 1.2050, -1.2411, -0.3297, -1.1018, 0.5817, 0.2070],
  35440. [ 0.6664, -0.3476, 1.6641, 0.0448, -0.5411, -0.1716, 0.1976, 0.0074],
  35441. [-1.9553, -2.0598, 1.1571, -0.9643, -0.3716, -1.0576, 0.3377, 0.1847],
  35442. [ 0.8637, -0.2172, 1.7630, -0.1238, -0.3981, -0.1083, 0.0655, 0.0336]],
  35443. device='cuda:0', grad_fn=<AddmmBackward>)
  35444. landmarks are: tensor([[[-2.2859, -2.2859, 1.1379, -1.2697, -0.2305, -1.5854, 0.1679,
  35445. 0.0159],
  35446. [ 0.5604, -0.4620, 1.7961, -0.1997, -0.3229, 0.1082, 0.4104,
  35447. 0.2545],
  35448. [ 0.6286, -0.4303, 1.6691, 0.3931, -0.5827, -0.0919, 0.6359,
  35449. 0.0261],
  35450. [ 0.6026, -0.4032, 1.7326, -0.7771, -0.3691, -0.8617, 0.9704,
  35451. 0.3050],
  35452. [ 0.5779, -0.4054, 1.1032, -1.4006, -0.3460, -1.1543, 0.5547,
  35453. 0.2622],
  35454. [ 0.5030, -0.4631, 1.5998, -0.0303, -0.6404, -0.1843, 0.1465,
  35455. -0.1181],
  35456. [-2.2859, -2.2859, 1.4266, -1.1389, -0.4499, -1.1235, 0.2891,
  35457. 0.3007],
  35458. [ 0.5708, -0.4075, 1.7961, -0.2305, -0.4210, -0.0996, 0.1219,
  35459. 0.0893]]], device='cuda:0')
  35460. loss_train_step before backward: tensor(0.0447, device='cuda:0', grad_fn=<MseLossBackward>)
  35461. loss_train_step after backward: tensor(0.0447, device='cuda:0', grad_fn=<MseLossBackward>)
  35462. loss_train: 0.3547267075628042
  35463. step: 26
  35464. running loss: 0.013643334906261701
  35465. Train Steps: 26/90 Loss: 0.0136 torch.Size([8, 600, 800])
  35466. torch.Size([8, 8])
  35467. tensor([[0.6364, 0.4165, 0.9088, 0.4367, 0.4075, 0.3150, 0.6448, 0.5297],
  35468. [0.6273, 0.4110, 0.8900, 0.3817, 0.4188, 0.2167, 0.5858, 0.4835],
  35469. [0.6300, 0.4013, 0.8938, 0.5350, 0.3675, 0.4600, 0.6456, 0.4973],
  35470. [0.6212, 0.4159, 0.8675, 0.5783, 0.4088, 0.4317, 0.5613, 0.5917],
  35471. [0.6169, 0.4052, 0.8742, 0.4779, 0.4288, 0.4917, 0.5763, 0.5617],
  35472. [0.6137, 0.4084, 0.8076, 0.3889, 0.3650, 0.3150, 0.5356, 0.5896],
  35473. [0.6250, 0.4236, 0.8638, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799],
  35474. [0.6127, 0.4118, 0.8650, 0.5083, 0.4088, 0.5367, 0.5300, 0.5456]],
  35475. device='cuda:0', dtype=torch.float64)
  35476. predictions are: tensor([[ 0.6620, -0.3677, 1.8076, -0.3847, -0.4292, -0.8024, 0.6340, 0.0763],
  35477. [ 0.5911, -0.4249, 1.7981, -0.6270, -0.3402, -1.2767, 0.5001, -0.1651],
  35478. [ 0.5506, -0.5222, 1.7383, 0.0278, -0.5927, -0.0961, 0.7650, -0.1522],
  35479. [ 0.6720, -0.3217, 1.5277, 0.2882, -0.4061, -0.1802, 0.2474, 0.3197],
  35480. [ 0.5696, -0.4438, 1.7033, -0.2212, -0.2888, 0.0774, 0.3652, 0.1547],
  35481. [ 0.5928, -0.4065, 1.4053, -0.6018, -0.6057, -0.8218, 0.1283, 0.2768],
  35482. [ 0.4583, -0.4911, 1.6556, -0.6127, -0.3981, -0.7983, 0.3458, 0.3686],
  35483. [ 0.5000, -0.4589, 1.6152, -0.0718, -0.3428, 0.2066, 0.2578, 0.1003]],
  35484. device='cuda:0', grad_fn=<AddmmBackward>)
  35485. landmarks are: tensor([[[ 0.6537, -0.3623, 1.9115, -0.2690, -0.4037, -0.8309, 0.6926,
  35486. 0.1608],
  35487. [ 0.6117, -0.3877, 1.8249, -0.5230, -0.3517, -1.2851, 0.4200,
  35488. -0.0529],
  35489. [ 0.6240, -0.4321, 1.8423, 0.1852, -0.5885, -0.1612, 0.6962,
  35490. 0.0111],
  35491. [ 0.5836, -0.3649, 1.7210, 0.3854, -0.3979, -0.2921, 0.3065,
  35492. 0.4470],
  35493. [ 0.5637, -0.4143, 1.7519, -0.0787, -0.3055, -0.0149, 0.3758,
  35494. 0.3084],
  35495. [ 0.5485, -0.3997, 1.4445, -0.4895, -0.6000, -0.8309, 0.1878,
  35496. 0.4374],
  35497. [ 0.6008, -0.3293, 1.7037, -0.5461, -0.4152, -0.8309, 0.3234,
  35498. 0.3928],
  35499. [ 0.5443, -0.3840, 1.7095, 0.0620, -0.3979, 0.1929, 0.1622,
  35500. 0.2341]]], device='cuda:0')
  35501. loss_train_step before backward: tensor(0.0073, device='cuda:0', grad_fn=<MseLossBackward>)
  35502. loss_train_step after backward: tensor(0.0073, device='cuda:0', grad_fn=<MseLossBackward>)
  35503. loss_train: 0.3619840261526406
  35504. step: 27
  35505. running loss: 0.013406815783431133
  35506. Train Steps: 27/90 Loss: 0.0134 torch.Size([8, 600, 800])
  35507. torch.Size([8, 8])
  35508. tensor([[0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5413, 0.5717],
  35509. [0.6263, 0.4233, 0.7924, 0.4626, 0.3788, 0.2883, 0.5573, 0.6047],
  35510. [0.6153, 0.4117, 0.8688, 0.5167, 0.4895, 0.5647, 0.5524, 0.5136],
  35511. [0.6179, 0.4118, 0.7278, 0.4237, 0.3588, 0.3400, 0.5675, 0.5917],
  35512. [0.6205, 0.4012, 0.8675, 0.3283, 0.3713, 0.3050, 0.5813, 0.5117],
  35513. [0.6257, 0.4167, 0.8775, 0.3433, 0.3563, 0.4133, 0.6200, 0.5667],
  35514. [0.6258, 0.4143, 0.8525, 0.2617, 0.4950, 0.1667, 0.6219, 0.4967],
  35515. [0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356]],
  35516. device='cuda:0', dtype=torch.float64)
  35517. predictions are: tensor([[ 0.5315, -0.4936, 1.6832, -0.1068, -0.3618, -0.0711, 0.2883, 0.1341],
  35518. [ 0.4543, -0.4824, 1.3927, -0.0890, -0.5153, -0.8565, 0.3237, 0.4404],
  35519. [ 0.5046, -0.4862, 1.6858, 0.0078, -0.1689, 0.1853, 0.3572, 0.0407],
  35520. [ 0.6492, -0.3580, 1.2518, -0.4897, -0.6171, -0.7439, 0.2897, 0.3618],
  35521. [ 0.4292, -0.5649, 1.7148, -0.7917, -0.5799, -0.9039, 0.4304, -0.0680],
  35522. [ 0.7026, -0.3457, 1.7546, -0.6548, -0.7039, -0.3940, 0.5502, 0.2667],
  35523. [ 0.4067, -0.5144, 1.6281, -1.1137, -0.1014, -1.4728, 0.5556, -0.0738],
  35524. [ 0.5445, -0.4663, 1.4199, -1.1006, -0.3598, -1.2067, 0.7418, 0.0817]],
  35525. device='cuda:0', grad_fn=<AddmmBackward>)
  35526. landmarks are: tensor([[[ 5.4319e-01, -4.4619e-01, 1.7557e+00, -3.8029e-02, -3.1132e-01,
  35527. -7.6520e-02, 2.1409e-01, 3.5458e-01],
  35528. [ 6.0687e-01, -3.3095e-01, 1.3742e+00, -1.4927e-01, -5.3649e-01,
  35529. -9.5412e-01, 2.8843e-01, 5.0705e-01],
  35530. [ 5.5635e-01, -3.8422e-01, 1.7268e+00, 1.0054e-01, -2.4997e-02,
  35531. 3.2255e-01, 2.6581e-01, 8.6245e-02],
  35532. [ 5.6801e-01, -3.8397e-01, 1.0756e+00, -3.2902e-01, -6.2887e-01,
  35533. -7.1547e-01, 3.3533e-01, 4.4696e-01],
  35534. [ 5.7991e-01, -4.3295e-01, 1.7210e+00, -7.6936e-01, -5.7113e-01,
  35535. -8.7714e-01, 3.9885e-01, 7.7444e-02],
  35536. [ 6.0404e-01, -3.6135e-01, 1.7672e+00, -7.0008e-01, -6.4042e-01,
  35537. -3.7675e-01, 5.7783e-01, 3.3149e-01],
  35538. [ 6.0479e-01, -3.7229e-01, 1.6517e+00, -1.0773e+00, 4.6189e-04,
  35539. -1.5161e+00, 5.8660e-01, 8.0947e-03],
  35540. [ 6.1742e-01, -4.2249e-01, 1.4975e+00, -1.1709e+00, -3.1736e-01,
  35541. -1.1806e+00, 6.5391e-01, 1.8793e-01]]], device='cuda:0')
  35542. loss_train_step before backward: tensor(0.0076, device='cuda:0', grad_fn=<MseLossBackward>)
  35543. loss_train_step after backward: tensor(0.0076, device='cuda:0', grad_fn=<MseLossBackward>)
  35544. loss_train: 0.36961828637868166
  35545. step: 28
  35546. running loss: 0.013200653084952916
  35547.  
  35548. Train Steps: 28/90 Loss: 0.0132 torch.Size([8, 600, 800])
  35549. torch.Size([8, 8])
  35550. tensor([[0.6154, 0.4048, 0.7100, 0.2067, 0.4338, 0.1667, 0.5413, 0.5220],
  35551. [0.6279, 0.4008, 0.8600, 0.4883, 0.4325, 0.5283, 0.7010, 0.5378],
  35552. [0.6211, 0.3993, 0.7650, 0.1933, 0.4575, 0.1550, 0.5965, 0.4895],
  35553. [0.6139, 0.4019, 0.7137, 0.2150, 0.4375, 0.1533, 0.5293, 0.5006],
  35554. [ nan, nan, 0.6512, 0.1717, 0.4100, 0.1983, 0.5253, 0.5240],
  35555. [0.6199, 0.4102, 0.8950, 0.4417, 0.4012, 0.5367, 0.6112, 0.5967],
  35556. [0.6267, 0.4065, 0.8313, 0.2467, 0.4788, 0.1733, 0.6312, 0.5133],
  35557. [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901]],
  35558. device='cuda:0', dtype=torch.float64)
  35559. predictions are: tensor([[ 0.4313, -0.5155, 1.1067, -1.0634, -0.3721, -1.3428, 0.2664, 0.1494],
  35560. [ 0.6846, -0.3799, 1.7791, 0.0894, -0.4743, 0.2629, 0.8621, 0.1941],
  35561. [ 0.4842, -0.4876, 1.2978, -1.2284, -0.2565, -1.3639, 0.4528, 0.0666],
  35562. [ 0.3270, -0.5922, 1.1034, -1.0974, -0.3443, -1.4218, 0.2423, 0.0650],
  35563. [-2.2664, -2.2828, 0.8286, -1.3220, -0.4161, -1.3312, 0.1259, 0.1718],
  35564. [ 0.7369, -0.3070, 1.8857, -0.0652, -0.5453, 0.3664, 0.5765, 0.3902],
  35565. [ 0.5477, -0.4333, 1.6124, -0.9366, -0.2165, -1.2759, 0.6049, 0.1203],
  35566. [ 0.4474, -0.5065, 1.5137, -0.7639, -0.6022, -0.9738, 0.1621, 0.0091]],
  35567. device='cuda:0', grad_fn=<AddmmBackward>)
  35568. landmarks are: tensor([[[ 0.5566, -0.4160, 0.9935, -1.3313, -0.2824, -1.5161, 0.2144,
  35569. 0.1253],
  35570. [ 0.6141, -0.4345, 1.6864, -0.0303, -0.2882, 0.1544, 0.9521,
  35571. 0.1982],
  35572. [ 0.5828, -0.4417, 1.2476, -1.3929, -0.1727, -1.5700, 0.4694,
  35573. -0.0248],
  35574. [ 0.5496, -0.4295, 1.0109, -1.2928, -0.2651, -1.5777, 0.1589,
  35575. 0.0261],
  35576. [-2.2859, -2.2859, 0.7222, -1.4930, -0.3921, -1.3698, 0.1404,
  35577. 0.1343],
  35578. [ 0.5772, -0.3913, 1.8480, -0.2459, -0.4326, 0.1929, 0.5374,
  35579. 0.4701],
  35580. [ 0.6085, -0.4084, 1.5536, -1.1466, -0.0746, -1.4853, 0.6298,
  35581. 0.0851],
  35582. [ 0.5515, -0.4129, 1.4785, -1.0080, -0.5192, -1.1004, 0.1034,
  35583. -0.0220]]], device='cuda:0')
  35584. loss_train_step before backward: tensor(0.0138, device='cuda:0', grad_fn=<MseLossBackward>)
  35585. loss_train_step after backward: tensor(0.0138, device='cuda:0', grad_fn=<MseLossBackward>)
  35586. loss_train: 0.3834594888612628
  35587. step: 29
  35588. running loss: 0.013222740995215958
  35589. Train Steps: 29/90 Loss: 0.0132 torch.Size([8, 600, 800])
  35590. torch.Size([8, 8])
  35591. tensor([[0.6198, 0.3997, 0.8582, 0.5361, 0.4117, 0.5016, 0.5942, 0.5134],
  35592. [0.6132, 0.4037, 0.6963, 0.2217, 0.4100, 0.1950, 0.5395, 0.5175],
  35593. [0.6173, 0.4114, 0.7325, 0.2500, 0.4213, 0.1917, 0.5338, 0.5700],
  35594. [0.6076, 0.3953, 0.8125, 0.3917, 0.3525, 0.4083, 0.5346, 0.4954],
  35595. [0.6163, 0.4006, 0.8788, 0.4683, 0.3663, 0.4883, 0.5887, 0.5017],
  35596. [0.6138, 0.4054, 0.8750, 0.4750, 0.4363, 0.5017, 0.5086, 0.5822],
  35597. [ nan, nan, 0.6412, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700],
  35598. [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350]],
  35599. device='cuda:0', dtype=torch.float64)
  35600. predictions are: tensor([[ 0.6854, -0.4110, 1.7386, 0.2029, -0.4442, 0.0123, 0.5923, 0.0250],
  35601. [ 0.4848, -0.4617, 1.0833, -1.2358, -0.3939, -1.4464, 0.3869, 0.0485],
  35602. [ 0.2746, -0.5979, 1.1664, -1.1019, -0.3426, -1.3615, 0.2957, 0.2825],
  35603. [ 0.5144, -0.4763, 1.6270, -0.4498, -0.6939, -0.4958, 0.4341, 0.0083],
  35604. [ 0.6393, -0.4014, 1.8480, -0.1163, -0.6229, -0.0050, 0.5811, 0.1108],
  35605. [ 0.6124, -0.3778, 1.7751, -0.0631, -0.2991, -0.0231, 0.2343, 0.2695],
  35606. [-2.0529, -2.1630, 0.8682, -1.3244, -0.3408, -1.5010, 0.1745, 0.3081],
  35607. [ 0.6295, -0.3719, 1.8619, -0.2724, -0.5806, -0.4473, 0.4335, 0.2207]],
  35608. device='cuda:0', grad_fn=<AddmmBackward>)
  35609. landmarks are: tensor([[[ 5.7673e-01, -4.3957e-01, 1.6782e+00, 1.9046e-01, -3.8437e-01,
  35610. 3.0841e-02, 4.5876e-01, 8.5521e-02],
  35611. [ 5.4648e-01, -4.2140e-01, 9.3002e-01, -1.2620e+00, -3.9215e-01,
  35612. -1.3852e+00, 2.0618e-01, 1.0428e-01],
  35613. [ 5.6518e-01, -3.8584e-01, 1.0975e+00, -1.1312e+00, -3.4018e-01,
  35614. -1.4006e+00, 1.7945e-01, 3.4688e-01],
  35615. [ 5.2061e-01, -4.6028e-01, 1.4670e+00, -4.7683e-01, -6.5774e-01,
  35616. -3.9985e-01, 1.8356e-01, 2.0831e-03],
  35617. [ 5.6091e-01, -4.3541e-01, 1.7730e+00, -1.2271e-01, -5.9423e-01,
  35618. -3.0331e-02, 4.3349e-01, 3.1255e-02],
  35619. [ 5.4908e-01, -4.1324e-01, 1.7557e+00, -9.1917e-02, -2.7090e-01,
  35620. 3.1255e-02, 6.3480e-02, 4.0319e-01],
  35621. [-2.2859e+00, -2.2859e+00, 6.7598e-01, -1.4083e+00, -3.2864e-01,
  35622. -1.4160e+00, 2.4873e-01, 3.4688e-01],
  35623. [ 6.0722e-01, -3.2394e-01, 1.8423e+00, -3.5366e-01, -4.9607e-01,
  35624. -3.9215e-01, 2.0831e-01, 1.8522e-01]]], device='cuda:0')
  35625. loss_train_step before backward: tensor(0.0104, device='cuda:0', grad_fn=<MseLossBackward>)
  35626. loss_train_step after backward: tensor(0.0104, device='cuda:0', grad_fn=<MseLossBackward>)
  35627. loss_train: 0.3938441062346101
  35628. step: 30
  35629. running loss: 0.013128136874487002
  35630. Train Steps: 30/90 Loss: 0.0131 torch.Size([8, 600, 800])
  35631. torch.Size([8, 8])
  35632. tensor([[0.6286, 0.4034, 0.8191, 0.2414, 0.4262, 0.2393, 0.6365, 0.5356],
  35633. [0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343],
  35634. [ nan, nan, 0.8363, 0.3317, 0.3563, 0.3367, 0.5329, 0.5142],
  35635. [0.6087, 0.3951, 0.8387, 0.5833, 0.4188, 0.4933, 0.5146, 0.4830],
  35636. [ nan, nan, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999],
  35637. [0.6339, 0.4159, 0.8400, 0.5617, 0.3825, 0.4150, 0.7343, 0.5748],
  35638. [0.6167, 0.4048, 0.6831, 0.3639, 0.3763, 0.3017, 0.5700, 0.5883],
  35639. [0.6176, 0.3911, 0.8738, 0.4217, 0.3488, 0.4033, 0.6025, 0.4817]],
  35640. device='cuda:0', dtype=torch.float64)
  35641. predictions are: tensor([[ 0.7876, -0.2612, 1.4754, -1.2204, -0.3086, -1.2458, 0.6964, 0.1785],
  35642. [ 0.6606, -0.3845, 1.8579, -0.6968, -0.3384, -0.8934, 0.8674, 0.1214],
  35643. [-1.1621, -1.5742, 1.4177, -0.9740, -0.6736, -0.8677, 0.1271, 0.1619],
  35644. [ 0.8018, -0.2644, 1.5804, 0.3024, -0.3906, -0.0824, 0.0951, 0.0740],
  35645. [-1.5449, -1.8026, 1.1856, -1.3089, -0.3308, -1.4579, 0.0138, 0.1254],
  35646. [ 0.8893, -0.2169, 1.6289, 0.1542, -0.5249, -0.3488, 0.8478, 0.2615],
  35647. [ 0.5511, -0.3979, 1.0464, -0.8318, -0.5398, -0.8926, 0.2521, 0.5029],
  35648. [ 0.6511, -0.3953, 1.7895, -0.4373, -0.6443, -0.5113, 0.4277, 0.0741]],
  35649. device='cuda:0', grad_fn=<AddmmBackward>)
  35650. landmarks are: tensor([[[ 0.6174, -0.4225, 1.4975, -1.1709, -0.3174, -1.1806, 0.6539,
  35651. 0.1879],
  35652. [ 0.6125, -0.4369, 1.9173, -0.5384, -0.2594, -0.8386, 0.9741,
  35653. 0.1821],
  35654. [-2.2859, -2.2859, 1.5767, -0.7540, -0.6404, -0.7309, 0.1753,
  35655. 0.0893],
  35656. [ 0.5258, -0.4610, 1.5882, 0.4085, -0.3517, -0.0072, 0.0910,
  35657. -0.0550],
  35658. [-2.2859, -2.2859, 1.3400, -1.0388, -0.3055, -1.4930, 0.1157,
  35659. 0.0231],
  35660. [ 0.6421, -0.3647, 1.5940, 0.3084, -0.5192, -0.3691, 1.1057,
  35661. 0.3692],
  35662. [ 0.5626, -0.4162, 0.8692, -0.6051, -0.5480, -0.8925, 0.3469,
  35663. 0.4316],
  35664. [ 0.5669, -0.4794, 1.7499, -0.3383, -0.6751, -0.4229, 0.4970,
  35665. -0.0611]]], device='cuda:0')
  35666. loss_train_step before backward: tensor(0.0529, device='cuda:0', grad_fn=<MseLossBackward>)
  35667. loss_train_step after backward: tensor(0.0529, device='cuda:0', grad_fn=<MseLossBackward>)
  35668. loss_train: 0.44675440434366465
  35669. step: 31
  35670. running loss: 0.01441143239818273
  35671. Train Steps: 31/90 Loss: 0.0144 torch.Size([8, 600, 800])
  35672. torch.Size([8, 8])
  35673. tensor([[0.6200, 0.4101, 0.8838, 0.5317, 0.3825, 0.3800, 0.5413, 0.5683],
  35674. [0.6202, 0.4053, 0.8638, 0.5283, 0.4546, 0.5108, 0.5900, 0.5133],
  35675. [0.6167, 0.4048, 0.6831, 0.3639, 0.3763, 0.3017, 0.5700, 0.5883],
  35676. [0.6197, 0.4090, 0.7825, 0.2500, 0.4200, 0.2483, 0.5988, 0.5667],
  35677. [0.6271, 0.4040, 0.9138, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413],
  35678. [0.6131, 0.4037, 0.6907, 0.2819, 0.3688, 0.2700, 0.5217, 0.5680],
  35679. [0.6200, 0.3913, 0.8788, 0.5217, 0.4075, 0.5100, 0.6060, 0.4913],
  35680. [0.6185, 0.4080, 0.8625, 0.3483, 0.3788, 0.2650, 0.5320, 0.5272]],
  35681. device='cuda:0', dtype=torch.float64)
  35682. predictions are: tensor([[ 0.6828, -0.3442, 1.7467, 0.0619, -0.5904, -0.5502, 0.3114, 0.3808],
  35683. [ 0.6513, -0.3840, 1.7281, 0.0559, -0.2999, 0.0749, 0.4490, 0.1396],
  35684. [ 0.3418, -0.5646, 1.0523, -0.8296, -0.6099, -0.9134, 0.3042, 0.4728],
  35685. [ 0.4111, -0.5128, 1.3451, -1.2935, -0.4238, -1.2114, 0.4419, 0.3051],
  35686. [ 0.6906, -0.3634, 2.1027, -0.7261, -0.1946, -1.2085, 1.1692, 0.1020],
  35687. [ 0.2817, -0.5988, 0.9430, -1.1393, -0.5916, -1.0603, 0.1292, 0.2920],
  35688. [ 0.5299, -0.5242, 1.7526, 0.0204, -0.4859, 0.0421, 0.3819, -0.0368],
  35689. [ 0.6846, -0.3526, 1.7553, -0.8015, -0.5090, -1.1177, 0.3066, 0.1300]],
  35690. device='cuda:0', grad_fn=<AddmmBackward>)
  35691. landmarks are: tensor([[[ 0.5777, -0.3915, 1.7961, 0.1698, -0.5192, -0.5307, 0.2141,
  35692. 0.3392],
  35693. [ 0.5786, -0.4141, 1.7037, 0.1544, -0.1862, 0.0736, 0.4393,
  35694. 0.0851],
  35695. [ 0.5626, -0.4162, 0.8692, -0.6051, -0.5480, -0.8925, 0.3469,
  35696. 0.4316],
  35697. [ 0.5764, -0.3969, 1.3284, -1.1312, -0.3460, -1.1389, 0.4797,
  35698. 0.3315],
  35699. [ 0.6108, -0.4201, 1.9346, -0.5538, -0.1497, -1.0773, 1.0545,
  35700. 0.2142],
  35701. [ 0.5459, -0.4215, 0.9043, -0.9838, -0.5827, -1.0388, 0.1236,
  35702. 0.3378],
  35703. [ 0.5776, -0.4784, 1.7730, 0.1236, -0.4037, 0.0697, 0.5132,
  35704. -0.0168],
  35705. [ 0.5711, -0.4015, 1.6979, -0.6770, -0.5365, -1.0619, 0.1712,
  35706. 0.1494]]], device='cuda:0')
  35707. loss_train_step before backward: tensor(0.0103, device='cuda:0', grad_fn=<MseLossBackward>)
  35708. loss_train_step after backward: tensor(0.0103, device='cuda:0', grad_fn=<MseLossBackward>)
  35709. loss_train: 0.4570535933598876
  35710. step: 32
  35711. running loss: 0.014282924792496487
  35712.  
  35713. Train Steps: 32/90 Loss: 0.0143 torch.Size([8, 600, 800])
  35714. torch.Size([8, 8])
  35715. tensor([[0.6245, 0.4115, 0.8700, 0.4883, 0.4625, 0.5517, 0.6100, 0.5217],
  35716. [0.6079, 0.3964, 0.7420, 0.2958, 0.3563, 0.2917, 0.5351, 0.4980],
  35717. [ nan, nan, 0.8363, 0.3317, 0.3563, 0.3367, 0.5329, 0.5142],
  35718. [0.6156, 0.4125, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084],
  35719. [0.6112, 0.4029, 0.8638, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567],
  35720. [0.6199, 0.4060, 0.8888, 0.4667, 0.3800, 0.5050, 0.6188, 0.5433],
  35721. [0.6219, 0.4114, 0.8175, 0.2817, 0.3925, 0.2783, 0.5900, 0.5350],
  35722. [0.6229, 0.4198, 0.7662, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783]],
  35723. device='cuda:0', dtype=torch.float64)
  35724. predictions are: tensor([[ 6.4178e-01, -3.5956e-01, 1.7213e+00, -5.5002e-03, -2.6230e-01,
  35725. 1.4525e-02, 5.9712e-01, 2.7523e-01],
  35726. [ 4.7999e-01, -4.8110e-01, 1.1534e+00, -9.4995e-01, -6.3442e-01,
  35727. -1.0915e+00, 2.5093e-01, -2.0075e-03],
  35728. [-1.9991e+00, -2.1325e+00, 1.3982e+00, -9.4794e-01, -7.2812e-01,
  35729. -9.9760e-01, 1.5390e-01, 1.4590e-01],
  35730. [ 5.6458e-01, -3.8236e-01, 1.7374e+00, -7.4091e-02, -2.8748e-01,
  35731. 1.2523e-01, 3.6424e-01, 2.4052e-01],
  35732. [ 4.8428e-01, -4.9028e-01, 1.7524e+00, -8.4647e-02, -1.1400e-01,
  35733. -2.1300e-01, 4.6986e-01, 2.6927e-01],
  35734. [ 5.5946e-01, -4.2733e-01, 1.7709e+00, -1.3355e-01, -5.6999e-01,
  35735. -1.6277e-01, 5.4718e-01, 2.3423e-01],
  35736. [ 4.1153e-01, -5.0251e-01, 1.4305e+00, -1.0455e+00, -5.0402e-01,
  35737. -1.1267e+00, 5.4868e-01, 2.4379e-01],
  35738. [ 4.6093e-01, -4.3216e-01, 1.2042e+00, -1.0118e+00, -2.2142e-01,
  35739. -1.4587e+00, 3.7473e-01, 3.7239e-01]], device='cuda:0',
  35740. grad_fn=<AddmmBackward>)
  35741. landmarks are: tensor([[[ 0.5987, -0.3852, 1.7326, -0.0303, -0.1497, 0.2622, 0.5316,
  35742. 0.1236],
  35743. [ 0.5219, -0.4550, 1.1415, -0.9196, -0.6404, -0.9387, 0.1856,
  35744. 0.0141],
  35745. [-2.2859, -2.2859, 1.5767, -0.7540, -0.6404, -0.7309, 0.1753,
  35746. 0.0893],
  35747. [ 0.5573, -0.3808, 1.8018, -0.0534, -0.1843, 0.4008, 0.3769,
  35748. 0.0622],
  35749. [ 0.5371, -0.4252, 1.7037, -0.0688, -0.0342, 0.0620, 0.3758,
  35750. 0.2853],
  35751. [ 0.5773, -0.4105, 1.8192, -0.1304, -0.5307, 0.0467, 0.5721,
  35752. 0.2237],
  35753. [ 0.5868, -0.3858, 1.4901, -0.9849, -0.4730, -1.0003, 0.4393,
  35754. 0.1852],
  35755. [ 0.5912, -0.3467, 1.2533, -1.0388, -0.1150, -1.3005, 0.3353,
  35756. 0.3854]]], device='cuda:0')
  35757. loss_train_step before backward: tensor(0.0128, device='cuda:0', grad_fn=<MseLossBackward>)
  35758. loss_train_step after backward: tensor(0.0128, device='cuda:0', grad_fn=<MseLossBackward>)
  35759. loss_train: 0.46980485040694475
  35760. step: 33
  35761. running loss: 0.014236510618392265
  35762. Train Steps: 33/90 Loss: 0.0142 torch.Size([8, 600, 800])
  35763. torch.Size([8, 8])
  35764. tensor([[0.6265, 0.4071, 0.8875, 0.3367, 0.3975, 0.3350, 0.6312, 0.5250],
  35765. [0.6148, 0.4053, 0.8750, 0.4550, 0.4850, 0.5218, 0.5863, 0.5567],
  35766. [0.6222, 0.4108, 0.8938, 0.4233, 0.3600, 0.3817, 0.5825, 0.5283],
  35767. [0.6277, 0.4029, 0.8250, 0.2433, 0.4325, 0.2100, 0.6366, 0.5207],
  35768. [0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204],
  35769. [0.6112, 0.4029, 0.8638, 0.4800, 0.4875, 0.5083, 0.5763, 0.5567],
  35770. [0.6276, 0.4235, 0.8888, 0.5333, 0.3800, 0.3117, 0.5427, 0.6164],
  35771. [0.6126, 0.3954, 0.8538, 0.4983, 0.4250, 0.4700, 0.5355, 0.5350]],
  35772. device='cuda:0', dtype=torch.float64)
  35773. predictions are: tensor([[ 0.6651, -0.3525, 1.8387, -0.8408, -0.5424, -0.8504, 0.6656, 0.1439],
  35774. [ 0.4286, -0.4925, 1.7124, -0.2640, -0.1381, -0.0479, 0.2317, 0.3010],
  35775. [ 0.7453, -0.2869, 1.7095, -0.4168, -0.7020, -0.5912, 0.4953, 0.2396],
  35776. [ 0.3159, -0.6062, 1.4369, -1.2623, -0.3676, -1.3907, 0.5356, 0.0811],
  35777. [ 0.6956, -0.3403, 1.5745, 0.1918, -0.4717, 0.0982, 0.7810, 0.1632],
  35778. [ 0.3613, -0.5702, 1.7576, -0.1592, -0.1135, -0.0804, 0.3595, 0.2578],
  35779. [ 0.6153, -0.3244, 1.6121, 0.1522, -0.5642, -0.8442, 0.2849, 0.5926],
  35780. [ 0.5132, -0.4749, 1.6085, -0.0092, -0.4199, -0.1877, 0.1662, 0.1904]],
  35781. device='cuda:0', grad_fn=<AddmmBackward>)
  35782. landmarks are: tensor([[[ 0.6078, -0.4057, 1.8134, -0.7309, -0.4499, -0.7386, 0.6298,
  35783. 0.1390],
  35784. [ 0.5538, -0.4139, 1.7557, -0.1843, -0.0459, 0.1242, 0.4219,
  35785. 0.2853],
  35786. [ 0.5880, -0.3887, 1.8423, -0.3306, -0.6231, -0.5230, 0.4046,
  35787. 0.1544],
  35788. [ 0.6132, -0.4248, 1.5247, -1.1620, -0.2882, -1.3159, 0.6545,
  35789. 0.1193],
  35790. [ 0.6042, -0.4273, 1.7198, 0.2184, -0.3478, 0.1149, 0.8062,
  35791. 0.1176],
  35792. [ 0.5371, -0.4252, 1.7037, -0.0688, -0.0342, 0.0620, 0.3758,
  35793. 0.2853],
  35794. [ 0.6131, -0.3299, 1.8192, 0.1775, -0.5307, -0.8463, 0.2206,
  35795. 0.5612],
  35796. [ 0.5435, -0.4597, 1.6575, 0.0159, -0.3229, -0.1150, 0.1877,
  35797. 0.1854]]], device='cuda:0')
  35798. loss_train_step before backward: tensor(0.0096, device='cuda:0', grad_fn=<MseLossBackward>)
  35799. loss_train_step after backward: tensor(0.0096, device='cuda:0', grad_fn=<MseLossBackward>)
  35800. loss_train: 0.4794104462489486
  35801. step: 34
  35802. running loss: 0.014100307242616135
  35803. Train Steps: 34/90 Loss: 0.0141 torch.Size([8, 600, 800])
  35804. torch.Size([8, 8])
  35805. tensor([[0.6241, 0.4143, 0.8938, 0.4650, 0.4075, 0.5350, 0.6250, 0.5550],
  35806. [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083],
  35807. [0.6226, 0.4103, 0.8575, 0.3450, 0.4388, 0.2067, 0.5787, 0.5383],
  35808. [0.6064, 0.4019, 0.8650, 0.4517, 0.4037, 0.5367, 0.5703, 0.5609],
  35809. [ nan, nan, 0.8850, 0.3000, 0.5363, 0.2250, 0.7343, 0.5771],
  35810. [0.6201, 0.4050, 0.7757, 0.2234, 0.4459, 0.1798, 0.5975, 0.5426],
  35811. [0.6109, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567],
  35812. [ nan, nan, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729]],
  35813. device='cuda:0', dtype=torch.float64)
  35814. predictions are: tensor([[ 0.7850, -0.2571, 1.8136, -0.0547, -0.3984, 0.1662, 0.5929, 0.2880],
  35815. [ 0.7754, -0.2955, 1.7141, 0.1585, -0.6337, -0.1329, 0.4963, 0.4172],
  35816. [ 0.7866, -0.2281, 1.6788, -0.6753, -0.2686, -1.2191, 0.3706, 0.1885],
  35817. [ 0.6878, -0.3163, 1.6834, -0.0576, -0.4463, 0.3079, 0.4258, 0.2013],
  35818. [-1.5394, -1.7654, 1.7289, -0.9718, 0.0645, -1.2452, 0.7936, 0.3608],
  35819. [ 0.6353, -0.3698, 1.2097, -1.1548, -0.2937, -1.3861, 0.3841, 0.0950],
  35820. [ 0.6473, -0.3535, 0.9732, -1.2908, -0.5212, -1.0675, 0.3539, 0.1694],
  35821. [-1.8065, -1.9688, 0.9073, -1.2550, -0.4676, -1.4095, -0.0827, 0.2193]],
  35822. device='cuda:0', grad_fn=<AddmmBackward>)
  35823. landmarks are: tensor([[[ 0.5968, -0.3725, 1.8423, -0.1381, -0.4037, 0.1852, 0.6009,
  35824. 0.2776],
  35825. [ 0.5713, -0.4105, 1.7557, 0.0467, -0.6520, -0.2690, 0.3988,
  35826. 0.5239],
  35827. [ 0.5898, -0.3908, 1.6748, -0.6924, -0.2594, -1.3313, 0.3873,
  35828. 0.2006],
  35829. [ 0.5151, -0.4296, 1.7095, -0.1997, -0.4210, 0.1929, 0.3484,
  35830. 0.3047],
  35831. [-2.2859, -2.2859, 1.8018, -0.9002, 0.1910, -1.2467, 1.1057,
  35832. 0.3799],
  35833. [ 0.5784, -0.4153, 1.2972, -1.2541, -0.2265, -1.4553, 0.4741,
  35834. 0.2203],
  35835. [ 0.5356, -0.4217, 1.0339, -1.4776, -0.5076, -1.1081, 0.4219,
  35836. 0.2853],
  35837. [-2.2859, -2.2859, 0.8824, -1.2727, -0.3691, -1.4237, 0.0943,
  35838. 0.3604]]], device='cuda:0')
  35839. loss_train_step before backward: tensor(0.0284, device='cuda:0', grad_fn=<MseLossBackward>)
  35840. loss_train_step after backward: tensor(0.0284, device='cuda:0', grad_fn=<MseLossBackward>)
  35841. loss_train: 0.5077718859538436
  35842. step: 35
  35843. running loss: 0.014507768170109818
  35844. Train Steps: 35/90 Loss: 0.0145 torch.Size([8, 600, 800])
  35845. torch.Size([8, 8])
  35846. tensor([[0.6102, 0.3999, 0.8750, 0.5133, 0.3825, 0.4750, 0.5637, 0.5083],
  35847. [0.6192, 0.4128, 0.8513, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633],
  35848. [0.6260, 0.4214, 0.8538, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983],
  35849. [0.6102, 0.4020, 0.8638, 0.3717, 0.3625, 0.5017, 0.6038, 0.5500],
  35850. [0.6161, 0.4055, 0.8675, 0.3867, 0.3713, 0.4033, 0.5195, 0.5162],
  35851. [0.6144, 0.4032, 0.8563, 0.3283, 0.3525, 0.4200, 0.5775, 0.5583],
  35852. [0.6286, 0.4078, 0.8063, 0.2267, 0.4788, 0.1533, 0.5953, 0.4913],
  35853. [0.6204, 0.4055, 0.8438, 0.5733, 0.4574, 0.4801, 0.5487, 0.5617]],
  35854. device='cuda:0', dtype=torch.float64)
  35855. predictions are: tensor([[ 0.5011, -0.4467, 1.7422, 0.0539, -0.4417, -0.0660, 0.3217, 0.0671],
  35856. [ 0.5066, -0.4254, 1.6587, 0.2854, -0.3504, 0.1462, 0.4665, 0.3451],
  35857. [ 0.6545, -0.3217, 1.6254, 0.3260, -0.5526, -0.4922, 0.4399, 0.5244],
  35858. [ 0.5252, -0.4480, 1.6938, -0.6012, -0.5887, -0.0047, 0.6975, 0.3762],
  35859. [ 0.2379, -0.6145, 1.7338, -0.5259, -0.5895, -0.5340, 0.2075, 0.1188],
  35860. [ 0.6845, -0.3041, 1.5760, -0.7952, -0.6356, -0.4136, 0.3729, 0.3453],
  35861. [ 0.4442, -0.4868, 1.3857, -1.2715, -0.0972, -1.6168, 0.4491, 0.0212],
  35862. [ 0.4995, -0.4227, 1.5819, 0.3630, -0.1269, -0.0922, 0.3253, 0.3093]],
  35863. device='cuda:0', grad_fn=<AddmmBackward>)
  35864. landmarks are: tensor([[[ 0.5326, -0.4390, 1.7557, 0.0851, -0.5192, -0.0919, 0.3180,
  35865. 0.0620],
  35866. [ 0.5742, -0.3792, 1.6460, 0.3084, -0.3460, 0.1467, 0.4162,
  35867. 0.3161],
  35868. [ 0.6055, -0.3393, 1.6575, 0.2545, -0.5942, -0.5461, 0.2949,
  35869. 0.4778],
  35870. [ 0.5324, -0.4293, 1.7037, -0.5692, -0.6115, 0.0313, 0.5028,
  35871. 0.2545],
  35872. [ 0.5598, -0.4129, 1.7210, -0.4999, -0.5711, -0.4229, 0.1136,
  35873. 0.0983],
  35874. [ 0.5520, -0.4237, 1.6691, -0.7694, -0.6577, -0.3460, 0.3815,
  35875. 0.2930],
  35876. [ 0.6177, -0.4022, 1.4381, -1.2390, -0.0746, -1.5777, 0.4639,
  35877. -0.0168],
  35878. [ 0.5795, -0.4129, 1.6113, 0.3623, -0.1733, -0.0684, 0.2487,
  35879. 0.3084]]], device='cuda:0')
  35880. loss_train_step before backward: tensor(0.0061, device='cuda:0', grad_fn=<MseLossBackward>)
  35881. loss_train_step after backward: tensor(0.0061, device='cuda:0', grad_fn=<MseLossBackward>)
  35882. loss_train: 0.5138542065396905
  35883. step: 36
  35884. running loss: 0.014273727959435847
  35885.  
  35886. Train Steps: 36/90 Loss: 0.0143 torch.Size([8, 600, 800])
  35887. torch.Size([8, 8])
  35888. tensor([[0.6201, 0.4151, 0.8588, 0.5467, 0.3700, 0.3950, 0.5637, 0.5933],
  35889. [0.6364, 0.4154, 0.8938, 0.3717, 0.4500, 0.2583, 0.6448, 0.5285],
  35890. [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5413, 0.5433],
  35891. [0.6205, 0.4062, 0.8337, 0.2683, 0.3675, 0.4283, 0.6338, 0.5250],
  35892. [0.6325, 0.4066, 0.8438, 0.5350, 0.3925, 0.5267, 0.7113, 0.5795],
  35893. [0.6332, 0.4118, 0.9238, 0.4267, 0.4012, 0.4733, 0.7525, 0.5436],
  35894. [0.6200, 0.4024, 0.8390, 0.3139, 0.3525, 0.3833, 0.6162, 0.5383],
  35895. [0.6125, 0.3974, 0.7725, 0.2517, 0.3538, 0.3317, 0.5887, 0.5500]],
  35896. device='cuda:0', dtype=torch.float64)
  35897. predictions are: tensor([[ 0.7667, -0.1900, 1.5827, 0.3425, -0.4211, -0.4493, 0.1190, 0.4417],
  35898. [ 0.6480, -0.3304, 1.8848, -0.5738, -0.0757, -1.2094, 0.5862, 0.2134],
  35899. [ 0.4521, -0.4451, 1.6456, -0.5933, -0.4995, -0.8002, -0.0086, 0.1792],
  35900. [ 0.3198, -0.5839, 1.5540, -0.9149, -0.5333, -0.2930, 0.5008, 0.2137],
  35901. [ 0.6390, -0.3685, 1.5938, 0.3059, -0.3143, 0.2177, 0.8638, 0.3779],
  35902. [ 0.6208, -0.3873, 1.8751, -0.2567, -0.3241, 0.0175, 0.9235, 0.2965],
  35903. [ 0.2887, -0.6082, 1.6301, -0.7104, -0.5966, -0.4797, 0.4536, 0.2508],
  35904. [ 0.2123, -0.6332, 1.2361, -1.0201, -0.5911, -0.7614, 0.2279, 0.2410]],
  35905. device='cuda:0', grad_fn=<AddmmBackward>)
  35906. landmarks are: tensor([[[ 0.5785, -0.3687, 1.6806, 0.2391, -0.5769, -0.4614, 0.3180,
  35907. 0.4547],
  35908. [ 0.6537, -0.3671, 1.8423, -0.5692, -0.2074, -1.0927, 0.6926,
  35909. 0.1554],
  35910. [ 0.5894, -0.3479, 1.7730, -0.6847, -0.5538, -0.7155, 0.2141,
  35911. 0.2237],
  35912. [ 0.5799, -0.4099, 1.5651, -1.0465, -0.5885, -0.3075, 0.6413,
  35913. 0.1390],
  35914. [ 0.6355, -0.4080, 1.6113, 0.1852, -0.4730, 0.1467, 0.9996,
  35915. 0.3905],
  35916. [ 0.6388, -0.3840, 1.9808, -0.3152, -0.4326, -0.0996, 1.1897,
  35917. 0.2249],
  35918. [ 0.5779, -0.4275, 1.5894, -0.8362, -0.6577, -0.5153, 0.5605,
  35919. 0.2006],
  35920. [ 0.5430, -0.4503, 1.2822, -1.1235, -0.6520, -0.7540, 0.4335,
  35921. 0.2545]]], device='cuda:0')
  35922. loss_train_step before backward: tensor(0.0156, device='cuda:0', grad_fn=<MseLossBackward>)
  35923. loss_train_step after backward: tensor(0.0156, device='cuda:0', grad_fn=<MseLossBackward>)
  35924. loss_train: 0.5294085452333093
  35925. step: 37
  35926. running loss: 0.01430833906035971
  35927. Train Steps: 37/90 Loss: 0.0143 torch.Size([8, 600, 800])
  35928. torch.Size([8, 8])
  35929. tensor([[0.6182, 0.3972, 0.8552, 0.5914, 0.3683, 0.4181, 0.5688, 0.5378],
  35930. [0.6229, 0.4198, 0.7662, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783],
  35931. [0.6109, 0.4036, 0.7188, 0.1750, 0.3850, 0.2550, 0.5863, 0.5567],
  35932. [0.6164, 0.4119, 0.7913, 0.2650, 0.3538, 0.3500, 0.5614, 0.5038],
  35933. [ nan, nan, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633],
  35934. [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483],
  35935. [0.6224, 0.3964, 0.8225, 0.5717, 0.4150, 0.4617, 0.5775, 0.5267],
  35936. [ nan, nan, 0.7648, 0.2722, 0.3962, 0.2183, 0.5060, 0.5422]],
  35937. device='cuda:0', dtype=torch.float64)
  35938. predictions are: tensor([[ 0.8430, -0.2584, 1.6744, 0.3633, -0.4802, -0.2443, 0.5070, 0.2324],
  35939. [ 0.6614, -0.2980, 1.3332, -0.9772, -0.0668, -1.1946, 0.3990, 0.4166],
  35940. [ 0.6807, -0.3319, 1.1142, -1.3562, -0.3954, -1.0487, 0.4462, 0.3111],
  35941. [ 0.4883, -0.4645, 1.4983, -1.0006, -0.5738, -0.6161, 0.4483, 0.1922],
  35942. [-2.2025, -2.2211, 1.3382, -0.9761, -0.4328, -0.9994, 0.1234, 0.2760],
  35943. [ 0.7368, -0.3126, 1.1983, -1.3563, -0.3093, -1.0961, 0.5613, 0.3036],
  35944. [ 0.7644, -0.3089, 1.6260, 0.3014, -0.3090, -0.0725, 0.5097, 0.2114],
  35945. [-2.1184, -2.1461, 1.3030, -1.0153, -0.4229, -1.1399, 0.1310, 0.2134]],
  35946. device='cuda:0', grad_fn=<AddmmBackward>)
  35947. landmarks are: tensor([[[ 0.5697, -0.4514, 1.6642, 0.4459, -0.5850, -0.3546, 0.3414,
  35948. 0.1982],
  35949. [ 0.5912, -0.3467, 1.2533, -1.0388, -0.1150, -1.3005, 0.3353,
  35950. 0.3854],
  35951. [ 0.5356, -0.4217, 1.0339, -1.4776, -0.5076, -1.1081, 0.4219,
  35952. 0.2853],
  35953. [ 0.5614, -0.3834, 1.3688, -1.0619, -0.6520, -0.6693, 0.3069,
  35954. 0.0412],
  35955. [-2.2859, -2.2859, 1.2030, -1.0288, -0.4961, -1.1081, 0.0813,
  35956. 0.3161],
  35957. [ 0.5784, -0.4085, 1.0859, -1.3929, -0.4037, -1.1158, 0.5605,
  35958. 0.2468],
  35959. [ 0.5891, -0.4550, 1.5132, 0.3546, -0.3691, -0.1535, 0.3815,
  35960. 0.1467],
  35961. [-2.2859, -2.2859, 1.2469, -1.0288, -0.4557, -1.2774, 0.0511,
  35962. 0.2183]]], device='cuda:0')
  35963. loss_train_step before backward: tensor(0.0101, device='cuda:0', grad_fn=<MseLossBackward>)
  35964. loss_train_step after backward: tensor(0.0101, device='cuda:0', grad_fn=<MseLossBackward>)
  35965. loss_train: 0.5395098840817809
  35966. step: 38
  35967. running loss: 0.014197628528467919
  35968. Train Steps: 38/90 Loss: 0.0142 torch.Size([8, 600, 800])
  35969. torch.Size([8, 8])
  35970. tensor([[0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879],
  35971. [0.6102, 0.4005, 0.8688, 0.5100, 0.4813, 0.5400, 0.5404, 0.5064],
  35972. [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351],
  35973. [0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967],
  35974. [0.6193, 0.4165, 0.8838, 0.4700, 0.4150, 0.4867, 0.5427, 0.5261],
  35975. [0.6269, 0.4073, 0.8900, 0.4933, 0.3625, 0.4183, 0.6288, 0.5150],
  35976. [0.6107, 0.4013, 0.8700, 0.4650, 0.5049, 0.5176, 0.5850, 0.5550],
  35977. [ nan, nan, 0.8488, 0.2300, 0.5563, 0.2100, 0.7390, 0.5679]],
  35978. device='cuda:0', dtype=torch.float64)
  35979. predictions are: tensor([[ 0.5074, -0.4429, 1.7396, -0.2308, -0.5302, 0.2456, 0.5146, 0.1426],
  35980. [ 0.4840, -0.4596, 1.6347, 0.0541, -0.1286, 0.0975, 0.1633, 0.1393],
  35981. [ 0.8040, -0.2417, 1.2190, -1.0917, -0.2926, -1.5079, 0.3536, 0.2265],
  35982. [ 0.4214, -0.4987, 1.6166, -0.2834, -0.6564, 0.0112, 0.3598, 0.2051],
  35983. [ 0.5247, -0.3758, 1.7237, -0.0731, -0.4058, -0.1434, 0.2149, 0.2700],
  35984. [ 0.5961, -0.3832, 1.7580, 0.0120, -0.6258, -0.3360, 0.6261, 0.2390],
  35985. [ 0.6072, -0.3421, 1.6672, -0.1749, -0.0391, -0.0249, 0.3669, 0.3491],
  35986. [-2.1279, -2.1510, 1.6978, -1.1259, 0.1169, -1.2127, 1.0293, 0.4119]],
  35987. device='cuda:0', grad_fn=<AddmmBackward>)
  35988. landmarks are: tensor([[[ 0.5676, -0.4327, 1.8252, -0.2095, -0.4886, 0.3084, 0.4727,
  35989. -0.0322],
  35990. [ 0.5328, -0.4361, 1.7268, 0.0697, -0.0630, 0.2083, 0.2103,
  35991. 0.0532],
  35992. [ 0.5664, -0.4321, 1.2862, -1.0003, -0.2189, -1.4608, 0.3883,
  35993. 0.1855],
  35994. [ 0.4974, -0.4482, 1.6633, -0.3306, -0.6173, 0.1313, 0.2925,
  35995. 0.0081],
  35996. [ 0.5746, -0.3623, 1.7961, -0.1150, -0.3691, -0.0380, 0.2208,
  35997. 0.1439],
  35998. [ 0.6096, -0.4046, 1.8249, -0.0072, -0.6115, -0.3537, 0.6182,
  35999. 0.0928],
  36000. [ 0.5351, -0.4321, 1.7326, -0.1381, 0.0463, 0.1049, 0.4162,
  36001. 0.2776],
  36002. [-2.2859, -2.2859, 1.6344, -1.2236, 0.2834, -1.3159, 1.1276,
  36003. 0.3371]]], device='cuda:0')
  36004. loss_train_step before backward: tensor(0.0078, device='cuda:0', grad_fn=<MseLossBackward>)
  36005. loss_train_step after backward: tensor(0.0078, device='cuda:0', grad_fn=<MseLossBackward>)
  36006. loss_train: 0.5473534110933542
  36007. step: 39
  36008. running loss: 0.014034702848547544
  36009. Train Steps: 39/90 Loss: 0.0140 torch.Size([8, 600, 800])
  36010. torch.Size([8, 8])
  36011. tensor([[0.6214, 0.4112, 0.7838, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083],
  36012. [0.6124, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485],
  36013. [0.6059, 0.4002, 0.7562, 0.2767, 0.3538, 0.3033, 0.5529, 0.5455],
  36014. [0.6156, 0.4125, 0.8850, 0.4833, 0.4550, 0.5817, 0.5765, 0.5084],
  36015. [0.6193, 0.4079, 0.7288, 0.2500, 0.4250, 0.2550, 0.5989, 0.6266],
  36016. [0.6222, 0.4169, 0.8638, 0.5650, 0.4313, 0.4783, 0.5637, 0.5633],
  36017. [0.6343, 0.4097, 0.9287, 0.4367, 0.4313, 0.3600, 0.7248, 0.5841],
  36018. [0.6268, 0.4094, 0.9038, 0.4300, 0.3700, 0.3483, 0.6262, 0.5167]],
  36019. device='cuda:0', dtype=torch.float64)
  36020. predictions are: tensor([[ 5.9239e-01, -4.0632e-01, 1.4173e+00, -1.3067e+00, -5.8054e-01,
  36021. -8.9081e-01, 4.0951e-01, 3.0051e-02],
  36022. [ 5.6247e-01, -3.9829e-01, 1.0312e+00, -8.5208e-01, -5.4845e-01,
  36023. -9.3977e-01, 1.0921e-01, 2.4333e-01],
  36024. [ 1.8700e-01, -6.6109e-01, 1.2660e+00, -9.8780e-01, -5.7842e-01,
  36025. -8.6583e-01, 2.8633e-01, 2.2913e-01],
  36026. [ 9.6982e-02, -6.9610e-01, 1.8964e+00, -3.7301e-02, -1.7227e-01,
  36027. 4.4360e-01, 2.9462e-01, 1.1563e-01],
  36028. [ 6.4631e-01, -3.3672e-01, 1.1264e+00, -1.1566e+00, -2.4963e-01,
  36029. -1.0645e+00, 4.9173e-01, 6.1092e-01],
  36030. [ 3.9081e-01, -4.9535e-01, 1.7839e+00, 4.0092e-01, -2.1903e-01,
  36031. -9.6561e-04, 2.6731e-01, 3.3483e-01],
  36032. [ 6.0249e-01, -4.3258e-01, 1.9517e+00, -2.0927e-01, -2.7999e-01,
  36033. -5.2175e-01, 1.0574e+00, 3.2433e-01],
  36034. [ 4.8791e-01, -4.7775e-01, 1.9963e+00, -2.2164e-01, -4.3256e-01,
  36035. -5.8975e-01, 5.4839e-01, 1.9789e-01]], device='cuda:0',
  36036. grad_fn=<AddmmBackward>)
  36037. landmarks are: tensor([[[ 0.5845, -0.3864, 1.3342, -1.3082, -0.6000, -0.8386, 0.3353,
  36038. 0.0620],
  36039. [ 0.5425, -0.3998, 0.9263, -0.8683, -0.6000, -1.0157, 0.0990,
  36040. 0.2476],
  36041. [ 0.5129, -0.4374, 1.2072, -1.0080, -0.6520, -0.8848, 0.2679,
  36042. 0.2335],
  36043. [ 0.5573, -0.3808, 1.8018, -0.0534, -0.1843, 0.4008, 0.3769,
  36044. 0.0622],
  36045. [ 0.5746, -0.4021, 1.0801, -1.1312, -0.3229, -1.1081, 0.4803,
  36046. 0.6084],
  36047. [ 0.5880, -0.3605, 1.7037, 0.3238, -0.2940, -0.0765, 0.3180,
  36048. 0.3161],
  36049. [ 0.6438, -0.3936, 2.0039, -0.2690, -0.2940, -0.6231, 1.0618,
  36050. 0.4119],
  36051. [ 0.6094, -0.3947, 1.8885, -0.2998, -0.5769, -0.6770, 0.6067,
  36052. 0.1005]]], device='cuda:0')
  36053. loss_train_step before backward: tensor(0.0117, device='cuda:0', grad_fn=<MseLossBackward>)
  36054. loss_train_step after backward: tensor(0.0117, device='cuda:0', grad_fn=<MseLossBackward>)
  36055. loss_train: 0.5590283134952188
  36056. step: 40
  36057. running loss: 0.01397570783738047
  36058.  
  36059. Train Steps: 40/90 Loss: 0.0140 torch.Size([8, 600, 800])
  36060. torch.Size([8, 8])
  36061. tensor([[0.6178, 0.4012, 0.8900, 0.4495, 0.3891, 0.5617, 0.5972, 0.4879],
  36062. [0.6139, 0.4019, 0.7137, 0.2150, 0.4375, 0.1533, 0.5293, 0.5006],
  36063. [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367],
  36064. [0.6201, 0.4151, 0.8588, 0.5467, 0.3700, 0.3950, 0.5637, 0.5933],
  36065. [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413],
  36066. [0.6153, 0.4119, 0.8463, 0.3833, 0.3600, 0.3200, 0.5106, 0.5563],
  36067. [0.6310, 0.4017, 0.8563, 0.5800, 0.3688, 0.4750, 0.6326, 0.5006],
  36068. [0.6223, 0.3990, 0.8037, 0.2800, 0.4000, 0.2283, 0.5864, 0.5208]],
  36069. device='cuda:0', dtype=torch.float64)
  36070. predictions are: tensor([[ 0.4518, -0.4911, 1.8600, -0.2607, -0.4129, 0.5143, 0.4992, 0.0568],
  36071. [-0.3478, -1.0317, 1.0285, -1.3610, -0.1993, -1.5180, 0.1647, 0.1402],
  36072. [ 0.6369, -0.3709, 1.4022, -0.9832, -0.4957, -0.6442, 0.4793, 0.2773],
  36073. [ 0.5346, -0.3425, 1.6216, 0.1690, -0.4701, -0.3177, 0.2371, 0.5411],
  36074. [ 0.7848, -0.2854, 1.9911, -0.3355, -0.1997, -0.7153, 1.0501, 0.2050],
  36075. [ 0.3135, -0.5431, 1.6285, -0.5310, -0.5323, -0.6793, 0.0580, 0.3620],
  36076. [ 0.3326, -0.6026, 1.6768, 0.3122, -0.5380, 0.0443, 0.6050, 0.0992],
  36077. [ 0.5161, -0.4441, 1.4498, -1.1177, -0.3233, -1.1196, 0.4261, 0.2058]],
  36078. device='cuda:0', grad_fn=<AddmmBackward>)
  36079. landmarks are: tensor([[[ 0.5676, -0.4327, 1.8252, -0.2095, -0.4886, 0.3084, 0.4727,
  36080. -0.0322],
  36081. [ 0.5496, -0.4295, 1.0109, -1.2928, -0.2651, -1.5777, 0.1589,
  36082. 0.0261],
  36083. [ 0.5697, -0.4442, 1.3529, -0.9515, -0.5774, -0.7801, 0.5253,
  36084. 0.1931],
  36085. [ 0.5785, -0.3687, 1.6806, 0.2391, -0.5769, -0.4614, 0.3180,
  36086. 0.4547],
  36087. [ 0.6224, -0.4345, 1.9404, -0.2921, -0.3171, -0.8771, 1.0655,
  36088. 0.2142],
  36089. [ 0.5561, -0.3834, 1.6229, -0.5153, -0.6231, -0.8079, 0.0727,
  36090. 0.2837],
  36091. [ 0.6286, -0.4303, 1.6691, 0.3931, -0.5827, -0.0919, 0.6359,
  36092. 0.0261],
  36093. [ 0.5885, -0.4429, 1.4266, -0.9926, -0.4383, -1.2313, 0.4228,
  36094. 0.1195]]], device='cuda:0')
  36095. loss_train_step before backward: tensor(0.0280, device='cuda:0', grad_fn=<MseLossBackward>)
  36096. loss_train_step after backward: tensor(0.0280, device='cuda:0', grad_fn=<MseLossBackward>)
  36097. loss_train: 0.5870048655197024
  36098. step: 41
  36099. running loss: 0.014317191841943962
  36100. Train Steps: 41/90 Loss: 0.0143 torch.Size([8, 600, 800])
  36101. torch.Size([8, 8])
  36102. tensor([[0.6200, 0.3993, 0.8519, 0.4923, 0.3962, 0.4717, 0.6013, 0.5433],
  36103. [0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058],
  36104. [0.6311, 0.3998, 0.7975, 0.5767, 0.3838, 0.4850, 0.7327, 0.5343],
  36105. [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123],
  36106. [0.6157, 0.4102, 0.8513, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890],
  36107. [0.6125, 0.4076, 0.8488, 0.3883, 0.3700, 0.3683, 0.5026, 0.5505],
  36108. [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650],
  36109. [0.6250, 0.4110, 0.7238, 0.2067, 0.4263, 0.1883, 0.5625, 0.5633]],
  36110. device='cuda:0', dtype=torch.float64)
  36111. predictions are: tensor([[ 0.5652, -0.4335, 1.8033, 0.0198, -0.3656, 0.0220, 0.6305, 0.2068],
  36112. [ 0.4282, -0.5123, 1.7713, -0.2921, -0.4786, 0.0357, 0.2857, 0.0134],
  36113. [ 0.5329, -0.4882, 1.5947, 0.2187, -0.4321, 0.0283, 1.0430, 0.1789],
  36114. [ 0.7813, -0.2962, 1.0969, -1.1801, -0.4818, -1.0343, 0.2377, 0.0690],
  36115. [ 0.2271, -0.6053, 1.7303, -0.5249, -0.5412, -0.5170, 0.1491, 0.3835],
  36116. [-0.0694, -0.8135, 1.7507, -0.5003, -0.5468, -0.5444, 0.1607, 0.2659],
  36117. [ 0.7761, -0.2279, 1.7220, -0.3759, -0.4916, -0.7065, 0.3742, 0.4468],
  36118. [ 0.3328, -0.5700, 1.1195, -1.3783, -0.1821, -1.4471, 0.4467, 0.2584]],
  36119. device='cuda:0', grad_fn=<AddmmBackward>)
  36120. landmarks are: tensor([[[ 0.5777, -0.4415, 1.6491, -0.0122, -0.4557, -0.1073, 0.4912,
  36121. 0.2237],
  36122. [ 0.5245, -0.4361, 1.5940, -0.2921, -0.5480, -0.0919, 0.2432,
  36123. 0.0502],
  36124. [ 0.6289, -0.4393, 1.3977, 0.3777, -0.5134, -0.0457, 1.0984,
  36125. 0.1821],
  36126. [ 0.5342, -0.4280, 1.0051, -1.0619, -0.5711, -1.0388, 0.2226,
  36127. 0.0802],
  36128. [ 0.5581, -0.3912, 1.6460, -0.5230, -0.6173, -0.5923, 0.0681,
  36129. 0.4348],
  36130. [ 0.5433, -0.4032, 1.6344, -0.4922, -0.5769, -0.5846, 0.0357,
  36131. 0.2567],
  36132. [ 0.5894, -0.3503, 1.6402, -0.3614, -0.5827, -0.7925, 0.3238,
  36133. 0.3238],
  36134. [ 0.6010, -0.3875, 1.0570, -1.3313, -0.3171, -1.4160, 0.3122,
  36135. 0.3161]]], device='cuda:0')
  36136. loss_train_step before backward: tensor(0.0210, device='cuda:0', grad_fn=<MseLossBackward>)
  36137. loss_train_step after backward: tensor(0.0210, device='cuda:0', grad_fn=<MseLossBackward>)
  36138. loss_train: 0.6079580811783671
  36139. step: 42
  36140. running loss: 0.014475192409008741
  36141. Train Steps: 42/90 Loss: 0.0145 torch.Size([8, 600, 800])
  36142. torch.Size([8, 8])
  36143. tensor([[ nan, nan, 0.6859, 0.2194, 0.4150, 0.1867, 0.5153, 0.5729],
  36144. [0.6201, 0.4102, 0.7288, 0.2417, 0.4150, 0.2383, 0.6100, 0.5500],
  36145. [0.6216, 0.4099, 0.7225, 0.2033, 0.4188, 0.2217, 0.5975, 0.5283],
  36146. [0.6153, 0.4119, 0.8463, 0.3833, 0.3600, 0.3200, 0.5106, 0.5563],
  36147. [ nan, nan, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563],
  36148. [0.6139, 0.4019, 0.7137, 0.2150, 0.4375, 0.1533, 0.5293, 0.5006],
  36149. [0.6224, 0.4097, 0.7438, 0.2267, 0.3850, 0.2850, 0.5988, 0.5250],
  36150. [0.6200, 0.3961, 0.8461, 0.5497, 0.4142, 0.4577, 0.5892, 0.5402]],
  36151. device='cuda:0', dtype=torch.float64)
  36152. predictions are: tensor([[-2.1093, -2.1695, 1.0704, -1.1241, -0.2829, -1.2708, 0.3400, 0.3245],
  36153. [ 0.9580, -0.1868, 1.1554, -1.1501, -0.3985, -1.0545, 0.5273, 0.2618],
  36154. [ 0.7270, -0.3383, 1.1407, -1.2378, -0.3266, -1.1439, 0.4433, 0.2077],
  36155. [ 0.6284, -0.3472, 1.6952, -0.4482, -0.5650, -0.6186, 0.1774, 0.2573],
  36156. [-2.4414, -2.3598, 1.4631, -0.6933, -0.5079, -0.7125, 0.2897, 0.2286],
  36157. [ 0.6782, -0.3799, 1.1089, -1.2278, -0.2241, -1.4500, 0.2807, 0.0738],
  36158. [ 0.7070, -0.3484, 1.2289, -1.1555, -0.4486, -0.8100, 0.5723, 0.2394],
  36159. [ 0.6015, -0.4189, 1.7105, 0.1990, -0.4040, 0.0405, 0.5011, 0.1426]],
  36160. device='cuda:0', grad_fn=<AddmmBackward>)
  36161. landmarks are: tensor([[[-2.2859, -2.2859, 0.8824, -1.2727, -0.3691, -1.4237, 0.0943,
  36162. 0.3604],
  36163. [ 0.5784, -0.3913, 1.0801, -1.1697, -0.3691, -1.1851, 0.5316,
  36164. 0.2545],
  36165. [ 0.5850, -0.3925, 1.0513, -1.3467, -0.3517, -1.2620, 0.4739,
  36166. 0.1544],
  36167. [ 0.5561, -0.3834, 1.6229, -0.5153, -0.6231, -0.8079, 0.0727,
  36168. 0.2837],
  36169. [-2.2859, -2.2859, 1.4006, -0.8105, -0.6115, -0.8232, 0.0419,
  36170. 0.2837],
  36171. [ 0.5496, -0.4295, 1.0109, -1.2928, -0.2651, -1.5777, 0.1589,
  36172. 0.0261],
  36173. [ 0.5891, -0.3937, 1.1494, -1.2390, -0.5076, -0.9695, 0.4797,
  36174. 0.1390],
  36175. [ 0.5780, -0.4565, 1.6221, 0.2532, -0.3728, -0.1718, 0.4357,
  36176. 0.2091]]], device='cuda:0')
  36177. loss_train_step before backward: tensor(0.0137, device='cuda:0', grad_fn=<MseLossBackward>)
  36178. loss_train_step after backward: tensor(0.0137, device='cuda:0', grad_fn=<MseLossBackward>)
  36179. loss_train: 0.621706354431808
  36180. step: 43
  36181. running loss: 0.014458287312367629
  36182. Train Steps: 43/90 Loss: 0.0145 torch.Size([8, 600, 800])
  36183. torch.Size([8, 8])
  36184. tensor([[0.6212, 0.4171, 0.7875, 0.3633, 0.3813, 0.2933, 0.5675, 0.5700],
  36185. [0.6264, 0.4071, 0.9038, 0.3867, 0.3663, 0.3917, 0.6338, 0.5283],
  36186. [ nan, nan, 0.8888, 0.3100, 0.5262, 0.2817, 0.7145, 0.6003],
  36187. [ nan, nan, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999],
  36188. [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332],
  36189. [ nan, nan, 0.7625, 0.2433, 0.3713, 0.2867, 0.5235, 0.5220],
  36190. [0.6325, 0.4165, 0.9000, 0.4617, 0.3813, 0.4900, 0.7485, 0.5447],
  36191. [0.6214, 0.4112, 0.7838, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083]],
  36192. device='cuda:0', dtype=torch.float64)
  36193. predictions are: tensor([[ 0.7925, -0.2197, 1.2258, -0.5560, -0.5129, -0.8534, 0.1539, 0.3959],
  36194. [ 0.9824, -0.1438, 1.7369, -0.4676, -0.5488, -0.3900, 0.4122, 0.0837],
  36195. [-0.5162, -1.0907, 1.6477, -0.7800, 0.1305, -1.0171, 0.9095, 0.4633],
  36196. [-2.3501, -2.3298, 1.2146, -1.0448, -0.2453, -1.2660, 0.2259, 0.1469],
  36197. [ 0.8942, -0.2418, 1.7042, -0.0787, -0.5784, -0.3624, 0.6176, 0.1548],
  36198. [-2.1278, -2.1791, 1.1879, -1.0004, -0.4103, -0.9609, 0.1842, 0.1987],
  36199. [ 0.9677, -0.1782, 1.7206, -0.1119, -0.5416, 0.0892, 0.8351, 0.1302],
  36200. [ 0.9036, -0.2227, 1.1927, -1.2856, -0.5745, -0.8649, 0.2551, 0.0385]],
  36201. device='cuda:0', grad_fn=<AddmmBackward>)
  36202. landmarks are: tensor([[[ 0.5832, -0.3593, 1.3515, -0.6077, -0.5249, -0.9310, 0.3353,
  36203. 0.3469],
  36204. [ 0.6073, -0.4058, 1.8885, -0.4999, -0.5942, -0.4768, 0.6413,
  36205. 0.1544],
  36206. [-2.2859, -2.2859, 1.8192, -0.8540, 0.1448, -0.9849, 1.0143,
  36207. 0.4867],
  36208. [-2.2859, -2.2859, 1.3400, -1.0388, -0.3055, -1.4930, 0.1157,
  36209. 0.0231],
  36210. [ 0.6520, -0.4056, 1.9173, -0.0765, -0.5596, -0.4537, 0.7949,
  36211. 0.1768],
  36212. [-2.2859, -2.2859, 1.2360, -1.1620, -0.5711, -0.9618, 0.1322,
  36213. 0.1253],
  36214. [ 0.6355, -0.3623, 1.8711, -0.1535, -0.5249, -0.0226, 1.1715,
  36215. 0.2302],
  36216. [ 0.5845, -0.3864, 1.3342, -1.3082, -0.6000, -0.8386, 0.3353,
  36217. 0.0620]]], device='cuda:0')
  36218. loss_train_step before backward: tensor(0.0917, device='cuda:0', grad_fn=<MseLossBackward>)
  36219. loss_train_step after backward: tensor(0.0917, device='cuda:0', grad_fn=<MseLossBackward>)
  36220. loss_train: 0.7134014395996928
  36221. step: 44
  36222. running loss: 0.0162136690818112
  36223.  
  36224. Train Steps: 44/90 Loss: 0.0162 torch.Size([8, 600, 800])
  36225. torch.Size([8, 8])
  36226. tensor([[0.6037, 0.4020, 0.8300, 0.4033, 0.3575, 0.4883, 0.5647, 0.5631],
  36227. [0.6204, 0.4007, 0.7838, 0.2100, 0.4475, 0.1733, 0.5825, 0.5167],
  36228. [ nan, nan, 0.6488, 0.1817, 0.4325, 0.1867, 0.5475, 0.5733],
  36229. [0.6149, 0.4054, 0.6713, 0.2333, 0.4025, 0.2017, 0.5213, 0.5695],
  36230. [0.6293, 0.4024, 0.8750, 0.5000, 0.4012, 0.5733, 0.7121, 0.5633],
  36231. [0.6252, 0.4158, 0.8988, 0.4083, 0.3788, 0.4783, 0.6225, 0.5633],
  36232. [0.6229, 0.4086, 0.7538, 0.2600, 0.4775, 0.1617, 0.5900, 0.5383],
  36233. [0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412]],
  36234. device='cuda:0', dtype=torch.float64)
  36235. predictions are: tensor([[ 0.2478, -0.5887, 1.5762, -0.3955, -0.6159, -0.0235, 0.3924, 0.1690],
  36236. [ 0.5105, -0.4776, 1.2312, -1.2097, -0.2561, -1.4255, 0.4250, 0.0495],
  36237. [-2.8672, -2.7021, 0.9996, -1.1260, -0.2092, -1.3250, 0.2963, 0.3326],
  36238. [ 0.4463, -0.5024, 0.8379, -1.1694, -0.4447, -1.2806, 0.0816, 0.2743],
  36239. [ 0.4706, -0.4984, 1.7752, 0.0275, -0.4758, 0.3607, 0.8918, 0.1198],
  36240. [ 0.3932, -0.5023, 1.7916, -0.3608, -0.5850, -0.0992, 0.6404, 0.2181],
  36241. [ 0.5354, -0.4506, 1.1401, -1.1430, -0.1944, -1.4443, 0.3866, 0.1853],
  36242. [ 0.5298, -0.4138, 1.7078, -0.5646, -0.5096, -0.9205, 0.3372, 0.1251]],
  36243. device='cuda:0', grad_fn=<AddmmBackward>)
  36244. landmarks are: tensor([[[ 0.5027, -0.4289, 1.5478, -0.4229, -0.6346, -0.0303, 0.3223,
  36245. 0.3148],
  36246. [ 0.5796, -0.4351, 1.3342, -1.3159, -0.2189, -1.4853, 0.4046,
  36247. 0.1005],
  36248. [-2.2859, -2.2859, 0.7106, -1.4468, -0.2882, -1.4237, 0.2430,
  36249. 0.3623],
  36250. [ 0.5544, -0.4133, 0.8145, -1.2082, -0.4268, -1.3544, 0.1221,
  36251. 0.3446],
  36252. [ 0.6207, -0.4273, 1.7557, 0.0236, -0.4326, 0.3623, 1.0033,
  36253. 0.3157],
  36254. [ 0.6017, -0.3654, 1.8654, -0.3998, -0.5365, -0.0765, 0.5894,
  36255. 0.3161],
  36256. [ 0.5911, -0.3984, 1.1956, -1.0850, -0.0804, -1.5392, 0.4393,
  36257. 0.2006],
  36258. [ 0.5784, -0.4006, 1.7911, -0.5701, -0.5192, -1.0331, 0.4137,
  36259. 0.2139]]], device='cuda:0')
  36260. loss_train_step before backward: tensor(0.0179, device='cuda:0', grad_fn=<MseLossBackward>)
  36261. loss_train_step after backward: tensor(0.0179, device='cuda:0', grad_fn=<MseLossBackward>)
  36262. loss_train: 0.7312849136069417
  36263. step: 45
  36264. running loss: 0.016250775857932036
  36265. Train Steps: 45/90 Loss: 0.0163 torch.Size([8, 600, 800])
  36266. torch.Size([8, 8])
  36267. tensor([[0.6109, 0.4009, 0.8725, 0.4700, 0.3550, 0.4167, 0.5650, 0.5117],
  36268. [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395],
  36269. [0.6161, 0.4040, 0.7525, 0.2133, 0.4125, 0.2067, 0.5787, 0.5117],
  36270. [0.6250, 0.4236, 0.8638, 0.3767, 0.4050, 0.3150, 0.5649, 0.5799],
  36271. [0.6198, 0.4101, 0.8838, 0.5283, 0.3763, 0.5267, 0.5913, 0.5567],
  36272. [0.6122, 0.4006, 0.8850, 0.4217, 0.4088, 0.5517, 0.6063, 0.5517],
  36273. [0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038],
  36274. [0.6098, 0.3991, 0.8638, 0.4717, 0.4263, 0.4967, 0.5212, 0.5650]],
  36275. device='cuda:0', dtype=torch.float64)
  36276. predictions are: tensor([[ 0.4556, -0.5099, 1.7766, -0.1818, -0.6755, -0.4903, 0.4060, 0.0858],
  36277. [ 0.5090, -0.4585, 1.5956, 0.1512, -0.6968, -0.7298, 0.3877, 0.2242],
  36278. [ 0.5739, -0.4423, 1.0018, -1.4543, -0.4832, -1.4066, 0.4122, 0.0719],
  36279. [ 0.1892, -0.6547, 1.6334, -0.6650, -0.4261, -0.9657, 0.4471, 0.4533],
  36280. [ 0.6028, -0.4209, 1.7416, 0.1237, -0.5934, -0.0068, 0.6176, 0.2581],
  36281. [ 0.5751, -0.4199, 1.8006, -0.3498, -0.4233, 0.1760, 0.6082, 0.1663],
  36282. [ 0.5965, -0.4214, 1.7018, -0.1034, -0.2478, -0.0326, 0.2991, -0.0696],
  36283. [ 0.4331, -0.5177, 1.6583, -0.1081, -0.3108, -0.0744, 0.3485, 0.2111]],
  36284. device='cuda:0', grad_fn=<AddmmBackward>)
  36285. landmarks are: tensor([[[ 0.5356, -0.4343, 1.7441, -0.1150, -0.6462, -0.3614, 0.3238,
  36286. 0.0774],
  36287. [ 0.5697, -0.4514, 1.7420, 0.2672, -0.6055, -0.6312, 0.3449,
  36288. 0.2058],
  36289. [ 0.5598, -0.4201, 1.1898, -1.3005, -0.3806, -1.3313, 0.3873,
  36290. 0.0774],
  36291. [ 0.6008, -0.3293, 1.7037, -0.5461, -0.4152, -0.8309, 0.3234,
  36292. 0.3928],
  36293. [ 0.5770, -0.3918, 1.7961, 0.1544, -0.5480, 0.1467, 0.4450,
  36294. 0.2853],
  36295. [ 0.5417, -0.4355, 1.8018, -0.3383, -0.3979, 0.2622, 0.5143,
  36296. 0.2622],
  36297. [ 0.5249, -0.4473, 1.7326, -0.0919, -0.2016, 0.1544, 0.1733,
  36298. 0.0412],
  36299. [ 0.5308, -0.4425, 1.7037, -0.1073, -0.3171, 0.0082, 0.1217,
  36300. 0.3238]]], device='cuda:0')
  36301. loss_train_step before backward: tensor(0.0118, device='cuda:0', grad_fn=<MseLossBackward>)
  36302. loss_train_step after backward: tensor(0.0118, device='cuda:0', grad_fn=<MseLossBackward>)
  36303. loss_train: 0.7431346783414483
  36304. step: 46
  36305. running loss: 0.016155101703074964
  36306. Train Steps: 46/90 Loss: 0.0162 torch.Size([8, 600, 800])
  36307. torch.Size([8, 8])
  36308. tensor([[0.6150, 0.4097, 0.7468, 0.3194, 0.3825, 0.2633, 0.5463, 0.5800],
  36309. [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212],
  36310. [0.6164, 0.4119, 0.7913, 0.2650, 0.3538, 0.3500, 0.5614, 0.5038],
  36311. [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395],
  36312. [0.6127, 0.4084, 0.8700, 0.4467, 0.3987, 0.4317, 0.5013, 0.5471],
  36313. [0.6200, 0.4059, 0.8700, 0.4900, 0.4163, 0.5000, 0.6162, 0.5467],
  36314. [0.6048, 0.3987, 0.7620, 0.3861, 0.3475, 0.4167, 0.5137, 0.5466],
  36315. [ nan, nan, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702]],
  36316. device='cuda:0', dtype=torch.float64)
  36317. predictions are: tensor([[ 0.5993, -0.3803, 1.1228, -0.8423, -0.6056, -1.1385, 0.0585, 0.3099],
  36318. [ 0.6539, -0.4069, 1.4465, -1.0566, -0.4854, -1.0856, 0.7644, 0.0711],
  36319. [ 0.5155, -0.4606, 1.3688, -1.0454, -0.7240, -0.7367, 0.3963, 0.0150],
  36320. [ 0.5882, -0.4315, 1.0366, -1.3716, -0.4745, -1.2091, 0.5045, 0.1624],
  36321. [ 0.4665, -0.4729, 1.7877, -0.1115, -0.4896, -0.3687, 0.1220, 0.2028],
  36322. [ 0.4432, -0.5455, 1.8228, 0.0551, -0.3761, -0.0258, 0.6413, 0.1544],
  36323. [ 0.5897, -0.4017, 1.2351, -0.5070, -0.7453, -0.4400, 0.1918, 0.1963],
  36324. [-1.4964, -1.7881, 1.8678, -0.7517, 0.1144, -1.2951, 1.0231, 0.3557]],
  36325. device='cuda:0', grad_fn=<AddmmBackward>)
  36326. landmarks are: tensor([[[ 0.5548, -0.3936, 1.1634, -0.8105, -0.5192, -1.0696, 0.2372,
  36327. 0.3931],
  36328. [ 0.6125, -0.4345, 1.4308, -1.1384, -0.4213, -1.0031, 0.7190,
  36329. 0.1214],
  36330. [ 0.5614, -0.3834, 1.3688, -1.0619, -0.6520, -0.6693, 0.3069,
  36331. 0.0412],
  36332. [ 0.5713, -0.4538, 1.1028, -1.3659, -0.3831, -1.1273, 0.5340,
  36333. 0.2058],
  36334. [ 0.5441, -0.3997, 1.7326, -0.2228, -0.4441, -0.2921, 0.0296,
  36335. 0.2409],
  36336. [ 0.5776, -0.4109, 1.7326, -0.0226, -0.3633, 0.0236, 0.5605,
  36337. 0.2391],
  36338. [ 0.5076, -0.4443, 1.2337, -0.5023, -0.6808, -0.3614, 0.0866,
  36339. 0.2386],
  36340. [-2.2859, -2.2859, 1.8942, -0.6693, 0.0871, -1.2236, 1.1130,
  36341. 0.3478]]], device='cuda:0')
  36342. loss_train_step before backward: tensor(0.0182, device='cuda:0', grad_fn=<MseLossBackward>)
  36343. loss_train_step after backward: tensor(0.0182, device='cuda:0', grad_fn=<MseLossBackward>)
  36344. loss_train: 0.7613579174503684
  36345. step: 47
  36346. running loss: 0.016199104626603584
  36347. Train Steps: 47/90 Loss: 0.0162 torch.Size([8, 600, 800])
  36348. torch.Size([8, 8])
  36349. tensor([[0.6202, 0.4079, 0.8025, 0.2500, 0.3763, 0.3217, 0.6125, 0.5533],
  36350. [0.6218, 0.4137, 0.7263, 0.2233, 0.4075, 0.2650, 0.6212, 0.5783],
  36351. [0.6282, 0.4034, 0.7830, 0.2080, 0.4532, 0.2080, 0.6404, 0.5323],
  36352. [0.6129, 0.3930, 0.8769, 0.5105, 0.4146, 0.5642, 0.6000, 0.5082],
  36353. [0.6226, 0.4103, 0.8575, 0.3450, 0.4388, 0.2067, 0.5787, 0.5383],
  36354. [0.6224, 0.4179, 0.8700, 0.5683, 0.4037, 0.4683, 0.5650, 0.5600],
  36355. [0.6057, 0.4011, 0.8750, 0.4267, 0.4400, 0.5800, 0.5845, 0.5585],
  36356. [0.6187, 0.4104, 0.7126, 0.2069, 0.3950, 0.2517, 0.5913, 0.5683]],
  36357. device='cuda:0', dtype=torch.float64)
  36358. predictions are: tensor([[ 0.4333, -0.5111, 1.4087, -1.0953, -0.6372, -0.8462, 0.4818, 0.1355],
  36359. [ 0.3784, -0.5491, 1.0189, -1.2366, -0.5795, -1.2066, 0.4777, 0.2933],
  36360. [ 0.6046, -0.4339, 1.2734, -1.2354, -0.3520, -1.3963, 0.5953, 0.0656],
  36361. [ 0.2975, -0.6681, 1.8023, 0.2010, -0.3973, 0.1365, 0.5395, 0.0553],
  36362. [ 0.5932, -0.3750, 1.7603, -0.6885, -0.3399, -1.3641, 0.3018, 0.1969],
  36363. [ 0.2940, -0.5934, 1.7302, 0.3709, -0.5093, -0.3573, 0.4433, 0.4143],
  36364. [ 0.3846, -0.5253, 1.8102, -0.2334, -0.3446, 0.1800, 0.4126, 0.1311],
  36365. [ 0.4028, -0.5231, 0.8992, -1.3941, -0.5818, -1.2945, 0.3011, 0.2570]],
  36366. device='cuda:0', grad_fn=<AddmmBackward>)
  36367. landmarks are: tensor([[[ 0.5788, -0.4020, 1.4208, -1.1312, -0.5480, -0.8002, 0.5432,
  36368. 0.2699],
  36369. [ 0.5863, -0.3749, 1.0686, -1.2543, -0.4037, -1.0619, 0.5836,
  36370. 0.3854],
  36371. [ 0.6158, -0.4225, 1.3307, -1.3253, -0.1924, -1.3252, 0.6721,
  36372. 0.1727],
  36373. [ 0.5450, -0.4706, 1.7643, 0.0722, -0.3708, 0.3200, 0.4854,
  36374. 0.0612],
  36375. [ 0.5898, -0.3908, 1.6748, -0.6924, -0.2594, -1.3313, 0.3873,
  36376. 0.2006],
  36377. [ 0.5891, -0.3557, 1.7326, 0.3392, -0.4210, -0.1227, 0.3238,
  36378. 0.3007],
  36379. [ 0.5115, -0.4332, 1.7557, -0.3152, -0.2536, 0.3931, 0.4139,
  36380. 0.2936],
  36381. [ 0.5718, -0.3905, 1.0053, -1.3305, -0.4614, -1.1235, 0.4450,
  36382. 0.3392]]], device='cuda:0')
  36383. loss_train_step before backward: tensor(0.0146, device='cuda:0', grad_fn=<MseLossBackward>)
  36384. loss_train_step after backward: tensor(0.0146, device='cuda:0', grad_fn=<MseLossBackward>)
  36385. loss_train: 0.7759437253698707
  36386. step: 48
  36387. running loss: 0.016165494278538972
  36388.  
  36389. Train Steps: 48/90 Loss: 0.0162 torch.Size([8, 600, 800])
  36390. torch.Size([8, 8])
  36391. tensor([[0.6264, 0.4069, 0.7900, 0.1650, 0.4275, 0.2267, 0.6290, 0.4947],
  36392. [0.6271, 0.4040, 0.9000, 0.3250, 0.4938, 0.2300, 0.7192, 0.5413],
  36393. [0.6275, 0.4048, 0.8488, 0.2883, 0.4463, 0.2033, 0.6321, 0.5155],
  36394. [0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412],
  36395. [0.6182, 0.4058, 0.8738, 0.4350, 0.3563, 0.3400, 0.5290, 0.5822],
  36396. [0.6115, 0.3998, 0.7063, 0.2383, 0.4037, 0.1950, 0.5320, 0.4993],
  36397. [0.6175, 0.3997, 0.6737, 0.2500, 0.4313, 0.1933, 0.5513, 0.5750],
  36398. [0.6182, 0.3930, 0.8841, 0.3892, 0.3556, 0.4967, 0.6222, 0.5279]],
  36399. device='cuda:0', dtype=torch.float64)
  36400. predictions are: tensor([[ 4.9754e-01, -5.1794e-01, 1.2568e+00, -1.4171e+00, -3.8776e-01,
  36401. -1.1685e+00, 5.8964e-01, 4.4017e-02],
  36402. [ 2.9641e-01, -6.1759e-01, 1.8639e+00, -7.7632e-01, -1.0200e-01,
  36403. -1.1415e+00, 9.5960e-01, 2.4079e-01],
  36404. [ 6.7223e-01, -3.8713e-01, 1.6328e+00, -9.0681e-01, -2.8939e-01,
  36405. -1.2221e+00, 7.3069e-01, 1.4676e-01],
  36406. [ 5.1888e-01, -4.4076e-01, 1.7888e+00, -4.8815e-01, -5.6462e-01,
  36407. -9.1237e-01, 3.5224e-01, 2.3299e-01],
  36408. [ 3.6893e-01, -5.5036e-01, 1.7233e+00, -2.2765e-01, -7.2718e-01,
  36409. -5.9332e-01, 1.3962e-01, 3.8474e-01],
  36410. [ 6.5524e-01, -3.6731e-01, 9.2200e-01, -1.1836e+00, -5.1188e-01,
  36411. -1.2763e+00, 6.7762e-02, -1.0445e-03],
  36412. [ 2.2790e-01, -6.3644e-01, 6.8836e-01, -1.2796e+00, -3.6872e-01,
  36413. -1.4031e+00, 1.8171e-01, 4.5589e-01],
  36414. [-1.6043e-02, -8.6411e-01, 1.7684e+00, -4.3701e-01, -6.8084e-01,
  36415. 4.7354e-02, 7.2844e-01, 1.9542e-01]], device='cuda:0',
  36416. grad_fn=<AddmmBackward>)
  36417. landmarks are: tensor([[[ 6.0739e-01, -4.0631e-01, 1.3630e+00, -1.5238e+00, -3.1132e-01,
  36418. -1.2390e+00, 6.1950e-01, -9.2270e-04],
  36419. [ 6.1083e-01, -4.2008e-01, 1.8711e+00, -7.8476e-01, -5.3118e-03,
  36420. -1.2236e+00, 1.0362e+00, 2.1421e-01],
  36421. [ 6.1259e-01, -4.1609e-01, 1.6344e+00, -9.5412e-01, -2.2471e-01,
  36422. -1.3467e+00, 6.3389e-01, 9.5262e-02],
  36423. [ 5.7841e-01, -4.0062e-01, 1.7911e+00, -5.7008e-01, -5.1916e-01,
  36424. -1.0331e+00, 4.1374e-01, 2.1391e-01],
  36425. [ 5.6969e-01, -4.1132e-01, 1.7499e+00, -2.7667e-01, -6.4042e-01,
  36426. -7.1547e-01, 1.5756e-01, 4.0319e-01],
  36427. [ 5.3837e-01, -4.3934e-01, 9.7621e-01, -1.1851e+00, -4.2102e-01,
  36428. -1.3852e+00, 1.7122e-01, 2.0118e-02],
  36429. [ 5.6634e-01, -4.3965e-01, 8.2610e-01, -1.1312e+00, -2.9400e-01,
  36430. -1.3929e+00, 2.6028e-01, 3.6998e-01],
  36431. [ 5.6966e-01, -4.7064e-01, 1.7976e+00, -4.8841e-01, -6.4332e-01,
  36432. 8.0865e-03, 5.8780e-01, 1.5252e-01]]], device='cuda:0')
  36433. loss_train_step before backward: tensor(0.0183, device='cuda:0', grad_fn=<MseLossBackward>)
  36434. loss_train_step after backward: tensor(0.0183, device='cuda:0', grad_fn=<MseLossBackward>)
  36435. loss_train: 0.7942405147477984
  36436. step: 49
  36437. running loss: 0.016208990096893847
  36438. Train Steps: 49/90 Loss: 0.0162 torch.Size([8, 600, 800])
  36439. torch.Size([8, 8])
  36440. tensor([[0.6271, 0.4024, 0.9000, 0.3517, 0.3700, 0.4517, 0.6931, 0.5285],
  36441. [0.6329, 0.4175, 0.8550, 0.2333, 0.5425, 0.2250, 0.7398, 0.5609],
  36442. [0.6058, 0.3978, 0.8287, 0.3600, 0.3400, 0.4050, 0.5529, 0.5461],
  36443. [0.6273, 0.4105, 0.8988, 0.4517, 0.3912, 0.2550, 0.5894, 0.4811],
  36444. [0.6325, 0.4165, 0.9000, 0.4617, 0.3813, 0.4900, 0.7485, 0.5447],
  36445. [0.6278, 0.4253, 0.8875, 0.5017, 0.4113, 0.2750, 0.5413, 0.6196],
  36446. [0.6128, 0.4118, 0.8638, 0.5333, 0.4625, 0.5267, 0.5193, 0.5475],
  36447. [0.6222, 0.4169, 0.8638, 0.5650, 0.4313, 0.4783, 0.5637, 0.5633]],
  36448. device='cuda:0', dtype=torch.float64)
  36449. predictions are: tensor([[ 0.3633, -0.6099, 1.7727, -0.7365, -0.6176, -0.1545, 0.9483, 0.1723],
  36450. [ 0.3857, -0.5608, 1.5930, -1.4546, 0.1408, -1.4649, 1.0670, 0.2022],
  36451. [ 0.3727, -0.5318, 1.4499, -0.7004, -0.8048, -0.4178, 0.1934, 0.1654],
  36452. [ 0.6114, -0.3969, 1.8067, -0.3324, -0.5716, -1.1282, 0.2881, -0.0128],
  36453. [ 0.6555, -0.3796, 1.7888, -0.1854, -0.6189, 0.0533, 0.9555, 0.2217],
  36454. [ 0.5391, -0.4041, 1.6146, -0.1128, -0.4412, -1.0803, 0.1859, 0.5246],
  36455. [ 0.6420, -0.3471, 1.5873, 0.1019, -0.2799, 0.1364, -0.0239, 0.1821],
  36456. [ 0.5307, -0.4357, 1.6244, 0.2666, -0.3907, -0.1204, 0.2242, 0.3382]],
  36457. device='cuda:0', grad_fn=<AddmmBackward>)
  36458. landmarks are: tensor([[[ 0.6108, -0.4273, 1.8711, -0.6616, -0.5769, -0.1997, 0.9156,
  36459. 0.1554],
  36460. [ 0.6372, -0.3575, 1.6633, -1.2082, 0.2199, -1.2467, 1.1313,
  36461. 0.3050],
  36462. [ 0.5124, -0.4483, 1.5420, -0.6231, -0.7155, -0.4152, 0.2679,
  36463. 0.2365],
  36464. [ 0.6116, -0.3898, 1.8654, -0.1997, -0.4788, -1.1081, 0.4367,
  36465. -0.0637],
  36466. [ 0.6355, -0.3623, 1.8711, -0.1535, -0.5249, -0.0226, 1.1715,
  36467. 0.2302],
  36468. [ 0.6139, -0.3216, 1.8134, 0.0313, -0.3864, -1.0157, 0.2144,
  36469. 0.5762],
  36470. [ 0.5447, -0.3838, 1.7037, 0.1775, -0.1497, 0.1467, 0.1128,
  36471. 0.2431],
  36472. [ 0.5880, -0.3605, 1.7037, 0.3238, -0.2940, -0.0765, 0.3180,
  36473. 0.3161]]], device='cuda:0')
  36474. loss_train_step before backward: tensor(0.0114, device='cuda:0', grad_fn=<MseLossBackward>)
  36475. loss_train_step after backward: tensor(0.0114, device='cuda:0', grad_fn=<MseLossBackward>)
  36476. loss_train: 0.8056448325514793
  36477. step: 50
  36478. running loss: 0.016112896651029586
  36479. Train Steps: 50/90 Loss: 0.0161 torch.Size([8, 600, 800])
  36480. torch.Size([8, 8])
  36481. tensor([[0.6296, 0.3989, 0.9000, 0.4150, 0.3613, 0.3867, 0.6400, 0.5100],
  36482. [0.6172, 0.4055, 0.8175, 0.2650, 0.3550, 0.3683, 0.5787, 0.5550],
  36483. [0.6153, 0.4119, 0.8463, 0.3833, 0.3600, 0.3200, 0.5106, 0.5563],
  36484. [0.6127, 0.4084, 0.8700, 0.4467, 0.3987, 0.4317, 0.5013, 0.5471],
  36485. [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650],
  36486. [0.6225, 0.4191, 0.8500, 0.4167, 0.3688, 0.3233, 0.5650, 0.5650],
  36487. [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390],
  36488. [0.6132, 0.3930, 0.8672, 0.4754, 0.3712, 0.5222, 0.5974, 0.5098]],
  36489. device='cuda:0', dtype=torch.float64)
  36490. predictions are: tensor([[ 6.5583e-01, -4.2216e-01, 1.9241e+00, -2.7688e-01, -5.5424e-01,
  36491. -6.0314e-01, 9.0408e-01, 1.5943e-01],
  36492. [ 4.8151e-01, -4.7833e-01, 1.4162e+00, -1.0737e+00, -6.1699e-01,
  36493. -6.1750e-01, 5.0363e-01, 2.4150e-01],
  36494. [ 5.1282e-01, -4.4040e-01, 1.6403e+00, -5.1735e-01, -5.4924e-01,
  36495. -9.2146e-01, 2.4885e-01, 2.8576e-01],
  36496. [ 5.4983e-01, -4.1066e-01, 1.7298e+00, -1.6783e-01, -4.0936e-01,
  36497. -3.1518e-01, 1.2596e-01, 2.7520e-01],
  36498. [ 7.2211e-01, -3.1429e-01, 1.2671e+00, -1.0859e+00, -4.3973e-01,
  36499. -1.0486e+00, 2.4254e-01, 2.3212e-01],
  36500. [ 4.2025e-01, -4.8194e-01, 1.6345e+00, -3.9314e-01, -5.3136e-01,
  36501. -8.7824e-01, 4.3722e-01, 4.3657e-01],
  36502. [ 5.2567e-01, -4.8861e-01, 1.7370e+00, -2.4383e-02, -3.7794e-01,
  36503. 1.3738e-02, 3.3758e-01, 1.4508e-01],
  36504. [ 6.6609e-01, -3.9081e-01, 1.7458e+00, 3.7987e-04, -4.4658e-01,
  36505. 1.1860e-01, 6.7305e-01, 1.6379e-01]], device='cuda:0',
  36506. grad_fn=<AddmmBackward>)
  36507. landmarks are: tensor([[[ 0.6220, -0.4436, 1.8711, -0.3691, -0.6173, -0.4999, 0.6702,
  36508. 0.0697],
  36509. [ 0.5647, -0.4129, 1.4901, -1.0619, -0.6462, -0.5846, 0.3873,
  36510. 0.2776],
  36511. [ 0.5561, -0.3834, 1.6229, -0.5153, -0.6231, -0.8079, 0.0727,
  36512. 0.2837],
  36513. [ 0.5441, -0.3997, 1.7326, -0.2228, -0.4441, -0.2921, 0.0296,
  36514. 0.2409],
  36515. [ 0.5320, -0.4189, 1.3053, -1.0773, -0.5711, -0.9849, 0.2267,
  36516. 0.3237],
  36517. [ 0.5894, -0.3503, 1.6402, -0.3614, -0.5827, -0.7925, 0.3238,
  36518. 0.3238],
  36519. [ 0.5250, -0.4661, 1.7383, -0.0765, -0.4268, -0.0226, 0.2535,
  36520. 0.2035],
  36521. [ 0.5466, -0.4706, 1.7198, -0.0903, -0.5712, 0.1261, 0.4733,
  36522. 0.0688]]], device='cuda:0')
  36523. loss_train_step before backward: tensor(0.0073, device='cuda:0', grad_fn=<MseLossBackward>)
  36524. loss_train_step after backward: tensor(0.0073, device='cuda:0', grad_fn=<MseLossBackward>)
  36525. loss_train: 0.8129367139190435
  36526. step: 51
  36527. running loss: 0.01593993556704007
  36528.  
  36529. Train Steps: 51/90 Loss: 0.0159 torch.Size([8, 600, 800])
  36530. torch.Size([8, 8])
  36531. tensor([[0.6278, 0.4253, 0.8875, 0.5017, 0.4113, 0.2750, 0.5413, 0.6196],
  36532. [0.6219, 0.4114, 0.8175, 0.2817, 0.3925, 0.2783, 0.5900, 0.5350],
  36533. [0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162],
  36534. [ nan, nan, 0.9050, 0.3500, 0.5138, 0.2300, 0.7359, 0.5702],
  36535. [0.6219, 0.4097, 0.8738, 0.3400, 0.3563, 0.4117, 0.5975, 0.5683],
  36536. [0.6222, 0.3937, 0.8350, 0.5617, 0.4138, 0.4600, 0.5800, 0.5233],
  36537. [0.6086, 0.3998, 0.8788, 0.4450, 0.4025, 0.4650, 0.5306, 0.5103],
  36538. [0.6204, 0.4013, 0.8075, 0.2400, 0.4313, 0.2050, 0.5800, 0.5150]],
  36539. device='cuda:0', dtype=torch.float64)
  36540. predictions are: tensor([[ 0.6138, -0.3162, 1.6757, -0.0385, -0.3359, -1.0177, 0.2893, 0.4970],
  36541. [ 0.3506, -0.4950, 1.3670, -1.0617, -0.4420, -0.9203, 0.3976, 0.2195],
  36542. [ 0.6430, -0.3678, 1.8338, -0.3944, -0.5642, -0.1219, 0.7280, 0.2425],
  36543. [-2.2223, -2.2294, 1.5857, -0.9523, 0.0569, -1.1593, 0.6875, 0.2897],
  36544. [ 0.6182, -0.3403, 1.7174, -0.7481, -0.6678, -0.3971, 0.4993, 0.2399],
  36545. [ 0.5614, -0.4385, 1.5351, 0.2646, -0.4268, -0.1418, 0.4188, 0.1910],
  36546. [ 0.5111, -0.4241, 1.6742, -0.2531, -0.4758, -0.0282, 0.2072, 0.1283],
  36547. [ 0.6298, -0.3722, 1.3168, -1.1760, -0.2614, -1.3129, 0.4119, 0.1058]],
  36548. device='cuda:0', grad_fn=<AddmmBackward>)
  36549. landmarks are: tensor([[[ 0.6139, -0.3216, 1.8134, 0.0313, -0.3864, -1.0157, 0.2144,
  36550. 0.5762],
  36551. [ 0.5868, -0.3858, 1.4901, -0.9849, -0.4730, -1.0003, 0.4393,
  36552. 0.1852],
  36553. [ 0.6059, -0.4442, 1.8921, -0.3422, -0.6230, -0.1461, 0.7489,
  36554. 0.0985],
  36555. [-2.2859, -2.2859, 1.8942, -0.6693, 0.0871, -1.2236, 1.1130,
  36556. 0.3478],
  36557. [ 0.5867, -0.3937, 1.7499, -0.7155, -0.6404, -0.3844, 0.4739,
  36558. 0.3392],
  36559. [ 0.5880, -0.4676, 1.5709, 0.3084, -0.3748, -0.1612, 0.3931,
  36560. 0.1313],
  36561. [ 0.5253, -0.4392, 1.7730, -0.2305, -0.4268, -0.1381, 0.1651,
  36562. 0.0712],
  36563. [ 0.5796, -0.4326, 1.4439, -1.1774, -0.2940, -1.3390, 0.3931,
  36564. 0.0928]]], device='cuda:0')
  36565. loss_train_step before backward: tensor(0.0099, device='cuda:0', grad_fn=<MseLossBackward>)
  36566. loss_train_step after backward: tensor(0.0099, device='cuda:0', grad_fn=<MseLossBackward>)
  36567. loss_train: 0.8228131141513586
  36568. step: 52
  36569. running loss: 0.015823329118295357
  36570. Train Steps: 52/90 Loss: 0.0158 torch.Size([8, 600, 800])
  36571. torch.Size([8, 8])
  36572. tensor([[0.6229, 0.4066, 0.8513, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350],
  36573. [0.6164, 0.4119, 0.7913, 0.2650, 0.3538, 0.3500, 0.5614, 0.5038],
  36574. [0.6082, 0.4024, 0.8738, 0.4017, 0.3688, 0.3950, 0.5306, 0.5136],
  36575. [ nan, nan, 0.7612, 0.3250, 0.4037, 0.2533, 0.5438, 0.5767],
  36576. [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882],
  36577. [0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090],
  36578. [0.6145, 0.3983, 0.8700, 0.5017, 0.4400, 0.4800, 0.5375, 0.5750],
  36579. [0.6289, 0.4019, 0.8113, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332]],
  36580. device='cuda:0', dtype=torch.float64)
  36581. predictions are: tensor([[ 0.6219, -0.3752, 1.6652, 0.2994, -0.1866, 0.0215, 0.3743, 0.2948],
  36582. [ 0.7644, -0.2511, 1.4535, -1.0984, -0.6169, -0.6842, 0.4383, 0.1393],
  36583. [ 0.5370, -0.4199, 1.7934, -0.4256, -0.5711, -0.4423, 0.2916, 0.1527],
  36584. [-1.5907, -1.8032, 1.1504, -1.0188, -0.3355, -1.0750, 0.1658, 0.3443],
  36585. [ 0.7614, -0.2678, 1.8735, -0.4168, -0.5463, -0.5598, 0.2533, 0.0554],
  36586. [ 0.7204, -0.3163, 1.9029, -0.5729, -0.3919, -1.0855, 0.6899, 0.2262],
  36587. [ 0.5421, -0.4263, 1.7555, -0.0450, -0.2282, -0.0415, 0.2501, 0.3798],
  36588. [ 0.6803, -0.3359, 1.5785, 0.1547, -0.4745, -0.0506, 0.9300, 0.3099]],
  36589. device='cuda:0', grad_fn=<AddmmBackward>)
  36590. landmarks are: tensor([[[ 0.5911, -0.4080, 1.6460, 0.3546, -0.2074, 0.0467, 0.4970,
  36591. 0.1852],
  36592. [ 0.5614, -0.3834, 1.3688, -1.0619, -0.6520, -0.6693, 0.3069,
  36593. 0.0412],
  36594. [ 0.5235, -0.4273, 1.7499, -0.4306, -0.5827, -0.4614, 0.1651,
  36595. 0.0862],
  36596. [-2.2859, -2.2859, 1.2303, -0.7848, -0.4210, -1.1158, 0.2256,
  36597. 0.3777],
  36598. [ 0.5595, -0.3988, 1.7672, -0.4460, -0.5538, -0.5384, 0.0828,
  36599. -0.0310],
  36600. [ 0.6165, -0.4249, 1.7961, -0.5384, -0.4499, -0.9695, 0.6401,
  36601. 0.0652],
  36602. [ 0.5525, -0.4463, 1.7326, 0.0313, -0.2536, -0.0688, 0.1968,
  36603. 0.3700],
  36604. [ 0.6191, -0.4297, 1.4612, 0.2391, -0.4961, 0.0313, 1.1166,
  36605. 0.1768]]], device='cuda:0')
  36606. loss_train_step before backward: tensor(0.0193, device='cuda:0', grad_fn=<MseLossBackward>)
  36607. loss_train_step after backward: tensor(0.0193, device='cuda:0', grad_fn=<MseLossBackward>)
  36608. loss_train: 0.8421134296804667
  36609. step: 53
  36610. running loss: 0.015888932635480503
  36611. Train Steps: 53/90 Loss: 0.0159 torch.Size([8, 600, 800])
  36612. torch.Size([8, 8])
  36613. tensor([[0.6115, 0.4081, 0.6725, 0.2433, 0.4088, 0.1933, 0.5167, 0.5544],
  36614. [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6063, 0.5617],
  36615. [0.6230, 0.4152, 0.7588, 0.2283, 0.4012, 0.2883, 0.6200, 0.5767],
  36616. [0.6106, 0.4022, 0.7125, 0.2650, 0.3713, 0.2700, 0.5431, 0.5123],
  36617. [0.6150, 0.3935, 0.8696, 0.5158, 0.4647, 0.5329, 0.6041, 0.5153],
  36618. [ nan, nan, 0.6412, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700],
  36619. [0.6222, 0.4169, 0.8638, 0.5650, 0.4313, 0.4783, 0.5637, 0.5633],
  36620. [0.6277, 0.4103, 0.8087, 0.5717, 0.4188, 0.4750, 0.5663, 0.6083]],
  36621. device='cuda:0', dtype=torch.float64)
  36622. predictions are: tensor([[ 0.6673, -0.3268, 1.1403, -1.1721, -0.3743, -1.4599, 0.3026, 0.2220],
  36623. [ 0.6780, -0.3399, 2.0108, -0.0608, -0.5956, -0.1336, 0.5727, 0.2934],
  36624. [ 0.3732, -0.5323, 1.4563, -1.1377, -0.4467, -0.9829, 0.7052, 0.2905],
  36625. [ 0.6293, -0.3675, 1.2879, -1.0773, -0.5225, -1.0735, 0.2881, 0.0723],
  36626. [ 0.4728, -0.4933, 1.8535, 0.0456, -0.1775, 0.2102, 0.4344, 0.1276],
  36627. [-2.0240, -2.1459, 0.9359, -1.3552, -0.3156, -1.4738, 0.2332, 0.2545],
  36628. [ 0.6209, -0.3482, 1.8802, 0.2570, -0.3235, -0.0597, 0.2850, 0.2831],
  36629. [ 0.6233, -0.3547, 1.7081, 0.2817, -0.3935, -0.0655, 0.3343, 0.4167]],
  36630. device='cuda:0', grad_fn=<AddmmBackward>)
  36631. landmarks are: tensor([[[ 0.5385, -0.4010, 0.8203, -1.1620, -0.3979, -1.3929, 0.1005,
  36632. 0.2747],
  36633. [ 0.5797, -0.3965, 1.8480, -0.0765, -0.6173, -0.1535, 0.5143,
  36634. 0.3084],
  36635. [ 0.5915, -0.3682, 1.2187, -1.2313, -0.4326, -0.9541, 0.5778,
  36636. 0.3777],
  36637. [ 0.5342, -0.4280, 1.0051, -1.0619, -0.5711, -1.0388, 0.2226,
  36638. 0.0802],
  36639. [ 0.5548, -0.4682, 1.7309, 0.0966, -0.1394, 0.1757, 0.5045,
  36640. 0.0942],
  36641. [-2.2859, -2.2859, 0.6760, -1.4083, -0.3286, -1.4160, 0.2487,
  36642. 0.3469],
  36643. [ 0.5880, -0.3605, 1.7037, 0.3238, -0.2940, -0.0765, 0.3180,
  36644. 0.3161],
  36645. [ 0.6134, -0.3910, 1.4497, 0.3546, -0.3517, -0.0919, 0.3296,
  36646. 0.5239]]], device='cuda:0')
  36647. loss_train_step before backward: tensor(0.0127, device='cuda:0', grad_fn=<MseLossBackward>)
  36648. loss_train_step after backward: tensor(0.0127, device='cuda:0', grad_fn=<MseLossBackward>)
  36649. loss_train: 0.8548482032492757
  36650. step: 54
  36651. running loss: 0.015830522282393993
  36652. Train Steps: 54/90 Loss: 0.0158 torch.Size([8, 600, 800])
  36653. torch.Size([8, 8])
  36654. tensor([[0.6200, 0.3993, 0.8519, 0.4923, 0.3962, 0.4717, 0.6013, 0.5433],
  36655. [0.6274, 0.4270, 0.8938, 0.4967, 0.3550, 0.4283, 0.5700, 0.5733],
  36656. [ nan, nan, 0.6688, 0.2513, 0.4113, 0.2117, 0.5193, 0.5933],
  36657. [ nan, nan, 0.7725, 0.2611, 0.3675, 0.2733, 0.5413, 0.5167],
  36658. [0.6182, 0.3972, 0.8720, 0.5527, 0.3638, 0.3582, 0.5696, 0.5395],
  36659. [0.6140, 0.4034, 0.8850, 0.4317, 0.4288, 0.5067, 0.5825, 0.5533],
  36660. [0.6097, 0.3988, 0.8650, 0.5250, 0.4213, 0.5200, 0.5675, 0.5050],
  36661. [0.6126, 0.4067, 0.8638, 0.5383, 0.4188, 0.4850, 0.5016, 0.5392]],
  36662. device='cuda:0', dtype=torch.float64)
  36663. predictions are: tensor([[ 0.7359, -0.2978, 1.7785, -0.0222, -0.3996, -0.1070, 0.5864, 0.2958],
  36664. [ 0.7449, -0.2164, 1.9379, -0.0315, -0.5511, -0.3084, 0.5034, 0.3601],
  36665. [-1.3691, -1.6667, 0.8612, -1.2613, -0.3751, -1.3698, 0.2615, 0.4043],
  36666. [-1.9919, -2.1095, 1.2635, -1.0819, -0.4544, -1.0682, 0.2091, 0.1249],
  36667. [ 0.7430, -0.2650, 1.8531, 0.1796, -0.5756, -0.6459, 0.5336, 0.1851],
  36668. [ 0.8209, -0.2462, 1.8957, -0.2638, -0.2654, 0.1714, 0.5637, 0.2248],
  36669. [ 0.7045, -0.3207, 1.8143, 0.1388, -0.2615, 0.2025, 0.2661, 0.0781],
  36670. [ 0.8117, -0.2041, 1.7769, 0.2153, -0.3169, 0.0384, 0.1559, 0.1889]],
  36671. device='cuda:0', grad_fn=<AddmmBackward>)
  36672. landmarks are: tensor([[[ 0.5777, -0.4415, 1.6491, -0.0122, -0.4557, -0.1073, 0.4912,
  36673. 0.2237],
  36674. [ 0.6121, -0.3138, 1.8423, 0.0082, -0.6462, -0.3075, 0.3469,
  36675. 0.3623],
  36676. [-2.2859, -2.2859, 0.8033, -1.1250, -0.3864, -1.3082, 0.1126,
  36677. 0.4543],
  36678. [-2.2859, -2.2859, 1.2820, -1.0801, -0.5885, -1.0234, 0.2141,
  36679. 0.1005],
  36680. [ 0.5697, -0.4514, 1.7420, 0.2672, -0.6055, -0.6312, 0.3449,
  36681. 0.2058],
  36682. [ 0.5499, -0.4225, 1.8018, -0.2921, -0.3055, 0.0543, 0.4046,
  36683. 0.2699],
  36684. [ 0.5303, -0.4440, 1.7095, 0.1390, -0.3402, 0.1159, 0.3353,
  36685. 0.0467],
  36686. [ 0.5436, -0.4076, 1.7037, 0.2006, -0.3517, -0.0457, 0.0311,
  36687. 0.2048]]], device='cuda:0')
  36688. loss_train_step before backward: tensor(0.0322, device='cuda:0', grad_fn=<MseLossBackward>)
  36689. loss_train_step after backward: tensor(0.0322, device='cuda:0', grad_fn=<MseLossBackward>)
  36690. loss_train: 0.8870698222890496
  36691. step: 55
  36692. running loss: 0.016128542223437264
  36693.  
  36694. Train Steps: 55/90 Loss: 0.0161 torch.Size([8, 600, 800])
  36695. torch.Size([8, 8])
  36696. tensor([[0.6204, 0.4049, 0.7975, 0.2700, 0.3937, 0.2567, 0.5700, 0.5183],
  36697. [0.6279, 0.4050, 0.8025, 0.2200, 0.4313, 0.2733, 0.6820, 0.5320],
  36698. [0.6142, 0.3982, 0.8650, 0.4883, 0.3912, 0.4317, 0.5315, 0.5350],
  36699. [0.6201, 0.3970, 0.8413, 0.4950, 0.4413, 0.5183, 0.6088, 0.5400],
  36700. [0.6205, 0.4081, 0.8950, 0.4017, 0.3788, 0.4700, 0.5963, 0.5667],
  36701. [0.6201, 0.4064, 0.8688, 0.5050, 0.4225, 0.5100, 0.6138, 0.5500],
  36702. [ nan, nan, 0.7850, 0.2700, 0.4288, 0.1717, 0.5199, 0.4999],
  36703. [ nan, nan, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552]],
  36704. device='cuda:0', dtype=torch.float64)
  36705. predictions are: tensor([[ 0.6181, -0.3578, 1.4846, -0.8818, -0.5126, -1.1168, 0.2677, 0.1355],
  36706. [ 0.7924, -0.2418, 1.4436, -1.0710, -0.3385, -1.0589, 0.8068, 0.2151],
  36707. [ 0.7669, -0.2631, 1.7523, 0.1436, -0.4913, -0.1816, 0.1077, 0.1780],
  36708. [ 0.7052, -0.3087, 1.6060, 0.0755, -0.2882, 0.1245, 0.3166, 0.2797],
  36709. [ 0.8958, -0.1547, 1.7925, -0.2482, -0.5676, -0.0407, 0.4489, 0.3177],
  36710. [ 0.8621, -0.2202, 1.7208, 0.1235, -0.3645, 0.0783, 0.3573, 0.2843],
  36711. [-1.8809, -1.9708, 1.1639, -1.0491, -0.3869, -1.3121, 0.1169, 0.1347],
  36712. [-2.1162, -2.1510, 1.5890, -1.0452, 0.0364, -1.0503, 0.8505, 0.3124]],
  36713. device='cuda:0', grad_fn=<AddmmBackward>)
  36714. landmarks are: tensor([[[ 5.7985e-01, -4.1555e-01, 1.3977e+00, -1.0388e+00, -4.6721e-01,
  36715. -1.1004e+00, 3.4688e-01, 1.0824e-01],
  36716. [ 6.1413e-01, -4.1527e-01, 1.4208e+00, -1.2697e+00, -2.9400e-01,
  36717. -1.0234e+00, 8.6439e-01, 1.7146e-01],
  36718. [ 5.5127e-01, -4.4673e-01, 1.7095e+00, -3.0331e-02, -4.7875e-01,
  36719. -2.9207e-01, 1.6917e-01, 1.8544e-01],
  36720. [ 5.7812e-01, -4.5219e-01, 1.5998e+00, 4.6189e-04, -2.4781e-01,
  36721. 1.0824e-01, 5.2587e-01, 2.0831e-01],
  36722. [ 5.7991e-01, -4.0115e-01, 1.8480e+00, -4.3064e-01, -5.3649e-01,
  36723. -1.1501e-01, 4.6813e-01, 3.3149e-01],
  36724. [ 5.7841e-01, -4.0878e-01, 1.7268e+00, 4.6651e-02, -3.3441e-01,
  36725. 6.9746e-02, 5.4896e-01, 2.5450e-01],
  36726. [-2.2859e+00, -2.2859e+00, 1.3400e+00, -1.0388e+00, -3.0554e-01,
  36727. -1.4930e+00, 1.1570e-01, 2.3124e-02],
  36728. [-2.2859e+00, -2.2859e+00, 1.7557e+00, -1.1466e+00, 8.7067e-02,
  36729. -1.0773e+00, 1.1239e+00, 2.7833e-01]]], device='cuda:0')
  36730. loss_train_step before backward: tensor(0.0196, device='cuda:0', grad_fn=<MseLossBackward>)
  36731. loss_train_step after backward: tensor(0.0196, device='cuda:0', grad_fn=<MseLossBackward>)
  36732. loss_train: 0.9066562773659825
  36733. step: 56
  36734. running loss: 0.01619029066724969
  36735. Train Steps: 56/90 Loss: 0.0162 torch.Size([8, 600, 800])
  36736. torch.Size([8, 8])
  36737. tensor([[0.6339, 0.4123, 0.8638, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436],
  36738. [0.6214, 0.4040, 0.8838, 0.3500, 0.3600, 0.5183, 0.6362, 0.5200],
  36739. [0.6224, 0.4061, 0.8988, 0.4300, 0.3838, 0.4750, 0.6112, 0.5483],
  36740. [0.6239, 0.4107, 0.8162, 0.2763, 0.3625, 0.3600, 0.5988, 0.5700],
  36741. [0.6102, 0.4005, 0.8688, 0.5100, 0.4813, 0.5400, 0.5404, 0.5064],
  36742. [0.6160, 0.4086, 0.8775, 0.3983, 0.3750, 0.3783, 0.5128, 0.4882],
  36743. [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600],
  36744. [0.6204, 0.4049, 0.7975, 0.2700, 0.3937, 0.2567, 0.5700, 0.5183]],
  36745. device='cuda:0', dtype=torch.float64)
  36746. predictions are: tensor([[ 0.6829, -0.3401, 1.7066, 0.2011, -0.3595, 0.1086, 0.9120, 0.4105],
  36747. [ 0.8049, -0.2839, 1.7544, -0.5074, -0.5299, 0.0459, 0.5987, 0.2734],
  36748. [ 0.5709, -0.4311, 1.8545, -0.0998, -0.4727, -0.0623, 0.4352, 0.2658],
  36749. [ 0.4772, -0.4543, 1.5151, -0.7104, -0.5211, -0.7383, 0.4856, 0.3452],
  36750. [ 0.5559, -0.4322, 1.7555, 0.1990, -0.0483, 0.0407, 0.1077, 0.1094],
  36751. [ 0.5763, -0.4048, 1.8463, -0.2421, -0.5570, -0.6607, 0.1426, 0.0537],
  36752. [ 0.6002, -0.3517, 1.1491, -0.9855, -0.5656, -0.8767, 0.4835, 0.3959],
  36753. [ 0.1484, -0.6880, 1.5266, -0.8485, -0.4765, -1.2423, 0.2979, 0.1321]],
  36754. device='cuda:0', grad_fn=<AddmmBackward>)
  36755. landmarks are: tensor([[[ 0.6421, -0.3816, 1.7037, 0.1929, -0.4037, 0.2391, 1.1861,
  36756. 0.2249],
  36757. [ 0.5841, -0.4199, 1.7961, -0.6693, -0.6231, 0.1082, 0.6529,
  36758. 0.1159],
  36759. [ 0.5888, -0.4101, 1.8654, -0.2998, -0.5134, -0.0919, 0.5374,
  36760. 0.2468],
  36761. [ 0.5960, -0.3888, 1.4840, -1.0095, -0.6115, -0.6231, 0.4797,
  36762. 0.3469],
  36763. [ 0.5328, -0.4361, 1.7268, 0.0697, -0.0630, 0.2083, 0.2103,
  36764. 0.0532],
  36765. [ 0.5595, -0.3988, 1.7672, -0.4460, -0.5538, -0.5384, 0.0828,
  36766. -0.0310],
  36767. [ 0.5466, -0.4080, 1.0668, -1.1764, -0.6289, -0.7617, 0.4855,
  36768. 0.3007],
  36769. [ 0.5798, -0.4156, 1.3977, -1.0388, -0.4672, -1.1004, 0.3469,
  36770. 0.1082]]], device='cuda:0')
  36771. loss_train_step before backward: tensor(0.0156, device='cuda:0', grad_fn=<MseLossBackward>)
  36772. loss_train_step after backward: tensor(0.0156, device='cuda:0', grad_fn=<MseLossBackward>)
  36773. loss_train: 0.9222497092559934
  36774. step: 57
  36775. running loss: 0.01617981946063146
  36776. Train Steps: 57/90 Loss: 0.0162 torch.Size([8, 600, 800])
  36777. torch.Size([8, 8])
  36778. tensor([[0.6204, 0.4013, 0.8075, 0.2400, 0.4313, 0.2050, 0.5800, 0.5150],
  36779. [0.6263, 0.4039, 0.9000, 0.4400, 0.3625, 0.4667, 0.6424, 0.4804],
  36780. [0.6189, 0.4029, 0.8375, 0.5767, 0.4745, 0.4829, 0.5551, 0.5598],
  36781. [0.6266, 0.4067, 0.8588, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267],
  36782. [0.6040, 0.4002, 0.7338, 0.2267, 0.3975, 0.2100, 0.5231, 0.4778],
  36783. [0.6197, 0.4051, 0.7812, 0.2650, 0.3513, 0.4050, 0.6112, 0.5500],
  36784. [0.6216, 0.4099, 0.7225, 0.2033, 0.4188, 0.2217, 0.5975, 0.5283],
  36785. [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5413, 0.5433]],
  36786. device='cuda:0', dtype=torch.float64)
  36787. predictions are: tensor([[ 0.4680, -0.5072, 1.4882, -0.9610, -0.2499, -1.2891, 0.5355, 0.2191],
  36788. [ 0.7412, -0.3457, 1.8500, -0.2045, -0.5993, -0.0566, 0.4874, 0.0658],
  36789. [ 0.4956, -0.4700, 1.5806, 0.3873, -0.1192, -0.0546, 0.2357, 0.3602],
  36790. [-0.0531, -0.8100, 1.7826, -0.8371, -0.3113, -1.0016, 0.6913, 0.2936],
  36791. [ 0.5888, -0.4092, 1.1607, -1.1879, -0.4697, -1.2295, 0.2869, 0.0801],
  36792. [ 0.7139, -0.3403, 1.3548, -0.9276, -0.6617, -0.3502, 0.5011, 0.3473],
  36793. [ 0.2308, -0.6215, 1.1084, -1.1688, -0.3470, -1.2640, 0.4619, 0.3078],
  36794. [ 0.5495, -0.4039, 1.7657, -0.5195, -0.6240, -0.7107, 0.2321, 0.2657]],
  36795. device='cuda:0', grad_fn=<AddmmBackward>)
  36796. landmarks are: tensor([[[ 0.5796, -0.4326, 1.4439, -1.1774, -0.2940, -1.3390, 0.3931,
  36797. 0.0928],
  36798. [ 0.6072, -0.4206, 1.8711, -0.2536, -0.6115, -0.1304, 0.6812,
  36799. -0.0670],
  36800. [ 0.5726, -0.4249, 1.5824, 0.3777, -0.0942, -0.0556, 0.2782,
  36801. 0.2997],
  36802. [ 0.6084, -0.4076, 1.6806, -0.9618, -0.2998, -0.9695, 0.6356,
  36803. 0.1467],
  36804. [ 0.5037, -0.4375, 1.1032, -1.2390, -0.4499, -1.3159, 0.1301,
  36805. -0.0791],
  36806. [ 0.5763, -0.4147, 1.3226, -1.0619, -0.6635, -0.4152, 0.5374,
  36807. 0.2545],
  36808. [ 0.5850, -0.3925, 1.0513, -1.3467, -0.3517, -1.2620, 0.4739,
  36809. 0.1544],
  36810. [ 0.5894, -0.3479, 1.7730, -0.6847, -0.5538, -0.7155, 0.2141,
  36811. 0.2237]]], device='cuda:0')
  36812. loss_train_step before backward: tensor(0.0198, device='cuda:0', grad_fn=<MseLossBackward>)
  36813. loss_train_step after backward: tensor(0.0198, device='cuda:0', grad_fn=<MseLossBackward>)
  36814. loss_train: 0.9420179659500718
  36815. step: 58
  36816. running loss: 0.016241689068104685
  36817. Train Steps: 58/90 Loss: 0.0162 torch.Size([8, 600, 800])
  36818. torch.Size([8, 8])
  36819. tensor([[0.6193, 0.3930, 0.8949, 0.4437, 0.3852, 0.5435, 0.6263, 0.5263],
  36820. [0.6260, 0.4214, 0.8538, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983],
  36821. [0.6126, 0.4039, 0.8237, 0.3967, 0.3625, 0.3600, 0.5894, 0.6138],
  36822. [ nan, nan, 0.6412, 0.1900, 0.4238, 0.1883, 0.5487, 0.5700],
  36823. [0.6199, 0.3952, 0.8573, 0.5374, 0.4075, 0.4687, 0.5942, 0.5377],
  36824. [0.6185, 0.4098, 0.8838, 0.4900, 0.4537, 0.5800, 0.6288, 0.5400],
  36825. [0.6136, 0.4085, 0.6688, 0.2317, 0.3862, 0.2367, 0.5517, 0.5783],
  36826. [0.6204, 0.4091, 0.8950, 0.4783, 0.3613, 0.4617, 0.6063, 0.5617]],
  36827. device='cuda:0', dtype=torch.float64)
  36828. predictions are: tensor([[ 0.5118, -0.4744, 1.7803, -0.3263, -0.4744, 0.1388, 0.4503, 0.0420],
  36829. [ 0.4170, -0.4748, 1.6589, 0.1730, -0.5230, -0.5177, 0.1883, 0.3262],
  36830. [ 0.4429, -0.4836, 1.5923, -0.5252, -0.5800, -0.5985, 0.3619, 0.4124],
  36831. [-2.6916, -2.5599, 0.9813, -1.2277, -0.3173, -1.4307, 0.2689, 0.2390],
  36832. [ 0.5146, -0.5080, 1.7030, 0.0749, -0.3763, -0.1399, 0.4489, 0.0463],
  36833. [ 0.4605, -0.4835, 1.7907, -0.1342, -0.1387, 0.3505, 0.4464, 0.1572],
  36834. [ 0.4982, -0.4772, 0.9685, -1.1748, -0.4223, -1.1938, 0.3804, 0.2206],
  36835. [ 0.5921, -0.4157, 1.8043, -0.1693, -0.5291, -0.1697, 0.5353, 0.2092]],
  36836. device='cuda:0', grad_fn=<AddmmBackward>)
  36837. landmarks are: tensor([[[ 0.5746, -0.4706, 1.8476, -0.2365, -0.5068, 0.2245, 0.6069,
  36838. 0.1449],
  36839. [ 0.6055, -0.3393, 1.6575, 0.2545, -0.5942, -0.5461, 0.2949,
  36840. 0.4778],
  36841. [ 0.5438, -0.4206, 1.5189, -0.4537, -0.6115, -0.6231, 0.4365,
  36842. 0.5491],
  36843. [-2.2859, -2.2859, 0.6760, -1.4083, -0.3286, -1.4160, 0.2487,
  36844. 0.3469],
  36845. [ 0.5775, -0.4607, 1.6741, 0.1962, -0.4036, -0.1212, 0.4588,
  36846. 0.1979],
  36847. [ 0.5709, -0.3933, 1.7961, -0.0226, -0.1901, 0.3931, 0.6182,
  36848. 0.2083],
  36849. [ 0.5483, -0.3991, 0.8030, -1.2159, -0.5018, -1.1928, 0.2624,
  36850. 0.3852],
  36851. [ 0.5797, -0.3965, 1.8480, -0.0765, -0.6173, -0.1535, 0.5143,
  36852. 0.3084]]], device='cuda:0')
  36853. loss_train_step before backward: tensor(0.0128, device='cuda:0', grad_fn=<MseLossBackward>)
  36854. loss_train_step after backward: tensor(0.0128, device='cuda:0', grad_fn=<MseLossBackward>)
  36855. loss_train: 0.95481295324862
  36856. step: 59
  36857. running loss: 0.016183270394044408
  36858.  
  36859. Train Steps: 59/90 Loss: 0.0162 torch.Size([8, 600, 800])
  36860. torch.Size([8, 8])
  36861. tensor([[0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689],
  36862. [0.6222, 0.4172, 0.8850, 0.5217, 0.3738, 0.4600, 0.5700, 0.5633],
  36863. [0.6266, 0.4070, 0.8712, 0.5600, 0.3713, 0.4783, 0.5775, 0.6100],
  36864. [0.6048, 0.3928, 0.8538, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719],
  36865. [0.6307, 0.4060, 0.8950, 0.5183, 0.3750, 0.3850, 0.6338, 0.4938],
  36866. [0.6084, 0.3981, 0.8588, 0.5233, 0.4600, 0.5367, 0.5680, 0.5006],
  36867. [0.6178, 0.4059, 0.7525, 0.2250, 0.4313, 0.1783, 0.5404, 0.5240],
  36868. [0.6134, 0.4090, 0.6926, 0.2819, 0.3538, 0.3233, 0.5563, 0.5667]],
  36869. device='cuda:0', dtype=torch.float64)
  36870. predictions are: tensor([[ 0.6028, -0.3939, 1.3621, -0.7628, -0.6650, -0.5480, 0.2328, 0.2807],
  36871. [ 0.6330, -0.3729, 1.8560, -0.0090, -0.5082, -0.1894, 0.3534, 0.3002],
  36872. [ 0.6191, -0.4331, 1.6896, 0.1688, -0.5121, -0.0415, 0.5741, 0.4011],
  36873. [ 0.1725, -0.7107, 1.6695, 0.1222, -0.4347, 0.0672, 0.2956, 0.1270],
  36874. [ 0.5586, -0.4837, 1.9110, -0.0396, -0.5220, -0.5785, 0.6986, -0.0429],
  36875. [ 0.4333, -0.5479, 1.6927, 0.0067, -0.1288, 0.1918, 0.2718, 0.0897],
  36876. [ 0.2199, -0.6341, 1.2996, -1.2787, -0.2926, -1.4296, 0.4385, 0.1184],
  36877. [ 0.6418, -0.3789, 1.0555, -1.0573, -0.6544, -0.7649, 0.3476, 0.3257]],
  36878. device='cuda:0', grad_fn=<AddmmBackward>)
  36879. landmarks are: tensor([[[ 0.5327, -0.4381, 1.2880, -0.6308, -0.6866, -0.5307, 0.2658,
  36880. 0.3417],
  36881. [ 0.5879, -0.3591, 1.8018, 0.1236, -0.5596, -0.1612, 0.3469,
  36882. 0.3161],
  36883. [ 0.6081, -0.4059, 1.7383, 0.3007, -0.5711, -0.0765, 0.3815,
  36884. 0.5316],
  36885. [ 0.5079, -0.4714, 1.6575, 0.2237, -0.4961, 0.0774, 0.1465,
  36886. -0.1061],
  36887. [ 0.6273, -0.4105, 1.8480, 0.1082, -0.5538, -0.5076, 0.6414,
  36888. -0.0049],
  36889. [ 0.5241, -0.4470, 1.6806, 0.1313, -0.1612, 0.1929, 0.3378,
  36890. 0.0261],
  36891. [ 0.5676, -0.4112, 1.1898, -1.2467, -0.2940, -1.4622, 0.2103,
  36892. 0.1343],
  36893. [ 0.5473, -0.3966, 0.9131, -0.9838, -0.6520, -0.7925, 0.2834,
  36894. 0.3315]]], device='cuda:0')
  36895. loss_train_step before backward: tensor(0.0120, device='cuda:0', grad_fn=<MseLossBackward>)
  36896. loss_train_step after backward: tensor(0.0120, device='cuda:0', grad_fn=<MseLossBackward>)
  36897. loss_train: 0.9668285604566336
  36898. step: 60
  36899. running loss: 0.016113809340943894
  36900. Train Steps: 60/90 Loss: 0.0161 torch.Size([8, 600, 800])
  36901. torch.Size([8, 8])
  36902. tensor([[0.6189, 0.3961, 0.7589, 0.2256, 0.3759, 0.3280, 0.6184, 0.5334],
  36903. [0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407],
  36904. [0.6364, 0.4092, 0.8525, 0.5633, 0.3875, 0.4950, 0.6599, 0.5285],
  36905. [0.6212, 0.4159, 0.8675, 0.5783, 0.4088, 0.4317, 0.5613, 0.5917],
  36906. [0.6122, 0.3993, 0.8738, 0.4667, 0.4517, 0.4879, 0.5155, 0.4927],
  36907. [0.6275, 0.4003, 0.9100, 0.3783, 0.4388, 0.3133, 0.7058, 0.5343],
  36908. [0.6124, 0.4083, 0.6954, 0.3069, 0.3650, 0.2750, 0.5163, 0.5485],
  36909. [0.6307, 0.3998, 0.8500, 0.5233, 0.3850, 0.5050, 0.7446, 0.5320]],
  36910. device='cuda:0', dtype=torch.float64)
  36911. predictions are: tensor([[ 0.5016, -0.4703, 1.2278, -1.2672, -0.5932, -0.8244, 0.4420, 0.1547],
  36912. [ 0.2497, -0.6129, 1.6548, -0.4681, -0.5996, -0.8265, -0.0047, 0.1981],
  36913. [ 0.5524, -0.4908, 1.6442, 0.1523, -0.4874, 0.0100, 0.6921, 0.1528],
  36914. [ 0.4261, -0.4861, 1.6044, 0.1678, -0.3794, -0.2528, 0.2993, 0.4646],
  36915. [ 0.5280, -0.4921, 1.7713, -0.2128, -0.2425, 0.0598, 0.1536, 0.0179],
  36916. [ 0.4659, -0.5584, 1.8527, -0.6520, -0.3224, -0.8461, 0.8426, 0.0868],
  36917. [ 0.4742, -0.4619, 0.9765, -0.9842, -0.6332, -0.9425, 0.0351, 0.2549],
  36918. [ 0.5845, -0.4672, 1.6357, 0.0128, -0.4951, 0.0762, 0.8753, 0.1771]],
  36919. device='cuda:0', grad_fn=<AddmmBackward>)
  36920. landmarks are: tensor([[[ 5.7296e-01, -4.5619e-01, 1.2195e+00, -1.2440e+00, -5.4966e-01,
  36921. -7.7109e-01, 5.7045e-01, 1.7788e-01],
  36922. [ 5.4700e-01, -3.9515e-01, 1.6377e+00, -4.2531e-01, -6.2887e-01,
  36923. -8.0785e-01, 2.4925e-02, 2.1157e-01],
  36924. [ 6.5365e-01, -3.9601e-01, 1.6517e+00, 3.1609e-01, -4.9607e-01,
  36925. 4.6189e-04, 7.6203e-01, 1.5543e-01],
  36926. [ 5.8360e-01, -3.6490e-01, 1.7210e+00, 3.8537e-01, -3.9792e-01,
  36927. -2.9207e-01, 3.0647e-01, 4.4696e-01],
  36928. [ 5.4166e-01, -4.4175e-01, 1.7499e+00, -1.3041e-01, -1.9942e-01,
  36929. -3.2367e-02, 9.5140e-02, -9.9401e-03],
  36930. [ 6.1248e-01, -4.3693e-01, 1.9173e+00, -5.3841e-01, -2.5935e-01,
  36931. -8.3865e-01, 9.7406e-01, 1.8214e-01],
  36932. [ 5.4249e-01, -3.9977e-01, 9.2628e-01, -8.6826e-01, -6.0000e-01,
  36933. -1.0157e+00, 9.8951e-02, 2.4764e-01],
  36934. [ 6.2730e-01, -4.3934e-01, 1.6402e+00, 1.3133e-01, -5.0762e-01,
  36935. 4.6651e-02, 1.1532e+00, 1.7146e-01]]], device='cuda:0')
  36936. loss_train_step before backward: tensor(0.0087, device='cuda:0', grad_fn=<MseLossBackward>)
  36937. loss_train_step after backward: tensor(0.0087, device='cuda:0', grad_fn=<MseLossBackward>)
  36938. loss_train: 0.9755781814455986
  36939. step: 61
  36940. running loss: 0.015993084941731125
  36941. Train Steps: 61/90 Loss: 0.0160 torch.Size([8, 600, 800])
  36942. torch.Size([8, 8])
  36943. tensor([[0.6079, 0.3964, 0.7420, 0.2958, 0.3563, 0.2917, 0.5351, 0.4980],
  36944. [ nan, nan, 0.8488, 0.2300, 0.5563, 0.2100, 0.7390, 0.5679],
  36945. [0.6192, 0.4128, 0.8513, 0.5617, 0.4200, 0.5267, 0.5850, 0.5633],
  36946. [0.6182, 0.3972, 0.8552, 0.5914, 0.3683, 0.4181, 0.5688, 0.5378],
  36947. [0.6111, 0.4019, 0.8350, 0.3867, 0.3500, 0.4283, 0.5480, 0.5719],
  36948. [0.6219, 0.4089, 0.8938, 0.4800, 0.3825, 0.4450, 0.5850, 0.5200],
  36949. [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153],
  36950. [0.6147, 0.4081, 0.8538, 0.3400, 0.3663, 0.3150, 0.5142, 0.4875]],
  36951. device='cuda:0', dtype=torch.float64)
  36952. predictions are: tensor([[ 0.4429, -0.5028, 1.0489, -1.0283, -0.5719, -0.9001, 0.1954, 0.0527],
  36953. [-2.7236, -2.5519, 1.5574, -1.1088, 0.1584, -1.0457, 1.0681, 0.3034],
  36954. [ 0.5035, -0.4616, 1.5655, 0.1601, -0.3356, 0.1247, 0.5148, 0.2402],
  36955. [ 0.3959, -0.5453, 1.4629, 0.1522, -0.5350, -0.4185, 0.3992, 0.2318],
  36956. [ 0.3227, -0.5639, 1.4088, -0.5852, -0.6078, -0.3931, 0.1804, 0.2821],
  36957. [ 0.5485, -0.4297, 1.6632, -0.2612, -0.4895, -0.2028, 0.5109, 0.1559],
  36958. [ 0.5043, -0.4713, 1.2385, -0.8179, -0.6154, -0.6846, 0.3148, 0.1545],
  36959. [ 0.5604, -0.4319, 1.6134, -0.8001, -0.5624, -0.7814, 0.1643, 0.0606]],
  36960. device='cuda:0', grad_fn=<AddmmBackward>)
  36961. landmarks are: tensor([[[ 0.5219, -0.4550, 1.1415, -0.9196, -0.6404, -0.9387, 0.1856,
  36962. 0.0141],
  36963. [-2.2859, -2.2859, 1.6344, -1.2236, 0.2834, -1.3159, 1.1276,
  36964. 0.3371],
  36965. [ 0.5742, -0.3792, 1.6460, 0.3084, -0.3460, 0.1467, 0.4162,
  36966. 0.3161],
  36967. [ 0.5697, -0.4514, 1.6642, 0.4459, -0.5850, -0.3546, 0.3414,
  36968. 0.1982],
  36969. [ 0.5367, -0.4294, 1.5709, -0.4999, -0.6693, -0.3075, 0.2455,
  36970. 0.3559],
  36971. [ 0.5865, -0.3973, 1.8423, -0.0688, -0.5192, -0.2305, 0.4162,
  36972. 0.1159],
  36973. [ 0.5614, -0.4514, 1.3252, -0.7240, -0.6237, -0.7215, 0.4299,
  36974. 0.0942],
  36975. [ 0.5532, -0.4008, 1.6575, -0.7155, -0.5942, -0.8309, 0.0890,
  36976. -0.0340]]], device='cuda:0')
  36977. loss_train_step before backward: tensor(0.0143, device='cuda:0', grad_fn=<MseLossBackward>)
  36978. loss_train_step after backward: tensor(0.0143, device='cuda:0', grad_fn=<MseLossBackward>)
  36979. loss_train: 0.989847207441926
  36980. step: 62
  36981. running loss: 0.015965277539385904
  36982. Train Steps: 62/90 Loss: 0.0160 torch.Size([8, 600, 800])
  36983. torch.Size([8, 8])
  36984. tensor([[0.6114, 0.4018, 0.7213, 0.1967, 0.3763, 0.2700, 0.5875, 0.5533],
  36985. [0.6225, 0.4116, 0.8662, 0.3517, 0.3663, 0.3233, 0.5837, 0.5317],
  36986. [0.6271, 0.4040, 0.9138, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413],
  36987. [0.6332, 0.4165, 0.9100, 0.3350, 0.4188, 0.3683, 0.7438, 0.5528],
  36988. [ nan, nan, 0.8675, 0.2833, 0.5350, 0.1983, 0.6678, 0.5621],
  36989. [0.6200, 0.4118, 0.8287, 0.4017, 0.3775, 0.2833, 0.5391, 0.5799],
  36990. [0.6261, 0.4045, 0.8865, 0.5369, 0.3895, 0.4859, 0.6683, 0.5249],
  36991. [0.6198, 0.4075, 0.8650, 0.5617, 0.4150, 0.4367, 0.5450, 0.5650]],
  36992. device='cuda:0', dtype=torch.float64)
  36993. predictions are: tensor([[ 5.8423e-01, -4.0921e-01, 8.4051e-01, -1.3952e+00, -5.2819e-01,
  36994. -9.3486e-01, 2.4505e-01, 2.1930e-01],
  36995. [ 6.5671e-01, -3.4788e-01, 1.5112e+00, -7.5554e-01, -6.2020e-01,
  36996. -6.4918e-01, 2.0874e-01, 1.8594e-01],
  36997. [ 4.3317e-01, -5.3625e-01, 1.7903e+00, -5.5146e-01, -1.9873e-01,
  36998. -9.7180e-01, 7.4523e-01, 9.3876e-02],
  36999. [ 2.9301e-01, -5.9894e-01, 1.7694e+00, -6.2905e-01, -4.0333e-01,
  37000. -5.2115e-01, 8.2788e-01, 2.0718e-01],
  37001. [-2.7556e+00, -2.5627e+00, 1.4485e+00, -1.1333e+00, -9.4608e-04,
  37002. -1.1536e+00, 7.4314e-01, 2.3110e-01],
  37003. [ 3.9862e-01, -4.9696e-01, 1.3710e+00, -4.8671e-01, -5.6165e-01,
  37004. -8.6880e-01, 1.3935e-02, 3.0212e-01],
  37005. [ 6.2111e-01, -4.1407e-01, 1.5811e+00, 4.3570e-02, -5.5322e-01,
  37006. 2.0704e-02, 6.7610e-01, 4.8456e-02],
  37007. [ 3.0005e-01, -5.6617e-01, 1.4875e+00, 2.0248e-01, -4.3813e-01,
  37008. -2.1301e-01, 2.2350e-01, 2.4070e-01]], device='cuda:0',
  37009. grad_fn=<AddmmBackward>)
  37010. landmarks are: tensor([[[ 0.5380, -0.4302, 1.0455, -1.3775, -0.5480, -1.0388, 0.4277,
  37011. 0.2699],
  37012. [ 0.5893, -0.3847, 1.7152, -0.6616, -0.5942, -0.7925, 0.4104,
  37013. 0.1698],
  37014. [ 0.6108, -0.4201, 1.9346, -0.5538, -0.1497, -1.0773, 1.0545,
  37015. 0.2142],
  37016. [ 0.6388, -0.3623, 1.9173, -0.7386, -0.3517, -0.5846, 1.1495,
  37017. 0.2676],
  37018. [-2.2859, -2.2859, 1.7210, -0.9772, 0.1852, -1.3698, 0.7986,
  37019. 0.3104],
  37020. [ 0.5779, -0.3840, 1.5420, -0.4306, -0.5423, -0.9772, 0.2041,
  37021. 0.3928],
  37022. [ 0.6059, -0.4177, 1.8087, 0.1941, -0.4868, -0.0414, 0.8010,
  37023. 0.1385],
  37024. [ 0.5770, -0.4036, 1.7095, 0.3084, -0.3691, -0.2690, 0.2314,
  37025. 0.3238]]], device='cuda:0')
  37026. loss_train_step before backward: tensor(0.0255, device='cuda:0', grad_fn=<MseLossBackward>)
  37027. loss_train_step after backward: tensor(0.0255, device='cuda:0', grad_fn=<MseLossBackward>)
  37028. loss_train: 1.015299854800105
  37029. step: 63
  37030. running loss: 0.01611587071111278
  37031.  
  37032. Train Steps: 63/90 Loss: 0.0161 torch.Size([8, 600, 800])
  37033. torch.Size([8, 8])
  37034. tensor([[0.6086, 0.4019, 0.7782, 0.4278, 0.3625, 0.4350, 0.5150, 0.5285],
  37035. [0.6143, 0.4034, 0.8800, 0.4833, 0.4512, 0.5367, 0.5289, 0.5097],
  37036. [0.6182, 0.3987, 0.7878, 0.2889, 0.3699, 0.3260, 0.6086, 0.5367],
  37037. [0.6095, 0.3970, 0.8688, 0.4767, 0.4860, 0.4879, 0.5191, 0.4940],
  37038. [0.6170, 0.4102, 0.7468, 0.3695, 0.3463, 0.3767, 0.5238, 0.5823],
  37039. [0.6296, 0.4008, 0.9150, 0.4317, 0.4263, 0.3050, 0.7256, 0.5413],
  37040. [0.6266, 0.4067, 0.8588, 0.2867, 0.4300, 0.2850, 0.6325, 0.5267],
  37041. [0.6289, 0.4019, 0.8113, 0.5467, 0.3875, 0.5017, 0.7367, 0.5332]],
  37042. device='cuda:0', dtype=torch.float64)
  37043. predictions are: tensor([[ 0.4983, -0.4283, 1.2735, -0.4185, -0.6757, -0.3920, -0.0064, 0.1818],
  37044. [ 0.4463, -0.4745, 1.7635, -0.0741, -0.2198, 0.0709, 0.3045, 0.0067],
  37045. [ 0.4358, -0.5093, 1.3346, -1.0143, -0.6237, -0.7973, 0.4648, 0.2807],
  37046. [ 0.4776, -0.4860, 1.7109, -0.0809, -0.1550, -0.0845, 0.2076, 0.0665],
  37047. [ 0.3352, -0.4959, 1.1055, -0.6256, -0.7360, -0.6334, 0.1003, 0.4328],
  37048. [ 0.3745, -0.5841, 1.9757, -0.3037, -0.3375, -0.9440, 1.0600, 0.1346],
  37049. [ 0.7117, -0.3051, 1.6728, -0.9691, -0.3679, -1.0242, 0.5702, 0.1927],
  37050. [ 0.6047, -0.4263, 1.4712, 0.2078, -0.5265, -0.0830, 1.0713, 0.1632]],
  37051. device='cuda:0', grad_fn=<AddmmBackward>)
  37052. landmarks are: tensor([[[ 0.5253, -0.4296, 1.3084, -0.3098, -0.6115, -0.2767, 0.0928,
  37053. 0.1552],
  37054. [ 0.5515, -0.4225, 1.7788, -0.0534, -0.2016, 0.1929, 0.1568,
  37055. 0.0682],
  37056. [ 0.5697, -0.4442, 1.3529, -0.9515, -0.5774, -0.7801, 0.5253,
  37057. 0.1931],
  37058. [ 0.5292, -0.4520, 1.7268, -0.0842, -0.0413, -0.0324, 0.1116,
  37059. -0.0039],
  37060. [ 0.5639, -0.3911, 1.1634, -0.5794, -0.6866, -0.5461, 0.1334,
  37061. 0.4036],
  37062. [ 0.6224, -0.4345, 1.9404, -0.2921, -0.3171, -0.8771, 1.0655,
  37063. 0.2142],
  37064. [ 0.6084, -0.4076, 1.6806, -0.9618, -0.2998, -0.9695, 0.6356,
  37065. 0.1467],
  37066. [ 0.6191, -0.4297, 1.4612, 0.2391, -0.4961, 0.0313, 1.1166,
  37067. 0.1768]]], device='cuda:0')
  37068. loss_train_step before backward: tensor(0.0063, device='cuda:0', grad_fn=<MseLossBackward>)
  37069. loss_train_step after backward: tensor(0.0063, device='cuda:0', grad_fn=<MseLossBackward>)
  37070. loss_train: 1.021567848045379
  37071. step: 64
  37072. running loss: 0.015961997625709046
  37073. Train Steps: 64/90 Loss: 0.0160 torch.Size([8, 600, 800])
  37074. torch.Size([8, 8])
  37075. tensor([[0.6260, 0.4161, 0.9000, 0.3833, 0.4450, 0.2133, 0.6237, 0.4986],
  37076. [0.6175, 0.4091, 0.7863, 0.2800, 0.3638, 0.3583, 0.6188, 0.5433],
  37077. [0.6179, 0.3993, 0.8925, 0.4789, 0.3879, 0.4900, 0.6041, 0.5279],
  37078. [0.6250, 0.4106, 0.8700, 0.3717, 0.3588, 0.4967, 0.6038, 0.5167],
  37079. [ nan, nan, 0.8488, 0.2300, 0.5563, 0.2100, 0.7390, 0.5679],
  37080. [0.6125, 0.3983, 0.8750, 0.4867, 0.4275, 0.4783, 0.5413, 0.5717],
  37081. [0.6214, 0.3982, 0.8938, 0.4517, 0.3663, 0.4083, 0.5863, 0.5050],
  37082. [0.6185, 0.4067, 0.8838, 0.4450, 0.4037, 0.4733, 0.5213, 0.5142]],
  37083. device='cuda:0', dtype=torch.float64)
  37084. predictions are: tensor([[ 0.4823, -0.4508, 1.8019, -0.3918, -0.2574, -1.2008, 0.5459, 0.0978],
  37085. [ 0.5608, -0.3837, 1.1220, -1.0028, -0.6592, -0.5652, 0.3373, 0.2505],
  37086. [ 0.4518, -0.4826, 1.6649, -0.0459, -0.5189, -0.0822, 0.5264, 0.1332],
  37087. [ 0.6273, -0.3566, 1.5237, -0.4565, -0.6031, -0.0520, 0.6042, 0.1593],
  37088. [-2.3707, -2.2975, 1.5311, -1.0986, 0.2021, -1.1821, 1.1018, 0.3396],
  37089. [ 0.4574, -0.4729, 1.5861, 0.0844, -0.3624, -0.0815, 0.3012, 0.2289],
  37090. [ 0.5393, -0.4603, 1.6396, -0.0924, -0.6100, -0.3728, 0.4136, 0.0178],
  37091. [ 0.6332, -0.3491, 1.5950, -0.1174, -0.4494, -0.1350, 0.1423, 0.0673]],
  37092. device='cuda:0', grad_fn=<AddmmBackward>)
  37093. landmarks are: tensor([[[ 0.6054, -0.3641, 1.8711, -0.5153, -0.2305, -1.3005, 0.5948,
  37094. 0.0171],
  37095. [ 0.5665, -0.3963, 1.3457, -0.9926, -0.6058, -0.6308, 0.5721,
  37096. 0.2237],
  37097. [ 0.5680, -0.4417, 1.8365, -0.0740, -0.4941, -0.0227, 0.5045,
  37098. 0.1525],
  37099. [ 0.6010, -0.3896, 1.7326, -0.5692, -0.6289, 0.0082, 0.5028,
  37100. 0.1005],
  37101. [-2.2859, -2.2859, 1.6344, -1.2236, 0.2834, -1.3159, 1.1276,
  37102. 0.3371],
  37103. [ 0.5432, -0.4462, 1.7557, -0.0380, -0.3113, -0.0765, 0.2141,
  37104. 0.3546],
  37105. [ 0.5844, -0.4466, 1.8423, -0.1997, -0.5942, -0.3998, 0.4219,
  37106. 0.0467],
  37107. [ 0.5708, -0.4075, 1.7961, -0.2305, -0.4210, -0.0996, 0.1219,
  37108. 0.0893]]], device='cuda:0')
  37109. loss_train_step before backward: tensor(0.0087, device='cuda:0', grad_fn=<MseLossBackward>)
  37110. loss_train_step after backward: tensor(0.0087, device='cuda:0', grad_fn=<MseLossBackward>)
  37111. loss_train: 1.0303018535487354
  37112. step: 65
  37113. running loss: 0.015850797746903622
  37114. Train Steps: 65/90 Loss: 0.0159 torch.Size([8, 600, 800])
  37115. torch.Size([8, 8])
  37116. tensor([[0.6070, 0.3979, 0.8575, 0.5083, 0.4350, 0.5400, 0.5180, 0.5466],
  37117. [0.6185, 0.4067, 0.8838, 0.4450, 0.4037, 0.4733, 0.5213, 0.5142],
  37118. [0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742],
  37119. [0.6201, 0.4151, 0.8588, 0.5467, 0.3700, 0.3950, 0.5637, 0.5933],
  37120. [0.6243, 0.4128, 0.7762, 0.2717, 0.3825, 0.3133, 0.6212, 0.5417],
  37121. [0.6257, 0.4034, 0.8287, 0.2333, 0.3925, 0.2717, 0.6330, 0.4901],
  37122. [0.6332, 0.4128, 0.9200, 0.3517, 0.4400, 0.3833, 0.7461, 0.5494],
  37123. [0.6160, 0.4093, 0.8314, 0.4417, 0.3675, 0.4583, 0.5250, 0.5808]],
  37124. device='cuda:0', dtype=torch.float64)
  37125. predictions are: tensor([[ 0.5700, -0.4073, 1.6418, 0.1862, -0.2623, 0.1250, 0.4004, 0.1653],
  37126. [ 0.7220, -0.3087, 1.7366, -0.1382, -0.4315, -0.1361, 0.2370, 0.0559],
  37127. [ 0.4944, -0.4977, 1.6870, -0.6288, -0.4603, -1.0396, 0.5128, -0.0098],
  37128. [ 0.5891, -0.3530, 1.6456, 0.2662, -0.5829, -0.4358, 0.4185, 0.4307],
  37129. [ 0.7564, -0.2906, 1.2859, -1.0564, -0.4996, -0.8074, 0.6509, 0.2876],
  37130. [ 0.5688, -0.4439, 1.4126, -1.1453, -0.4822, -0.9829, 0.5486, -0.0029],
  37131. [ 0.5284, -0.4687, 1.8423, -0.5432, -0.2972, -0.5283, 1.1121, 0.2081],
  37132. [ 0.4286, -0.4599, 1.5429, -0.1226, -0.6268, -0.2642, 0.2406, 0.3169]],
  37133. device='cuda:0', grad_fn=<AddmmBackward>)
  37134. landmarks are: tensor([[[ 0.5178, -0.4480, 1.6748, 0.0620, -0.2767, 0.2083, 0.1067,
  37135. 0.2386],
  37136. [ 0.5708, -0.4075, 1.7961, -0.2305, -0.4210, -0.0996, 0.1219,
  37137. 0.0893],
  37138. [ 0.6134, -0.4218, 1.7268, -0.6154, -0.4730, -1.0850, 0.5463,
  37139. -0.0957],
  37140. [ 0.5785, -0.3687, 1.6806, 0.2391, -0.5769, -0.4614, 0.3180,
  37141. 0.4547],
  37142. [ 0.5977, -0.3792, 1.2995, -1.0311, -0.5192, -0.8386, 0.5836,
  37143. 0.2160],
  37144. [ 0.6042, -0.4225, 1.5420, -1.2082, -0.4730, -1.0311, 0.6380,
  37145. -0.0220],
  37146. [ 0.6388, -0.3792, 1.9635, -0.6616, -0.2536, -0.5153, 1.1605,
  37147. 0.2516],
  37148. [ 0.5592, -0.3956, 1.5543, -0.2456, -0.5885, -0.1689, 0.1392,
  37149. 0.3968]]], device='cuda:0')
  37150. loss_train_step before backward: tensor(0.0066, device='cuda:0', grad_fn=<MseLossBackward>)
  37151. loss_train_step after backward: tensor(0.0066, device='cuda:0', grad_fn=<MseLossBackward>)
  37152. loss_train: 1.0369146969169378
  37153. step: 66
  37154. running loss: 0.015710828741165726
  37155. Train Steps: 66/90 Loss: 0.0157 torch.Size([8, 600, 800])
  37156. torch.Size([8, 8])
  37157. tensor([[0.6136, 0.4085, 0.6688, 0.2317, 0.3862, 0.2367, 0.5517, 0.5783],
  37158. [0.6136, 0.4117, 0.8700, 0.5167, 0.4188, 0.5083, 0.5147, 0.5495],
  37159. [0.6299, 0.4303, 0.7963, 0.3933, 0.4850, 0.2283, 0.5480, 0.6222],
  37160. [0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750],
  37161. [0.6196, 0.4094, 0.7562, 0.2817, 0.3937, 0.3183, 0.6013, 0.6183],
  37162. [0.6129, 0.4063, 0.8738, 0.5250, 0.4313, 0.4733, 0.5230, 0.5874],
  37163. [0.6229, 0.4198, 0.7662, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783],
  37164. [0.6085, 0.4008, 0.8588, 0.5200, 0.4959, 0.4977, 0.5175, 0.5283]],
  37165. device='cuda:0', dtype=torch.float64)
  37166. predictions are: tensor([[ 0.4583, -0.4992, 0.8629, -1.1960, -0.6042, -1.1037, 0.4087, 0.1988],
  37167. [ 0.5696, -0.4237, 1.8692, 0.1697, -0.4560, 0.0712, 0.5130, 0.0453],
  37168. [ 0.6720, -0.3059, 1.6680, -0.4913, -0.1965, -1.1177, 0.5406, 0.3097],
  37169. [ 0.2959, -0.5816, 1.4801, -0.9764, -0.5421, -1.1075, 0.2888, 0.1507],
  37170. [ 0.6170, -0.3856, 1.3985, -0.9501, -0.5622, -0.8307, 0.7079, 0.3880],
  37171. [ 0.6378, -0.3512, 1.8180, 0.2329, -0.4365, -0.0178, 0.4217, 0.2185],
  37172. [ 0.4949, -0.4244, 1.3233, -1.0124, -0.2559, -1.2040, 0.5447, 0.3054],
  37173. [ 0.5707, -0.4398, 1.8753, 0.1660, -0.1342, 0.0073, 0.5037, 0.0035]],
  37174. device='cuda:0', grad_fn=<AddmmBackward>)
  37175. landmarks are: tensor([[[ 0.5483, -0.3991, 0.8030, -1.2159, -0.5018, -1.1928, 0.2624,
  37176. 0.3852],
  37177. [ 0.5482, -0.3841, 1.7326, 0.1005, -0.3517, 0.0620, 0.0912,
  37178. 0.2522],
  37179. [ 0.6237, -0.2983, 1.3919, -0.4691, -0.0457, -1.2313, 0.2453,
  37180. 0.5882],
  37181. [ 0.5695, -0.3927, 1.3226, -0.9002, -0.4672, -1.1928, 0.1737,
  37182. 0.3700],
  37183. [ 0.5759, -0.3948, 1.2072, -0.9849, -0.4672, -0.8156, 0.4912,
  37184. 0.5701],
  37185. [ 0.5453, -0.4091, 1.7499, 0.1390, -0.2940, -0.0996, 0.1300,
  37186. 0.4272],
  37187. [ 0.5912, -0.3467, 1.2533, -1.0388, -0.1150, -1.3005, 0.3353,
  37188. 0.3854],
  37189. [ 0.5245, -0.4347, 1.6806, 0.1159, 0.0046, 0.0129, 0.1044,
  37190. 0.1544]]], device='cuda:0')
  37191. loss_train_step before backward: tensor(0.0235, device='cuda:0', grad_fn=<MseLossBackward>)
  37192. loss_train_step after backward: tensor(0.0235, device='cuda:0', grad_fn=<MseLossBackward>)
  37193. loss_train: 1.060399990528822
  37194. step: 67
  37195. running loss: 0.015826865530280926
  37196.  
  37197. Train Steps: 67/90 Loss: 0.0158 torch.Size([8, 600, 800])
  37198. torch.Size([8, 8])
  37199. tensor([[0.6229, 0.4066, 0.8513, 0.5717, 0.4500, 0.5050, 0.6025, 0.5350],
  37200. [0.6361, 0.4071, 0.9100, 0.4783, 0.3738, 0.3967, 0.6670, 0.5332],
  37201. [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6088, 0.5583],
  37202. [0.6214, 0.4116, 0.7750, 0.2317, 0.4487, 0.1883, 0.6200, 0.5400],
  37203. [0.6150, 0.3949, 0.8800, 0.4033, 0.3825, 0.4900, 0.5875, 0.5483],
  37204. [0.6274, 0.4099, 0.8625, 0.3233, 0.4400, 0.1983, 0.5876, 0.4869],
  37205. [0.6218, 0.4185, 0.7338, 0.2650, 0.4625, 0.1950, 0.5687, 0.5800],
  37206. [0.6203, 0.4072, 0.8892, 0.3523, 0.3783, 0.3017, 0.5898, 0.5478]],
  37207. device='cuda:0', dtype=torch.float64)
  37208. predictions are: tensor([[ 0.6211, -0.3934, 1.7261, 0.4591, -0.2666, 0.0805, 0.5491, 0.1778],
  37209. [ 0.6831, -0.3731, 1.9187, -0.0112, -0.6001, -0.3599, 0.7898, 0.1492],
  37210. [ 0.3914, -0.5190, 0.9761, -1.2904, -0.5236, -1.0034, 0.4221, 0.2727],
  37211. [ 0.5367, -0.4494, 1.2859, -1.1676, -0.2218, -1.2896, 0.5564, 0.1938],
  37212. [ 0.7263, -0.3094, 1.7982, -0.3888, -0.5521, 0.0118, 0.5007, 0.1631],
  37213. [ 0.6938, -0.3546, 1.8083, -0.7449, -0.2507, -1.3090, 0.5148, 0.0247],
  37214. [ 0.5000, -0.4189, 1.0990, -1.0047, -0.2257, -1.2594, 0.3236, 0.4026],
  37215. [ 0.6410, -0.3702, 1.9326, -0.6053, -0.5425, -0.7602, 0.4749, 0.2303]],
  37216. device='cuda:0', grad_fn=<AddmmBackward>)
  37217. landmarks are: tensor([[[ 0.5911, -0.4080, 1.6460, 0.3546, -0.2074, 0.0467, 0.4970,
  37218. 0.1852],
  37219. [ 0.6520, -0.4056, 1.9173, -0.0765, -0.5596, -0.4537, 0.7949,
  37220. 0.1768],
  37221. [ 0.5992, -0.3968, 0.9307, -1.3497, -0.4730, -1.0465, 0.5259,
  37222. 0.2930],
  37223. [ 0.5842, -0.3848, 1.2938, -1.2159, -0.2132, -1.4160, 0.5778,
  37224. 0.2083],
  37225. [ 0.5546, -0.4620, 1.7788, -0.4229, -0.5192, -0.0226, 0.4277,
  37226. 0.2468],
  37227. [ 0.6119, -0.3927, 1.6979, -0.7925, -0.2536, -1.3698, 0.4282,
  37228. -0.0368],
  37229. [ 0.5862, -0.3530, 1.1032, -1.0619, -0.1497, -1.3852, 0.3411,
  37230. 0.3931],
  37231. [ 0.5792, -0.4052, 1.8214, -0.6587, -0.5384, -0.8924, 0.4381,
  37232. 0.2442]]], device='cuda:0')
  37233. loss_train_step before backward: tensor(0.0052, device='cuda:0', grad_fn=<MseLossBackward>)
  37234. loss_train_step after backward: tensor(0.0052, device='cuda:0', grad_fn=<MseLossBackward>)
  37235. loss_train: 1.0655750310979784
  37236. step: 68
  37237. running loss: 0.015670221045558506
  37238. Train Steps: 68/90 Loss: 0.0157 torch.Size([8, 600, 800])
  37239. torch.Size([8, 8])
  37240. tensor([[0.6201, 0.4151, 0.8588, 0.5467, 0.3700, 0.3950, 0.5637, 0.5933],
  37241. [0.6229, 0.4198, 0.7662, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783],
  37242. [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317],
  37243. [0.6179, 0.4008, 0.8600, 0.4015, 0.3932, 0.2515, 0.5711, 0.5438],
  37244. [0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204],
  37245. [0.6168, 0.4055, 0.7575, 0.2950, 0.4050, 0.2717, 0.5938, 0.6167],
  37246. [0.6353, 0.4128, 0.8488, 0.2600, 0.5525, 0.1616, 0.6694, 0.5540],
  37247. [0.6198, 0.4164, 0.8700, 0.5067, 0.4625, 0.5650, 0.5464, 0.5197]],
  37248. device='cuda:0', dtype=torch.float64)
  37249. predictions are: tensor([[ 6.4932e-01, -3.1109e-01, 1.7652e+00, 1.8150e-01, -6.3398e-01,
  37250. -3.7820e-01, 2.5626e-01, 3.6480e-01],
  37251. [ 5.5427e-01, -3.8678e-01, 1.2450e+00, -1.0782e+00, -2.0523e-01,
  37252. -1.2230e+00, 4.0710e-01, 3.8396e-01],
  37253. [ 5.8606e-01, -4.1393e-01, 1.8375e+00, 1.7360e-01, -6.3783e-01,
  37254. -4.7665e-01, 3.0656e-01, 1.1668e-01],
  37255. [ 7.0116e-01, -3.3979e-01, 1.6863e+00, -5.5139e-01, -4.7769e-01,
  37256. -1.0744e+00, 2.7354e-01, 1.6098e-01],
  37257. [ 6.5969e-01, -3.8419e-01, 1.7464e+00, 1.3380e-01, -3.9393e-01,
  37258. 1.2854e-01, 9.0110e-01, 1.3022e-01],
  37259. [ 5.2570e-01, -4.2665e-01, 1.2532e+00, -1.0416e+00, -5.1630e-01,
  37260. -9.5149e-01, 4.5284e-01, 4.6819e-01],
  37261. [ 5.8049e-01, -4.0814e-01, 1.6135e+00, -1.3326e+00, 2.0178e-01,
  37262. -1.3246e+00, 9.6435e-01, 2.3764e-01],
  37263. [ 6.0887e-01, -3.5483e-01, 1.8828e+00, -1.1708e-03, -2.4791e-01,
  37264. 1.8476e-01, 4.1590e-01, 7.2394e-02]], device='cuda:0',
  37265. grad_fn=<AddmmBackward>)
  37266. landmarks are: tensor([[[ 0.5785, -0.3687, 1.6806, 0.2391, -0.5769, -0.4614, 0.3180,
  37267. 0.4547],
  37268. [ 0.5912, -0.3467, 1.2533, -1.0388, -0.1150, -1.3005, 0.3353,
  37269. 0.3854],
  37270. [ 0.5901, -0.4157, 1.7557, 0.1929, -0.5423, -0.5923, 0.3584,
  37271. 0.1698],
  37272. [ 0.5680, -0.4345, 1.6864, -0.4315, -0.4698, -1.1241, 0.3518,
  37273. 0.2261],
  37274. [ 0.6042, -0.4273, 1.7198, 0.2184, -0.3478, 0.1149, 0.8062,
  37275. 0.1176],
  37276. [ 0.5631, -0.4129, 1.2129, -0.9233, -0.4152, -1.0311, 0.4566,
  37277. 0.5624],
  37278. [ 0.6487, -0.3792, 1.6344, -1.0850, 0.2659, -1.5397, 0.8059,
  37279. 0.2730],
  37280. [ 0.5770, -0.3624, 1.7326, 0.0543, -0.1497, 0.3238, 0.2378,
  37281. 0.1146]]], device='cuda:0')
  37282. loss_train_step before backward: tensor(0.0071, device='cuda:0', grad_fn=<MseLossBackward>)
  37283. loss_train_step after backward: tensor(0.0071, device='cuda:0', grad_fn=<MseLossBackward>)
  37284. loss_train: 1.0726689579896629
  37285. step: 69
  37286. running loss: 0.01554592692738642
  37287. Train Steps: 69/90 Loss: 0.0155 torch.Size([8, 600, 800])
  37288. torch.Size([8, 8])
  37289. tensor([[0.6048, 0.3928, 0.8538, 0.5433, 0.3875, 0.5117, 0.5266, 0.4719],
  37290. [0.6055, 0.4015, 0.7425, 0.2033, 0.4113, 0.1883, 0.5217, 0.4823],
  37291. [0.6148, 0.3996, 0.8488, 0.3867, 0.3488, 0.4067, 0.5863, 0.5000],
  37292. [0.6136, 0.3955, 0.8400, 0.3267, 0.3500, 0.4200, 0.5863, 0.5467],
  37293. [0.6223, 0.4130, 0.8100, 0.2983, 0.3525, 0.3900, 0.5694, 0.5103],
  37294. [0.6361, 0.4102, 0.8650, 0.2983, 0.4888, 0.2000, 0.6702, 0.5459],
  37295. [0.6271, 0.4040, 0.9138, 0.3750, 0.4625, 0.2617, 0.7232, 0.5413],
  37296. [0.6214, 0.4112, 0.7838, 0.2117, 0.3650, 0.3133, 0.5675, 0.5083]],
  37297. device='cuda:0', dtype=torch.float64)
  37298. predictions are: tensor([[ 0.5832, -0.4019, 1.7111, 0.4521, -0.3889, 0.0626, 0.1651, 0.1931],
  37299. [ 0.5105, -0.4393, 0.9998, -1.1845, -0.3595, -1.3779, 0.1922, 0.1342],
  37300. [ 0.6528, -0.3713, 1.6885, -0.3394, -0.6116, -0.3599, 0.3463, 0.2027],
  37301. [ 0.6792, -0.3349, 1.6055, -0.6538, -0.5749, -0.4010, 0.4729, 0.3349],
  37302. [ 0.8564, -0.2081, 1.5162, -0.6773, -0.5229, -0.5072, 0.3051, 0.3393],
  37303. [ 0.6010, -0.4154, 1.7584, -0.8380, 0.0861, -1.3183, 0.9652, 0.2800],
  37304. [ 0.6342, -0.4167, 2.0447, -0.4159, -0.0362, -1.1658, 1.0665, 0.2960],
  37305. [ 0.6628, -0.3548, 1.3875, -1.2671, -0.5098, -0.9115, 0.4795, 0.2078]],
  37306. device='cuda:0', grad_fn=<AddmmBackward>)
  37307. landmarks are: tensor([[[ 0.5079, -0.4714, 1.6575, 0.2237, -0.4961, 0.0774, 0.1465,
  37308. -0.1061],
  37309. [ 0.5109, -0.4316, 1.1436, -1.3467, -0.3864, -1.4160, 0.1239,
  37310. -0.0580],
  37311. [ 0.5538, -0.4400, 1.6344, -0.4999, -0.6751, -0.4075, 0.4219,
  37312. 0.0236],
  37313. [ 0.5484, -0.4591, 1.5940, -0.7771, -0.6693, -0.3460, 0.4219,
  37314. 0.2391],
  37315. [ 0.5886, -0.3784, 1.4554, -0.9079, -0.6577, -0.4845, 0.3440,
  37316. 0.0712],
  37317. [ 0.6520, -0.3912, 1.7095, -0.9079, -0.0284, -1.3621, 0.8096,
  37318. 0.2356],
  37319. [ 0.6108, -0.4201, 1.9346, -0.5538, -0.1497, -1.0773, 1.0545,
  37320. 0.2142],
  37321. [ 0.5845, -0.3864, 1.3342, -1.3082, -0.6000, -0.8386, 0.3353,
  37322. 0.0620]]], device='cuda:0')
  37323. loss_train_step before backward: tensor(0.0132, device='cuda:0', grad_fn=<MseLossBackward>)
  37324. loss_train_step after backward: tensor(0.0132, device='cuda:0', grad_fn=<MseLossBackward>)
  37325. loss_train: 1.0858652363531291
  37326. step: 70
  37327. running loss: 0.015512360519330417
  37328. Train Steps: 70/90 Loss: 0.0155 torch.Size([8, 600, 800])
  37329. torch.Size([8, 8])
  37330. tensor([[0.6277, 0.4036, 0.8688, 0.3617, 0.3925, 0.2600, 0.6132, 0.4742],
  37331. [0.6076, 0.3958, 0.8700, 0.4667, 0.4546, 0.5046, 0.5231, 0.4960],
  37332. [0.6275, 0.4071, 0.8075, 0.2100, 0.4700, 0.1533, 0.6148, 0.4834],
  37333. [0.6189, 0.4029, 0.8375, 0.5767, 0.4745, 0.4829, 0.5551, 0.5598],
  37334. [0.6140, 0.4070, 0.8700, 0.5000, 0.4612, 0.4900, 0.5260, 0.5852],
  37335. [0.6147, 0.4112, 0.7988, 0.3200, 0.3775, 0.2767, 0.5150, 0.5550],
  37336. [ nan, nan, 0.7515, 0.2708, 0.3987, 0.2267, 0.5162, 0.5567],
  37337. [0.6276, 0.4002, 0.8800, 0.5533, 0.3575, 0.4400, 0.6132, 0.4672]],
  37338. device='cuda:0', dtype=torch.float64)
  37339. predictions are: tensor([[ 0.7524, -0.3138, 1.7652, -0.7036, -0.4591, -1.0202, 0.6242, 0.0644],
  37340. [ 0.7036, -0.3337, 1.8266, -0.1566, -0.2323, 0.0085, 0.2252, 0.1692],
  37341. [ 0.7085, -0.3449, 1.4596, -1.4295, -0.0171, -1.5700, 0.9127, 0.0854],
  37342. [ 0.6374, -0.3348, 1.6523, 0.3809, -0.1125, -0.1319, 0.3967, 0.3868],
  37343. [ 0.5919, -0.3661, 1.7851, 0.0314, -0.1617, -0.0299, 0.3320, 0.4521],
  37344. [ 0.8570, -0.1832, 1.4189, -0.8228, -0.5323, -0.9292, 0.0790, 0.3344],
  37345. [-1.1760, -1.5135, 1.1183, -1.1627, -0.3278, -1.3243, 0.1392, 0.3514],
  37346. [ 0.7340, -0.3614, 1.7749, 0.2905, -0.6121, -0.2406, 0.7003, 0.1101]],
  37347. device='cuda:0', grad_fn=<AddmmBackward>)
  37348. landmarks are: tensor([[[ 0.6134, -0.4218, 1.7268, -0.6154, -0.4730, -1.0850, 0.5463,
  37349. -0.0957],
  37350. [ 0.5205, -0.4577, 1.7326, -0.1304, -0.1862, 0.0447, 0.1301,
  37351. 0.0051],
  37352. [ 0.6126, -0.4057, 1.4439, -1.3159, -0.1150, -1.5777, 0.5537,
  37353. -0.0530],
  37354. [ 0.5726, -0.4249, 1.5824, 0.3777, -0.0942, -0.0556, 0.2782,
  37355. 0.2997],
  37356. [ 0.5500, -0.4060, 1.7326, 0.0236, -0.1554, -0.0226, 0.1438,
  37357. 0.4171],
  37358. [ 0.5532, -0.3864, 1.4035, -0.8079, -0.5423, -1.0080, 0.0928,
  37359. 0.2776],
  37360. [-2.2859, -2.2859, 1.1854, -1.0352, -0.4441, -1.2390, 0.0986,
  37361. 0.2853],
  37362. [ 0.6128, -0.4375, 1.7788, 0.2699, -0.6346, -0.2536, 0.5463,
  37363. -0.1278]]], device='cuda:0')
  37364. loss_train_step before backward: tensor(0.0401, device='cuda:0', grad_fn=<MseLossBackward>)
  37365. loss_train_step after backward: tensor(0.0401, device='cuda:0', grad_fn=<MseLossBackward>)
  37366. loss_train: 1.1259842771105468
  37367. step: 71
  37368. running loss: 0.015858933480430236
  37369.  
  37370. Train Steps: 71/90 Loss: 0.0159 torch.Size([8, 600, 800])
  37371. torch.Size([8, 8])
  37372. tensor([[0.6200, 0.4112, 0.8862, 0.4100, 0.3638, 0.4917, 0.6088, 0.6050],
  37373. [0.6151, 0.4085, 0.8750, 0.4367, 0.3887, 0.4367, 0.5066, 0.5846],
  37374. [0.6108, 0.4011, 0.8037, 0.3400, 0.3700, 0.2933, 0.5658, 0.5617],
  37375. [0.6236, 0.4084, 0.7738, 0.2133, 0.3663, 0.3233, 0.5813, 0.5567],
  37376. [0.6138, 0.4020, 0.8800, 0.4717, 0.4375, 0.4717, 0.5502, 0.5611],
  37377. [0.6339, 0.4123, 0.8638, 0.5367, 0.4075, 0.5467, 0.7517, 0.5436],
  37378. [0.6161, 0.4055, 0.8675, 0.3867, 0.3713, 0.4033, 0.5195, 0.5162],
  37379. [ nan, nan, 0.8488, 0.2300, 0.5563, 0.2100, 0.7390, 0.5679]],
  37380. device='cuda:0', dtype=torch.float64)
  37381. predictions are: tensor([[ 0.7596, -0.2490, 1.7706, -0.2143, -0.4673, -0.0846, 0.3435, 0.3957],
  37382. [ 0.6366, -0.3010, 1.7074, -0.1447, -0.4508, -0.3138, -0.0450, 0.2564],
  37383. [ 0.6255, -0.3668, 1.4270, -0.6066, -0.5383, -0.9385, 0.2893, 0.1861],
  37384. [ 0.6542, -0.3571, 1.2980, -1.1232, -0.4624, -0.9424, 0.4861, 0.2497],
  37385. [ 0.6544, -0.3270, 1.7445, -0.0034, -0.2148, -0.1705, 0.1259, 0.2557],
  37386. [ 0.6998, -0.3413, 1.7491, 0.2611, -0.3304, 0.1221, 1.0424, 0.2210],
  37387. [ 0.7470, -0.2772, 1.7329, -0.3159, -0.5494, -0.5700, -0.0321, 0.0129],
  37388. [-1.4784, -1.7442, 1.6703, -1.2390, 0.4985, -1.6062, 1.2554, 0.3567]],
  37389. device='cuda:0', grad_fn=<AddmmBackward>)
  37390. landmarks are: tensor([[[ 0.5777, -0.3868, 1.8076, -0.3921, -0.6058, -0.0149, 0.5259,
  37391. 0.5085],
  37392. [ 0.5553, -0.3992, 1.7557, -0.2690, -0.4903, -0.2690, 0.0542,
  37393. 0.4145],
  37394. [ 0.5355, -0.4335, 1.4266, -0.7155, -0.5769, -0.9310, 0.3275,
  37395. 0.3087],
  37396. [ 0.5946, -0.3995, 1.2880, -1.3005, -0.5942, -0.7925, 0.3988,
  37397. 0.2853],
  37398. [ 0.5491, -0.4290, 1.7788, -0.1073, -0.2651, -0.1073, 0.2555,
  37399. 0.3057],
  37400. [ 0.6421, -0.3816, 1.7037, 0.1929, -0.4037, 0.2391, 1.1861,
  37401. 0.2249],
  37402. [ 0.5598, -0.4129, 1.7210, -0.4999, -0.5711, -0.4229, 0.1136,
  37403. 0.0983],
  37404. [-2.2859, -2.2859, 1.6344, -1.2236, 0.2834, -1.3159, 1.1276,
  37405. 0.3371]]], device='cuda:0')
  37406. loss_train_step before backward: tensor(0.0265, device='cuda:0', grad_fn=<MseLossBackward>)
  37407. loss_train_step after backward: tensor(0.0265, device='cuda:0', grad_fn=<MseLossBackward>)
  37408. loss_train: 1.1525277267210186
  37409. step: 72
  37410. running loss: 0.016007329537791923
  37411. Train Steps: 72/90 Loss: 0.0160 torch.Size([8, 600, 800])
  37412. torch.Size([8, 8])
  37413. tensor([[0.6132, 0.4118, 0.8200, 0.3633, 0.3563, 0.5400, 0.5787, 0.5136],
  37414. [ nan, nan, 0.8750, 0.2467, 0.5138, 0.2617, 0.7382, 0.5552],
  37415. [0.6110, 0.4047, 0.8700, 0.4483, 0.3713, 0.3967, 0.5088, 0.5517],
  37416. [0.6185, 0.4042, 0.7700, 0.2250, 0.4062, 0.2117, 0.5763, 0.5150],
  37417. [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051],
  37418. [0.6270, 0.4267, 0.7150, 0.3317, 0.4900, 0.1817, 0.5560, 0.6183],
  37419. [0.6264, 0.4248, 0.8938, 0.4183, 0.3875, 0.4100, 0.5400, 0.5350],
  37420. [0.6276, 0.4095, 0.8237, 0.2250, 0.4662, 0.1783, 0.6171, 0.4869]],
  37421. device='cuda:0', dtype=torch.float64)
  37422. predictions are: tensor([[ 5.9427e-01, -3.5149e-01, 1.5071e+00, -5.8090e-01, -4.9336e-01,
  37423. 8.7826e-02, 4.6292e-01, 2.3218e-01],
  37424. [-1.9318e+00, -2.0346e+00, 1.7207e+00, -1.2192e+00, 2.7853e-01,
  37425. -1.4549e+00, 1.0914e+00, 3.3141e-01],
  37426. [ 5.7028e-01, -3.5766e-01, 1.7352e+00, -9.0277e-02, -5.7938e-01,
  37427. -3.2921e-01, -1.3246e-01, 2.5256e-01],
  37428. [ 5.9285e-01, -4.0147e-01, 1.2382e+00, -1.1689e+00, -3.9334e-01,
  37429. -1.2938e+00, 3.6060e-01, 1.1863e-01],
  37430. [ 6.5827e-01, -3.9911e-01, 1.8256e+00, -1.6101e-03, -5.0889e-01,
  37431. -4.0250e-01, 5.7856e-01, 1.1106e-01],
  37432. [ 6.9346e-01, -2.9639e-01, 1.0066e+00, -7.7031e-01, -1.1378e-01,
  37433. -1.3372e+00, 2.4999e-01, 5.3997e-01],
  37434. [ 6.0116e-01, -3.3656e-01, 1.8278e+00, -2.2940e-01, -5.3385e-01,
  37435. -3.2537e-01, 7.8322e-02, 2.2241e-01],
  37436. [ 6.2932e-01, -3.6683e-01, 1.5071e+00, -1.2220e+00, -3.4216e-03,
  37437. -1.4243e+00, 6.1705e-01, 7.4610e-02]], device='cuda:0',
  37438. grad_fn=<AddmmBackward>)
  37439. landmarks are: tensor([[[ 0.5466, -0.3840, 1.5016, -0.6077, -0.6404, 0.2083, 0.3871,
  37440. 0.0862],
  37441. [-2.2859, -2.2859, 1.7557, -1.1466, 0.0871, -1.0773, 1.1239,
  37442. 0.2783],
  37443. [ 0.5363, -0.4168, 1.7326, -0.2151, -0.5711, -0.4537, 0.0640,
  37444. 0.2622],
  37445. [ 0.5707, -0.4189, 1.2707, -1.2467, -0.4095, -1.3082, 0.3758,
  37446. 0.0928],
  37447. [ 0.6204, -0.4336, 1.8654, -0.0688, -0.6058, -0.5230, 0.6503,
  37448. 0.0472],
  37449. [ 0.6101, -0.3152, 1.0166, -0.7540, -0.0226, -1.4468, 0.2823,
  37450. 0.5702],
  37451. [ 0.6072, -0.3239, 1.8423, -0.3537, -0.4961, -0.3921, 0.2083,
  37452. 0.1852],
  37453. [ 0.6127, -0.3944, 1.5189, -1.2467, -0.1323, -1.4622, 0.5646,
  37454. -0.0369]]], device='cuda:0')
  37455. loss_train_step before backward: tensor(0.0106, device='cuda:0', grad_fn=<MseLossBackward>)
  37456. loss_train_step after backward: tensor(0.0106, device='cuda:0', grad_fn=<MseLossBackward>)
  37457. loss_train: 1.1630904129706323
  37458. step: 73
  37459. running loss: 0.015932745383159346
  37460. Train Steps: 73/90 Loss: 0.0159 torch.Size([8, 600, 800])
  37461. torch.Size([8, 8])
  37462. tensor([[0.6182, 0.4058, 0.8738, 0.4350, 0.3563, 0.3400, 0.5290, 0.5822],
  37463. [0.6164, 0.3972, 0.7818, 0.3381, 0.3599, 0.3387, 0.5880, 0.5153],
  37464. [0.6314, 0.4107, 0.8750, 0.5100, 0.3788, 0.4900, 0.7121, 0.5864],
  37465. [0.6109, 0.4015, 0.7668, 0.3639, 0.3513, 0.3667, 0.5200, 0.5641],
  37466. [0.6226, 0.4001, 0.8438, 0.5733, 0.3862, 0.4250, 0.5750, 0.5283],
  37467. [0.6163, 0.4001, 0.8788, 0.5033, 0.4012, 0.4633, 0.5338, 0.5767],
  37468. [0.6201, 0.4065, 0.7300, 0.1933, 0.4075, 0.2533, 0.6162, 0.5483],
  37469. [0.6153, 0.4119, 0.8463, 0.3833, 0.3600, 0.3200, 0.5106, 0.5563]],
  37470. device='cuda:0', dtype=torch.float64)
  37471. predictions are: tensor([[ 0.6199, -0.3633, 1.7836, -0.3836, -0.5077, -0.7211, 0.0684, 0.2759],
  37472. [ 0.6889, -0.3536, 1.4425, -0.8366, -0.5263, -0.7809, 0.4121, 0.0883],
  37473. [ 0.6030, -0.4272, 1.9410, 0.0582, -0.2889, -0.1175, 1.1050, 0.3419],
  37474. [ 0.5360, -0.4069, 1.3305, -0.7654, -0.5238, -0.6782, 0.1789, 0.2832],
  37475. [ 0.7144, -0.3519, 1.7441, 0.3128, -0.3675, -0.3465, 0.3809, 0.1663],
  37476. [ 0.5369, -0.4524, 1.8689, -0.0673, -0.2455, -0.2268, 0.1903, 0.2530],
  37477. [ 0.6427, -0.3706, 1.1777, -1.4482, -0.2613, -1.2697, 0.5720, 0.2102],
  37478. [ 0.6546, -0.3343, 1.7212, -0.5872, -0.4739, -0.9253, 0.0056, 0.2401]],
  37479. device='cuda:0', grad_fn=<AddmmBackward>)
  37480. landmarks are: tensor([[[ 0.5697, -0.4113, 1.7499, -0.2767, -0.6404, -0.7155, 0.1576,
  37481. 0.4032],
  37482. [ 0.5614, -0.4514, 1.3252, -0.7240, -0.6237, -0.7215, 0.4299,
  37483. 0.0942],
  37484. [ 0.6306, -0.3888, 1.7557, 0.0697, -0.5365, -0.0226, 1.0033,
  37485. 0.4226],
  37486. [ 0.5357, -0.4315, 1.2557, -0.6051, -0.6635, -0.5923, 0.1159,
  37487. 0.3198],
  37488. [ 0.5900, -0.4377, 1.6113, 0.3623, -0.5018, -0.3229, 0.3700,
  37489. 0.1544],
  37490. [ 0.5610, -0.4381, 1.7730, 0.0390, -0.4326, -0.1458, 0.1794,
  37491. 0.3777],
  37492. [ 0.5784, -0.4085, 1.0859, -1.3929, -0.4037, -1.1158, 0.5605,
  37493. 0.2468],
  37494. [ 0.5561, -0.3834, 1.6229, -0.5153, -0.6231, -0.8079, 0.0727,
  37495. 0.2837]]], device='cuda:0')
  37496. loss_train_step before backward: tensor(0.0091, device='cuda:0', grad_fn=<MseLossBackward>)
  37497. loss_train_step after backward: tensor(0.0091, device='cuda:0', grad_fn=<MseLossBackward>)
  37498. loss_train: 1.1722305989824235
  37499. step: 74
  37500. running loss: 0.015840954040303022
  37501. Train Steps: 74/90 Loss: 0.0158 torch.Size([8, 600, 800])
  37502. torch.Size([8, 8])
  37503. tensor([[0.6139, 0.4019, 0.7137, 0.2150, 0.4375, 0.1533, 0.5293, 0.5006],
  37504. [0.6200, 0.4055, 0.8011, 0.2988, 0.3842, 0.2798, 0.5949, 0.5183],
  37505. [ nan, nan, 0.8525, 0.2217, 0.5413, 0.2367, 0.7367, 0.5482],
  37506. [0.6138, 0.4054, 0.8750, 0.4750, 0.4363, 0.5017, 0.5086, 0.5822],
  37507. [0.6275, 0.4157, 0.8337, 0.5800, 0.3763, 0.4200, 0.5547, 0.6125],
  37508. [0.6128, 0.4022, 0.8738, 0.5067, 0.4983, 0.5231, 0.5364, 0.5064],
  37509. [0.6097, 0.4024, 0.8488, 0.3717, 0.3875, 0.5517, 0.5836, 0.5591],
  37510. [0.6239, 0.4061, 0.8850, 0.4600, 0.4225, 0.5200, 0.6138, 0.5450]],
  37511. device='cuda:0', dtype=torch.float64)
  37512. predictions are: tensor([[ 0.7868, -0.2607, 1.0193, -1.2032, -0.3471, -1.6748, 0.2052, 0.0175],
  37513. [ 0.5743, -0.4024, 1.4428, -0.8914, -0.5607, -0.9998, 0.4252, 0.0937],
  37514. [-2.1559, -2.1935, 1.5947, -1.3129, 0.2300, -1.5728, 1.0261, 0.3095],
  37515. [ 0.5379, -0.3878, 1.7228, -0.1518, -0.3012, -0.0963, 0.0473, 0.2776],
  37516. [ 0.6823, -0.3246, 1.6219, 0.3853, -0.6079, -0.4308, 0.2746, 0.4834],
  37517. [ 0.4335, -0.4857, 1.7522, -0.0481, -0.0376, -0.1753, 0.0980, 0.1554],
  37518. [ 0.4807, -0.4354, 1.6261, -0.5759, -0.4750, 0.1213, 0.4935, 0.1963],
  37519. [ 0.5461, -0.4275, 1.8113, -0.2441, -0.3507, 0.0115, 0.4388, 0.1897]],
  37520. device='cuda:0', grad_fn=<AddmmBackward>)
  37521. landmarks are: tensor([[[ 0.5496, -0.4295, 1.0109, -1.2928, -0.2651, -1.5777, 0.1589,
  37522. 0.0261],
  37523. [ 0.5781, -0.4129, 1.4142, -0.9057, -0.5115, -0.9937, 0.4621,
  37524. 0.1080],
  37525. [-2.2859, -2.2859, 1.6517, -1.2620, 0.2141, -1.1928, 1.1166,
  37526. 0.2463],
  37527. [ 0.5491, -0.4132, 1.7557, -0.0919, -0.2709, 0.0313, 0.0635,
  37528. 0.4032],
  37529. [ 0.6124, -0.3658, 1.5651, 0.3931, -0.5480, -0.3460, 0.2761,
  37530. 0.5431],
  37531. [ 0.5446, -0.4280, 1.7499, 0.0543, 0.0156, 0.1301, 0.1918,
  37532. 0.0532],
  37533. [ 0.5301, -0.4273, 1.6344, -0.5692, -0.4961, 0.2622, 0.4098,
  37534. 0.2966],
  37535. [ 0.5960, -0.4102, 1.8018, -0.1612, -0.3344, 0.1159, 0.5490,
  37536. 0.2314]]], device='cuda:0')
  37537. loss_train_step before backward: tensor(0.0091, device='cuda:0', grad_fn=<MseLossBackward>)
  37538. loss_train_step after backward: tensor(0.0091, device='cuda:0', grad_fn=<MseLossBackward>)
  37539. loss_train: 1.1813519136048853
  37540. step: 75
  37541. running loss: 0.01575135884806514
  37542.  
  37543. Train Steps: 75/90 Loss: 0.0158 torch.Size([8, 600, 800])
  37544. torch.Size([8, 8])
  37545. tensor([[0.6117, 0.4018, 0.6562, 0.1967, 0.3738, 0.2550, 0.5280, 0.5103],
  37546. [0.6264, 0.4049, 0.8988, 0.4633, 0.3813, 0.4983, 0.6326, 0.4843],
  37547. [0.6124, 0.4030, 0.8650, 0.4867, 0.4999, 0.5106, 0.5137, 0.5773],
  37548. [0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038],
  37549. [0.6201, 0.4082, 0.8827, 0.3715, 0.3825, 0.2712, 0.5845, 0.5412],
  37550. [0.6222, 0.4072, 0.7164, 0.2166, 0.3738, 0.3167, 0.6100, 0.5533],
  37551. [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6088, 0.5583],
  37552. [0.6240, 0.4217, 0.8150, 0.3133, 0.4425, 0.2650, 0.5650, 0.5817]],
  37553. device='cuda:0', dtype=torch.float64)
  37554. predictions are: tensor([[ 4.3204e-01, -4.9423e-01, 8.7678e-01, -1.1577e+00, -4.9900e-01,
  37555. -1.2377e+00, 1.9630e-01, 1.8164e-01],
  37556. [ 6.9072e-01, -3.8656e-01, 2.0222e+00, -1.1292e-01, -5.2003e-01,
  37557. -2.3396e-02, 6.1781e-01, 6.3428e-02],
  37558. [ 3.8911e-01, -5.3941e-01, 1.7964e+00, -3.4572e-02, -9.5070e-03,
  37559. -9.4753e-02, 7.4348e-02, 3.2933e-01],
  37560. [ 6.1090e-01, -4.0105e-01, 1.8377e+00, -9.8769e-02, -1.8835e-01,
  37561. -1.7130e-03, 5.7579e-02, 8.1385e-02],
  37562. [ 6.4133e-01, -3.7246e-01, 1.8869e+00, -5.2754e-01, -5.2379e-01,
  37563. -1.0953e+00, 4.3551e-01, 2.3086e-01],
  37564. [ 6.0569e-01, -3.8838e-01, 1.1288e+00, -1.1754e+00, -5.1845e-01,
  37565. -9.1244e-01, 5.3057e-01, 2.8769e-01],
  37566. [ 5.0548e-01, -4.6197e-01, 1.0753e+00, -1.1920e+00, -4.6685e-01,
  37567. -1.2029e+00, 4.8190e-01, 3.0180e-01],
  37568. [ 5.9589e-01, -3.6515e-01, 1.5408e+00, -7.3261e-01, -3.0263e-01,
  37569. -1.2108e+00, 2.8190e-01, 4.1174e-01]], device='cuda:0',
  37570. grad_fn=<AddmmBackward>)
  37571. landmarks are: tensor([[[ 0.5393, -0.4299, 0.7453, -1.3775, -0.5596, -1.1081, 0.1527,
  37572. 0.0712],
  37573. [ 0.6075, -0.4159, 1.8654, -0.1458, -0.5249, 0.0159, 0.6359,
  37574. -0.0490],
  37575. [ 0.5428, -0.4244, 1.7095, -0.0380, 0.0232, 0.0725, 0.0866,
  37576. 0.3806],
  37577. [ 0.5249, -0.4473, 1.7326, -0.0919, -0.2016, 0.1544, 0.1733,
  37578. 0.0412],
  37579. [ 0.5784, -0.4006, 1.7911, -0.5701, -0.5192, -1.0331, 0.4137,
  37580. 0.2139],
  37581. [ 0.5878, -0.4052, 1.0229, -1.2855, -0.5596, -0.8232, 0.5316,
  37582. 0.2699],
  37583. [ 0.5992, -0.3968, 0.9307, -1.3497, -0.4730, -1.0465, 0.5259,
  37584. 0.2930],
  37585. [ 0.5964, -0.3380, 1.4785, -0.8386, -0.2420, -1.0619, 0.3238,
  37586. 0.4008]]], device='cuda:0')
  37587. loss_train_step before backward: tensor(0.0076, device='cuda:0', grad_fn=<MseLossBackward>)
  37588. loss_train_step after backward: tensor(0.0076, device='cuda:0', grad_fn=<MseLossBackward>)
  37589. loss_train: 1.1889062826521695
  37590. step: 76
  37591. running loss: 0.015643503719107492
  37592. Train Steps: 76/90 Loss: 0.0156 torch.Size([8, 600, 800])
  37593. torch.Size([8, 8])
  37594. tensor([[0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904],
  37595. [ nan, nan, 0.7981, 0.3194, 0.3625, 0.3167, 0.5040, 0.5563],
  37596. [0.6213, 0.4131, 0.8438, 0.3550, 0.3513, 0.4400, 0.5716, 0.5123],
  37597. [0.6043, 0.4022, 0.6887, 0.1983, 0.3775, 0.2483, 0.5480, 0.5136],
  37598. [0.6275, 0.4008, 0.8047, 0.2484, 0.4037, 0.2777, 0.6506, 0.5212],
  37599. [0.6179, 0.4008, 0.7505, 0.2678, 0.4368, 0.1891, 0.5831, 0.5263],
  37600. [ nan, nan, 0.7425, 0.2117, 0.3937, 0.2433, 0.5438, 0.5567],
  37601. [0.6249, 0.4142, 0.8350, 0.3283, 0.3613, 0.3700, 0.6188, 0.5400]],
  37602. device='cuda:0', dtype=torch.float64)
  37603. predictions are: tensor([[ 0.6656, -0.3252, 1.6314, -0.9552, -0.1234, -1.3587, 0.5424, 0.1389],
  37604. [-1.9349, -2.0047, 1.2804, -0.8710, -0.5206, -0.9053, 0.0334, 0.3029],
  37605. [ 0.6856, -0.3197, 1.6577, -0.5465, -0.5838, -0.3089, 0.2376, 0.2427],
  37606. [ 0.6226, -0.3830, 0.9294, -1.1398, -0.4260, -1.1295, 0.2986, 0.2081],
  37607. [ 0.6480, -0.3885, 1.4777, -0.9221, -0.3241, -0.9524, 0.6381, 0.1960],
  37608. [ 0.8055, -0.2722, 1.2072, -0.9814, -0.2422, -1.3769, 0.3189, 0.2356],
  37609. [-1.9116, -2.0022, 1.1542, -1.1836, -0.3305, -1.1824, 0.2033, 0.3075],
  37610. [ 0.7482, -0.2765, 1.6254, -0.7034, -0.5661, -0.6050, 0.4760, 0.3313]],
  37611. device='cuda:0', grad_fn=<AddmmBackward>)
  37612. landmarks are: tensor([[[ 0.6126, -0.3871, 1.6229, -1.0773, -0.2132, -1.3698, 0.5829,
  37613. -0.0209],
  37614. [-2.2859, -2.2859, 1.4006, -0.8105, -0.6115, -0.8232, 0.0419,
  37615. 0.2837],
  37616. [ 0.5841, -0.3778, 1.6113, -0.6462, -0.6635, -0.2536, 0.3542,
  37617. 0.0802],
  37618. [ 0.5053, -0.4281, 0.8954, -1.3698, -0.5423, -1.1389, 0.2453,
  37619. 0.0862],
  37620. [ 0.6125, -0.4345, 1.4308, -1.1384, -0.4213, -1.0031, 0.7190,
  37621. 0.1214],
  37622. [ 0.5680, -0.4345, 1.1806, -1.0490, -0.2683, -1.4127, 0.4074,
  37623. 0.1449],
  37624. [-2.2859, -2.2859, 1.1436, -1.3082, -0.4672, -1.1620, 0.2256,
  37625. 0.2853],
  37626. [ 0.6006, -0.3728, 1.5709, -0.7694, -0.6173, -0.5769, 0.5721,
  37627. 0.2083]]], device='cuda:0')
  37628. loss_train_step before backward: tensor(0.0152, device='cuda:0', grad_fn=<MseLossBackward>)
  37629. loss_train_step after backward: tensor(0.0152, device='cuda:0', grad_fn=<MseLossBackward>)
  37630. loss_train: 1.2041279622353613
  37631. step: 77
  37632. running loss: 0.01563802548357612
  37633. Train Steps: 77/90 Loss: 0.0156 torch.Size([8, 600, 800])
  37634. torch.Size([8, 8])
  37635. tensor([[0.6207, 0.4110, 0.8738, 0.5000, 0.4800, 0.5633, 0.6300, 0.5433],
  37636. [0.6138, 0.4101, 0.8800, 0.5083, 0.4637, 0.5950, 0.5587, 0.5077],
  37637. [0.6164, 0.3956, 0.8757, 0.5088, 0.4300, 0.5320, 0.6022, 0.5202],
  37638. [0.6185, 0.4098, 0.8838, 0.4900, 0.4537, 0.5800, 0.6288, 0.5400],
  37639. [0.6132, 0.4037, 0.6963, 0.2217, 0.4100, 0.1950, 0.5395, 0.5175],
  37640. [0.6086, 0.3940, 0.8712, 0.4783, 0.4025, 0.4900, 0.5498, 0.5390],
  37641. [0.6272, 0.4120, 0.9038, 0.4117, 0.3725, 0.3200, 0.6175, 0.5250],
  37642. [0.6230, 0.4152, 0.7588, 0.2283, 0.4012, 0.2883, 0.6200, 0.5767]],
  37643. device='cuda:0', dtype=torch.float64)
  37644. predictions are: tensor([[ 0.6204, -0.3751, 1.7329, -0.1243, -0.1641, 0.0857, 0.3867, 0.3459],
  37645. [ 0.4157, -0.4951, 1.7252, -0.0288, -0.2015, 0.1722, 0.2050, 0.1587],
  37646. [ 0.5428, -0.4652, 1.6959, -0.0345, -0.3847, 0.0054, 0.4109, 0.1474],
  37647. [ 0.4291, -0.5127, 1.7436, -0.1355, -0.2256, 0.2061, 0.4410, 0.2915],
  37648. [ 0.5248, -0.4014, 0.9346, -1.2211, -0.4824, -1.5760, 0.1073, 0.0700],
  37649. [ 0.6467, -0.4007, 1.6932, -0.1175, -0.4838, -0.1200, 0.0742, 0.1444],
  37650. [ 0.4221, -0.5190, 1.8474, -0.4413, -0.6660, -0.8941, 0.4888, 0.2100],
  37651. [ 0.5255, -0.4244, 1.2112, -1.0893, -0.5817, -1.1285, 0.5264, 0.3707]],
  37652. device='cuda:0', grad_fn=<AddmmBackward>)
  37653. landmarks are: tensor([[[ 0.5811, -0.3878, 1.7499, 0.0236, -0.0688, 0.3161, 0.6240,
  37654. 0.2237],
  37655. [ 0.5491, -0.3918, 1.7788, 0.0620, -0.1439, 0.4624, 0.2946,
  37656. 0.0592],
  37657. [ 0.5614, -0.4586, 1.7587, 0.0641, -0.2998, 0.1712, 0.4958,
  37658. 0.1170],
  37659. [ 0.5709, -0.3933, 1.7961, -0.0226, -0.1901, 0.3931, 0.6182,
  37660. 0.2083],
  37661. [ 0.5465, -0.4214, 0.9300, -1.2620, -0.3921, -1.3852, 0.2062,
  37662. 0.1043],
  37663. [ 0.5250, -0.4661, 1.7383, -0.0765, -0.4268, -0.0226, 0.2535,
  37664. 0.2035],
  37665. [ 0.6111, -0.3828, 1.8885, -0.3844, -0.5654, -0.8079, 0.5663,
  37666. 0.1390],
  37667. [ 0.5915, -0.3682, 1.2187, -1.2313, -0.4326, -0.9541, 0.5778,
  37668. 0.3777]]], device='cuda:0')
  37669. loss_train_step before backward: tensor(0.0122, device='cuda:0', grad_fn=<MseLossBackward>)
  37670. loss_train_step after backward: tensor(0.0122, device='cuda:0', grad_fn=<MseLossBackward>)
  37671. loss_train: 1.2163738389499485
  37672. step: 78
  37673. running loss: 0.015594536396794213
  37674. Train Steps: 78/90 Loss: 0.0156 torch.Size([8, 600, 800])
  37675. torch.Size([8, 8])
  37676. tensor([[0.6286, 0.4086, 0.8408, 0.2801, 0.4163, 0.2800, 0.6725, 0.5393],
  37677. [0.6202, 0.3983, 0.8555, 0.5429, 0.3842, 0.4370, 0.5866, 0.5398],
  37678. [ nan, nan, 0.6992, 0.2791, 0.3950, 0.2383, 0.5483, 0.5819],
  37679. [0.6157, 0.4102, 0.8513, 0.3817, 0.3613, 0.3667, 0.5096, 0.5890],
  37680. [0.6218, 0.4185, 0.7338, 0.2650, 0.4625, 0.1950, 0.5687, 0.5800],
  37681. [0.6207, 0.4110, 0.8738, 0.5000, 0.4800, 0.5633, 0.6300, 0.5433],
  37682. [0.6224, 0.4061, 0.8988, 0.4300, 0.3838, 0.4750, 0.6112, 0.5483],
  37683. [0.6101, 0.4042, 0.7775, 0.2617, 0.3713, 0.2817, 0.5440, 0.5650]],
  37684. device='cuda:0', dtype=torch.float64)
  37685. predictions are: tensor([[ 0.4975, -0.4646, 1.5513, -0.9666, -0.3757, -1.0631, 0.8349, 0.2213],
  37686. [ 0.6141, -0.4448, 1.6689, 0.2204, -0.5464, -0.2051, 0.5138, 0.0906],
  37687. [-2.0015, -2.1097, 0.9544, -1.1119, -0.4661, -1.1842, 0.0765, 0.2852],
  37688. [ 0.3614, -0.5117, 1.6381, -0.5652, -0.6371, -0.5385, 0.0116, 0.2458],
  37689. [ 0.4667, -0.4367, 1.1219, -1.0317, -0.2074, -1.4519, 0.2099, 0.2799],
  37690. [ 0.6129, -0.3851, 1.7800, -0.1275, -0.1141, 0.2926, 0.4631, 0.2179],
  37691. [ 0.5707, -0.4434, 1.8278, -0.3480, -0.5337, 0.0171, 0.5358, 0.1088],
  37692. [ 0.4578, -0.4612, 1.2927, -1.0640, -0.5458, -1.0070, 0.0641, 0.2277]],
  37693. device='cuda:0', grad_fn=<AddmmBackward>)
  37694. landmarks are: tensor([[[ 0.6174, -0.3984, 1.5975, -0.9921, -0.3633, -0.9926, 0.8205,
  37695. 0.2050],
  37696. [ 0.5786, -0.4463, 1.6655, 0.2216, -0.5115, -0.2675, 0.4236,
  37697. 0.2075],
  37698. [-2.2859, -2.2859, 0.9438, -0.9967, -0.4614, -1.1851, 0.2468,
  37699. 0.4019],
  37700. [ 0.5581, -0.3912, 1.6460, -0.5230, -0.6173, -0.5923, 0.0681,
  37701. 0.4348],
  37702. [ 0.5862, -0.3530, 1.1032, -1.0619, -0.1497, -1.3852, 0.3411,
  37703. 0.3931],
  37704. [ 0.5811, -0.3878, 1.7499, 0.0236, -0.0688, 0.3161, 0.6240,
  37705. 0.2237],
  37706. [ 0.5888, -0.4101, 1.8654, -0.2998, -0.5134, -0.0919, 0.5374,
  37707. 0.2468],
  37708. [ 0.5320, -0.4189, 1.3053, -1.0773, -0.5711, -0.9849, 0.2267,
  37709. 0.3237]]], device='cuda:0')
  37710. loss_train_step before backward: tensor(0.0080, device='cuda:0', grad_fn=<MseLossBackward>)
  37711. loss_train_step after backward: tensor(0.0080, device='cuda:0', grad_fn=<MseLossBackward>)
  37712. loss_train: 1.224423078354448
  37713. step: 79
  37714. running loss: 0.015499026308284152
  37715.  
  37716. Train Steps: 79/90 Loss: 0.0155 torch.Size([8, 600, 800])
  37717. torch.Size([8, 8])
  37718. tensor([[0.6091, 0.3997, 0.8314, 0.4334, 0.3788, 0.4550, 0.5213, 0.5656],
  37719. [0.6250, 0.4103, 0.8950, 0.4400, 0.3912, 0.5650, 0.6050, 0.5133],
  37720. [0.6225, 0.4196, 0.8788, 0.3467, 0.3750, 0.3400, 0.5413, 0.5433],
  37721. [0.6153, 0.4119, 0.8463, 0.3833, 0.3600, 0.3200, 0.5106, 0.5563],
  37722. [0.6138, 0.4101, 0.8800, 0.5083, 0.4637, 0.5950, 0.5587, 0.5077],
  37723. [0.6175, 0.4013, 0.7734, 0.2783, 0.4475, 0.1786, 0.5790, 0.5351],
  37724. [0.6255, 0.4017, 0.8688, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901],
  37725. [0.6143, 0.4034, 0.8800, 0.4833, 0.4512, 0.5367, 0.5289, 0.5097]],
  37726. device='cuda:0', dtype=torch.float64)
  37727. predictions are: tensor([[ 0.4846, -0.4914, 1.4329, -0.3428, -0.5375, -0.1980, 0.2619, 0.3136],
  37728. [ 0.5042, -0.4608, 1.7244, -0.3396, -0.4670, 0.3175, 0.5917, 0.2142],
  37729. [ 0.4605, -0.4590, 1.5435, -0.6849, -0.6888, -0.7761, 0.2342, 0.2381],
  37730. [ 0.2250, -0.6154, 1.5077, -0.5573, -0.6753, -0.8241, 0.0768, 0.3247],
  37731. [ 0.4591, -0.4868, 1.6747, -0.0019, -0.1548, 0.3512, 0.3674, 0.1820],
  37732. [ 0.7229, -0.3085, 1.1871, -1.0721, -0.2958, -1.5675, 0.3066, 0.1513],
  37733. [ 0.5734, -0.4881, 1.6056, -0.7844, -0.6470, -0.6891, 0.7640, 0.0590],
  37734. [ 0.4164, -0.5204, 1.6723, -0.1330, -0.2016, 0.1220, 0.1374, 0.1163]],
  37735. device='cuda:0', grad_fn=<AddmmBackward>)
  37736. landmarks are: tensor([[[ 0.5275, -0.4396, 1.5543, -0.2841, -0.5365, -0.1843, 0.1221,
  37737. 0.3265],
  37738. [ 0.6010, -0.3909, 1.8480, -0.2536, -0.4788, 0.3238, 0.5085,
  37739. 0.0851],
  37740. [ 0.5894, -0.3479, 1.7730, -0.6847, -0.5538, -0.7155, 0.2141,
  37741. 0.2237],
  37742. [ 0.5561, -0.3834, 1.6229, -0.5153, -0.6231, -0.8079, 0.0727,
  37743. 0.2837],
  37744. [ 0.5491, -0.3918, 1.7788, 0.0620, -0.1439, 0.4624, 0.2946,
  37745. 0.0592],
  37746. [ 0.5664, -0.4321, 1.2862, -1.0003, -0.2189, -1.4608, 0.3883,
  37747. 0.1855],
  37748. [ 0.6031, -0.4307, 1.7268, -0.8002, -0.6058, -0.6462, 0.6442,
  37749. -0.0220],
  37750. [ 0.5515, -0.4225, 1.7788, -0.0534, -0.2016, 0.1929, 0.1568,
  37751. 0.0682]]], device='cuda:0')
  37752. loss_train_step before backward: tensor(0.0099, device='cuda:0', grad_fn=<MseLossBackward>)
  37753. loss_train_step after backward: tensor(0.0099, device='cuda:0', grad_fn=<MseLossBackward>)
  37754. loss_train: 1.234301685821265
  37755. step: 80
  37756. running loss: 0.015428771072765812
  37757. Train Steps: 80/90 Loss: 0.0154 torch.Size([8, 600, 800])
  37758. torch.Size([8, 8])
  37759. tensor([[0.6284, 0.4029, 0.8838, 0.3783, 0.3975, 0.2850, 0.6335, 0.5090],
  37760. [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150],
  37761. [0.6087, 0.3951, 0.8387, 0.5833, 0.4188, 0.4933, 0.5146, 0.4830],
  37762. [0.6143, 0.4034, 0.8800, 0.4833, 0.4512, 0.5367, 0.5289, 0.5097],
  37763. [0.6257, 0.4024, 0.8672, 0.5422, 0.4196, 0.5198, 0.6694, 0.5204],
  37764. [ nan, nan, 0.8363, 0.3317, 0.3563, 0.3367, 0.5329, 0.5142],
  37765. [0.6109, 0.3943, 0.8525, 0.4950, 0.4338, 0.4800, 0.5449, 0.5383],
  37766. [ nan, nan, 0.8300, 0.3150, 0.3588, 0.3383, 0.5208, 0.5194]],
  37767. device='cuda:0', dtype=torch.float64)
  37768. predictions are: tensor([[ 0.7503, -0.2881, 1.6015, -0.6986, -0.4646, -1.0654, 0.6464, 0.1602],
  37769. [ 0.7804, -0.2480, 1.6838, -0.2485, -0.3454, 0.3726, 0.5508, 0.1698],
  37770. [ 0.6938, -0.3236, 1.5166, 0.2860, -0.4027, -0.0619, 0.1601, 0.1312],
  37771. [ 0.5815, -0.3628, 1.6414, -0.2030, -0.2527, 0.0957, 0.1165, 0.0985],
  37772. [ 0.6586, -0.3652, 1.5467, 0.0677, -0.4092, 0.0632, 0.8244, 0.2094],
  37773. [-2.3185, -2.3273, 1.3828, -0.8933, -0.6300, -0.7669, 0.1151, 0.2224],
  37774. [ 0.6937, -0.3142, 1.5366, -0.1562, -0.3201, -0.1382, 0.1785, 0.2307],
  37775. [-2.0555, -2.1445, 1.3345, -0.8968, -0.6410, -0.7471, 0.0783, 0.2306]],
  37776. device='cuda:0', grad_fn=<AddmmBackward>)
  37777. landmarks are: tensor([[[ 6.1645e-01, -4.2487e-01, 1.7961e+00, -5.3841e-01, -4.4988e-01,
  37778. -9.6952e-01, 6.4006e-01, 6.5205e-02],
  37779. [ 6.0139e-01, -3.8830e-01, 1.8192e+00, -1.1501e-01, -2.8822e-01,
  37780. 4.0077e-01, 5.2009e-01, 9.2841e-02],
  37781. [ 5.2575e-01, -4.6105e-01, 1.5882e+00, 4.0847e-01, -3.5173e-01,
  37782. -7.2363e-03, 9.1027e-02, -5.5027e-02],
  37783. [ 5.5155e-01, -4.2249e-01, 1.7788e+00, -5.3426e-02, -2.0162e-01,
  37784. 1.9292e-01, 1.5683e-01, 6.8210e-02],
  37785. [ 6.0425e-01, -4.2731e-01, 1.7198e+00, 2.1845e-01, -3.4783e-01,
  37786. 1.1492e-01, 8.0616e-01, 1.1755e-01],
  37787. [-2.2859e+00, -2.2859e+00, 1.5767e+00, -7.5396e-01, -6.4042e-01,
  37788. -7.3087e-01, 1.7534e-01, 8.9251e-02],
  37789. [ 5.3603e-01, -4.6490e-01, 1.6517e+00, 4.6189e-04, -2.8245e-01,
  37790. -6.8822e-02, 2.3086e-01, 2.0046e-01],
  37791. [-2.2859e+00, -2.2859e+00, 1.5478e+00, -8.3095e-01, -6.2887e-01,
  37792. -7.2317e-01, 1.1982e-01, 1.1330e-01]]], device='cuda:0')
  37793. loss_train_step before backward: tensor(0.0122, device='cuda:0', grad_fn=<MseLossBackward>)
  37794. loss_train_step after backward: tensor(0.0122, device='cuda:0', grad_fn=<MseLossBackward>)
  37795. loss_train: 1.246514733415097
  37796. step: 81
  37797. running loss: 0.015389070782902432
  37798. Train Steps: 81/90 Loss: 0.0154 torch.Size([8, 600, 800])
  37799. torch.Size([8, 8])
  37800. tensor([[0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407],
  37801. [0.6196, 0.4088, 0.8888, 0.4583, 0.4500, 0.5683, 0.6138, 0.5883],
  37802. [0.6211, 0.3935, 0.8636, 0.4841, 0.4417, 0.5126, 0.6331, 0.5268],
  37803. [0.6251, 0.4108, 0.8888, 0.4700, 0.4325, 0.5817, 0.6075, 0.5150],
  37804. [0.6086, 0.3981, 0.8700, 0.4750, 0.4512, 0.5283, 0.5324, 0.5038],
  37805. [0.6227, 0.4049, 0.8750, 0.5367, 0.3775, 0.3667, 0.5725, 0.5317],
  37806. [0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750],
  37807. [0.6078, 0.4033, 0.8019, 0.3055, 0.3450, 0.4200, 0.6025, 0.5550]],
  37808. device='cuda:0', dtype=torch.float64)
  37809. predictions are: tensor([[ 0.5458, -0.3904, 1.5569, -0.4111, -0.6724, -0.8364, 0.0369, 0.1736],
  37810. [ 0.5075, -0.4487, 1.7388, -0.2161, -0.2107, 0.4441, 0.5468, 0.3925],
  37811. [ 0.4614, -0.5432, 1.6451, -0.0553, -0.2670, 0.1172, 0.6347, 0.1695],
  37812. [ 0.6589, -0.3599, 1.7563, -0.1110, -0.3010, 0.5124, 0.5206, 0.1081],
  37813. [ 0.6011, -0.4098, 1.6841, -0.1285, -0.2343, 0.1414, 0.1193, -0.0113],
  37814. [ 0.6166, -0.4051, 1.6784, 0.2026, -0.5937, -0.5582, 0.3913, 0.1671],
  37815. [-0.2581, -0.9259, 1.3314, -0.9020, -0.5419, -1.0859, 0.0955, 0.2768],
  37816. [ 0.4362, -0.5123, 1.3979, -0.9372, -0.7266, -0.3351, 0.5126, 0.2498]],
  37817. device='cuda:0', grad_fn=<AddmmBackward>)
  37818. landmarks are: tensor([[[ 0.5470, -0.3952, 1.6377, -0.4253, -0.6289, -0.8079, 0.0249,
  37819. 0.2116],
  37820. [ 0.5759, -0.3978, 1.8192, -0.1689, -0.2074, 0.3392, 0.5490,
  37821. 0.4316],
  37822. [ 0.5828, -0.4682, 1.7031, -0.0497, -0.2458, 0.0818, 0.6381,
  37823. 0.1474],
  37824. [ 0.6014, -0.3883, 1.8192, -0.1150, -0.2882, 0.4008, 0.5201,
  37825. 0.0928],
  37826. [ 0.5249, -0.4473, 1.7326, -0.0919, -0.2016, 0.1544, 0.1733,
  37827. 0.0412],
  37828. [ 0.5901, -0.4157, 1.7557, 0.1929, -0.5423, -0.5923, 0.3584,
  37829. 0.1698],
  37830. [ 0.5695, -0.3927, 1.3226, -0.9002, -0.4672, -1.1928, 0.1737,
  37831. 0.3700],
  37832. [ 0.5215, -0.4232, 1.4182, -0.8747, -0.6924, -0.3460, 0.4970,
  37833. 0.2776]]], device='cuda:0')
  37834. loss_train_step before backward: tensor(0.0178, device='cuda:0', grad_fn=<MseLossBackward>)
  37835. loss_train_step after backward: tensor(0.0178, device='cuda:0', grad_fn=<MseLossBackward>)
  37836. loss_train: 1.264311611186713
  37837. step: 82
  37838. running loss: 0.015418434282764792
  37839. Train Steps: 82/90 Loss: 0.0154 torch.Size([8, 600, 800])
  37840. torch.Size([8, 8])
  37841. tensor([[0.6264, 0.4067, 0.9050, 0.4183, 0.3775, 0.4600, 0.6308, 0.4862],
  37842. [0.6329, 0.4055, 0.9050, 0.4783, 0.3613, 0.3917, 0.6464, 0.5019],
  37843. [0.6117, 0.4019, 0.8538, 0.4067, 0.3513, 0.3583, 0.5663, 0.5133],
  37844. [0.6250, 0.4013, 0.8525, 0.5417, 0.4037, 0.5117, 0.6325, 0.5017],
  37845. [0.6128, 0.4118, 0.8638, 0.5333, 0.4625, 0.5267, 0.5193, 0.5475],
  37846. [0.6201, 0.3970, 0.8413, 0.4950, 0.4413, 0.5183, 0.6088, 0.5400],
  37847. [0.6260, 0.4253, 0.8938, 0.4400, 0.3675, 0.3650, 0.5675, 0.5767],
  37848. [0.6161, 0.4024, 0.8838, 0.4583, 0.3688, 0.3733, 0.5311, 0.5344]],
  37849. device='cuda:0', dtype=torch.float64)
  37850. predictions are: tensor([[ 0.5646, -0.4298, 1.7529, -0.4515, -0.6126, -0.0544, 0.5670, 0.0215],
  37851. [ 0.6456, -0.4035, 1.7761, -0.2404, -0.5597, -0.4027, 0.7305, 0.0774],
  37852. [ 0.5289, -0.4356, 1.5772, -0.5750, -0.6524, -0.5157, 0.1538, 0.1797],
  37853. [ 0.3057, -0.6639, 1.6080, 0.1656, -0.4046, 0.2433, 0.5498, 0.0604],
  37854. [ 0.4928, -0.4453, 1.6299, 0.1164, -0.1336, 0.3010, 0.0359, 0.2095],
  37855. [ 0.4195, -0.5548, 1.5906, -0.0317, -0.2286, 0.2824, 0.4368, 0.2617],
  37856. [ 0.5216, -0.3977, 1.7365, -0.2763, -0.5934, -0.4298, 0.2651, 0.4191],
  37857. [ 0.5687, -0.4265, 1.7573, -0.2624, -0.5796, -0.3919, 0.1728, 0.1689]],
  37858. device='cuda:0', grad_fn=<AddmmBackward>)
  37859. landmarks are: tensor([[[ 6.0722e-01, -4.0747e-01, 1.8942e+00, -3.5366e-01, -5.4226e-01,
  37860. -1.6120e-01, 6.2772e-01, -3.9998e-02],
  37861. [ 6.3718e-01, -4.1286e-01, 1.8942e+00, -7.6520e-02, -6.1732e-01,
  37862. -4.7683e-01, 6.9989e-01, 3.2524e-02],
  37863. [ 5.3926e-01, -4.2941e-01, 1.6575e+00, -4.0754e-01, -6.6351e-01,
  37864. -6.3079e-01, 3.2956e-01, 8.5142e-02],
  37865. [ 6.0095e-01, -4.3212e-01, 1.6517e+00, 2.1601e-01, -4.2102e-01,
  37866. 7.7444e-02, 6.3557e-01, 3.1255e-02],
  37867. [ 5.4475e-01, -3.8383e-01, 1.7037e+00, 1.7752e-01, -1.4965e-01,
  37868. 1.4673e-01, 1.1283e-01, 2.4313e-01],
  37869. [ 5.7812e-01, -4.5219e-01, 1.5998e+00, 4.6189e-04, -2.4781e-01,
  37870. 1.0824e-01, 5.2587e-01, 2.0831e-01],
  37871. [ 6.0577e-01, -3.2156e-01, 1.8423e+00, -2.5358e-01, -5.8845e-01,
  37872. -6.0000e-01, 3.3533e-01, 3.7768e-01],
  37873. [ 5.5978e-01, -4.2731e-01, 1.7961e+00, -1.6890e-01, -5.8268e-01,
  37874. -5.6151e-01, 1.6711e-01, 1.8243e-01]]], device='cuda:0')
  37875. loss_train_step before backward: tensor(0.0094, device='cuda:0', grad_fn=<MseLossBackward>)
  37876. loss_train_step after backward: tensor(0.0094, device='cuda:0', grad_fn=<MseLossBackward>)
  37877. loss_train: 1.2737219003029168
  37878. step: 83
  37879. running loss: 0.015346046991601407
  37880.  
  37881. Train Steps: 83/90 Loss: 0.0153 torch.Size([8, 600, 800])
  37882. torch.Size([8, 8])
  37883. tensor([[0.6275, 0.4013, 0.8850, 0.2833, 0.4975, 0.2233, 0.7058, 0.5343],
  37884. [0.6085, 0.4005, 0.8400, 0.4317, 0.3763, 0.4750, 0.5476, 0.5058],
  37885. [0.6115, 0.3998, 0.7063, 0.2383, 0.4037, 0.1950, 0.5320, 0.4993],
  37886. [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083],
  37887. [0.6090, 0.4045, 0.7250, 0.2100, 0.4075, 0.2300, 0.5476, 0.5663],
  37888. [0.6218, 0.4185, 0.7338, 0.2650, 0.4625, 0.1950, 0.5687, 0.5800],
  37889. [0.6026, 0.3979, 0.8550, 0.4233, 0.3613, 0.5233, 0.5582, 0.4967],
  37890. [0.6058, 0.3986, 0.8324, 0.4626, 0.3838, 0.4983, 0.5147, 0.5466]],
  37891. device='cuda:0', dtype=torch.float64)
  37892. predictions are: tensor([[ 0.5970, -0.4386, 1.7881, -0.9258, 0.0162, -1.1350, 1.0138, 0.0702],
  37893. [ 0.6098, -0.4173, 1.6972, -0.2384, -0.5802, 0.1849, 0.2896, 0.0095],
  37894. [ 0.4448, -0.4789, 1.0836, -1.2496, -0.4350, -1.2242, 0.1473, -0.0186],
  37895. [ 0.7261, -0.3231, 1.8236, 0.1307, -0.6321, 0.0100, 0.5721, 0.3982],
  37896. [-0.4759, -1.1006, 1.1537, -1.1627, -0.4673, -0.9932, 0.2593, 0.3161],
  37897. [ 0.4587, -0.4426, 1.1685, -1.0159, -0.1973, -1.2388, 0.2722, 0.3378],
  37898. [ 0.6435, -0.4056, 1.7502, -0.2111, -0.6102, 0.3916, 0.4004, -0.0266],
  37899. [ 0.5203, -0.4551, 1.6436, -0.0169, -0.4993, 0.3264, 0.1976, 0.1663]],
  37900. device='cuda:0', grad_fn=<AddmmBackward>)
  37901. landmarks are: tensor([[[ 0.6125, -0.4321, 1.8018, -0.9772, 0.0120, -1.2543, 0.9741,
  37902. 0.1821],
  37903. [ 0.5245, -0.4361, 1.5940, -0.2921, -0.5480, -0.0919, 0.2432,
  37904. 0.0502],
  37905. [ 0.5384, -0.4393, 0.9762, -1.1851, -0.4210, -1.3852, 0.1712,
  37906. 0.0201],
  37907. [ 0.5713, -0.4105, 1.7557, 0.0467, -0.6520, -0.2690, 0.3988,
  37908. 0.5239],
  37909. [ 0.5269, -0.4176, 1.0628, -1.3159, -0.4037, -1.2236, 0.2432,
  37910. 0.3297],
  37911. [ 0.5862, -0.3530, 1.1032, -1.0619, -0.1497, -1.3852, 0.3411,
  37912. 0.3931],
  37913. [ 0.4974, -0.4482, 1.6633, -0.3306, -0.6173, 0.1313, 0.2925,
  37914. 0.0081],
  37915. [ 0.5124, -0.4446, 1.5587, -0.1493, -0.5134, 0.0159, 0.0912,
  37916. 0.2386]]], device='cuda:0')
  37917. loss_train_step before backward: tensor(0.0350, device='cuda:0', grad_fn=<MseLossBackward>)
  37918. loss_train_step after backward: tensor(0.0350, device='cuda:0', grad_fn=<MseLossBackward>)
  37919. loss_train: 1.3087662127800286
  37920. step: 84
  37921. running loss: 0.015580550152143198
  37922. Train Steps: 84/90 Loss: 0.0156 torch.Size([8, 600, 800])
  37923. torch.Size([8, 8])
  37924. tensor([[0.6229, 0.4198, 0.7662, 0.2700, 0.4700, 0.2133, 0.5675, 0.5783],
  37925. [0.6133, 0.4094, 0.8495, 0.4028, 0.3588, 0.3200, 0.5003, 0.5407],
  37926. [0.6144, 0.4032, 0.8563, 0.3283, 0.3525, 0.4200, 0.5775, 0.5583],
  37927. [0.6261, 0.4029, 0.8720, 0.3364, 0.3665, 0.3753, 0.6531, 0.5183],
  37928. [0.6210, 0.4164, 0.7202, 0.2930, 0.4025, 0.2483, 0.5687, 0.5567],
  37929. [0.6201, 0.4098, 0.8575, 0.3617, 0.3700, 0.3550, 0.5988, 0.5667],
  37930. [0.6095, 0.4002, 0.8533, 0.5168, 0.5031, 0.5094, 0.5125, 0.5433],
  37931. [0.6275, 0.4050, 0.9038, 0.3767, 0.3838, 0.3533, 0.7074, 0.5575]],
  37932. device='cuda:0', dtype=torch.float64)
  37933. predictions are: tensor([[ 3.2658e-01, -5.4547e-01, 1.2428e+00, -1.0113e+00, -1.1611e-01,
  37934. -1.1311e+00, 2.8831e-01, 3.0673e-01],
  37935. [ 4.7822e-01, -4.6463e-01, 1.6447e+00, -3.4109e-01, -5.8065e-01,
  37936. -7.2646e-01, 7.7841e-02, 1.1476e-01],
  37937. [ 4.7345e-01, -4.9782e-01, 1.6934e+00, -7.2943e-01, -6.7776e-01,
  37938. -2.3362e-01, 3.6246e-01, 1.4206e-01],
  37939. [ 6.8065e-01, -4.0213e-01, 1.7472e+00, -7.7854e-01, -5.4823e-01,
  37940. -4.4851e-01, 7.6350e-01, 1.3053e-03],
  37941. [ 7.1257e-02, -7.2028e-01, 1.1654e+00, -9.3299e-01, -4.1678e-01,
  37942. -8.6892e-01, 3.4502e-01, 2.8765e-01],
  37943. [ 5.8301e-01, -4.2550e-01, 1.6716e+00, -5.7195e-01, -5.2895e-01,
  37944. -4.7976e-01, 4.6565e-01, 2.3277e-01],
  37945. [ 5.5077e-01, -4.6933e-01, 1.6813e+00, 2.6087e-01, 1.6187e-02,
  37946. 3.4759e-01, 3.8679e-03, 1.3716e-01],
  37947. [ 4.8412e-01, -5.3995e-01, 1.8609e+00, -4.5671e-01, -4.9464e-01,
  37948. -4.9666e-01, 1.0809e+00, 2.0112e-01]], device='cuda:0',
  37949. grad_fn=<AddmmBackward>)
  37950. landmarks are: tensor([[[ 0.5912, -0.3467, 1.2533, -1.0388, -0.1150, -1.3005, 0.3353,
  37951. 0.3854],
  37952. [ 0.5470, -0.3952, 1.6377, -0.4253, -0.6289, -0.8079, 0.0249,
  37953. 0.2116],
  37954. [ 0.5520, -0.4237, 1.6691, -0.7694, -0.6577, -0.3460, 0.3815,
  37955. 0.2930],
  37956. [ 0.6059, -0.4249, 1.7420, -0.7321, -0.5931, -0.5523, 0.7307,
  37957. 0.1080],
  37958. [ 0.5824, -0.3625, 1.0405, -0.9325, -0.4268, -1.1389, 0.3411,
  37959. 0.2853],
  37960. [ 0.5783, -0.3933, 1.6748, -0.6154, -0.5769, -0.6462, 0.4797,
  37961. 0.3315],
  37962. [ 0.5295, -0.4373, 1.6553, 0.1011, 0.0380, 0.0671, 0.0813,
  37963. 0.2237],
  37964. [ 0.6125, -0.4153, 1.8885, -0.5461, -0.5134, -0.6539, 0.9814,
  37965. 0.2890]]], device='cuda:0')
  37966. loss_train_step before backward: tensor(0.0155, device='cuda:0', grad_fn=<MseLossBackward>)
  37967. loss_train_step after backward: tensor(0.0155, device='cuda:0', grad_fn=<MseLossBackward>)
  37968. loss_train: 1.3242887216620147
  37969. step: 85
  37970. running loss: 0.015579867313670762
  37971. Train Steps: 85/90 Loss: 0.0156 torch.Size([8, 600, 800])
  37972. torch.Size([8, 8])
  37973. tensor([[0.6126, 0.4073, 0.8750, 0.5133, 0.3800, 0.4333, 0.4986, 0.5378],
  37974. [0.6197, 0.4118, 0.8688, 0.5517, 0.4037, 0.5233, 0.5875, 0.5600],
  37975. [0.6246, 0.4090, 0.6964, 0.2027, 0.3925, 0.2683, 0.6088, 0.5583],
  37976. [0.6040, 0.4002, 0.7338, 0.2267, 0.3975, 0.2100, 0.5231, 0.4778],
  37977. [0.6260, 0.4214, 0.8538, 0.5500, 0.3663, 0.3767, 0.5587, 0.5983],
  37978. [0.6147, 0.4081, 0.8538, 0.3400, 0.3663, 0.3150, 0.5142, 0.4875],
  37979. [0.6195, 0.4101, 0.8650, 0.3917, 0.3575, 0.4033, 0.5800, 0.5650],
  37980. [0.6207, 0.4081, 0.7662, 0.2067, 0.3962, 0.3200, 0.6312, 0.5300]],
  37981. device='cuda:0', dtype=torch.float64)
  37982. predictions are: tensor([[ 0.4749, -0.4938, 1.8200, 0.1421, -0.4529, -0.0682, 0.2061, 0.1641],
  37983. [ 0.4511, -0.5150, 1.8038, 0.3654, -0.4010, 0.4584, 0.4593, 0.1613],
  37984. [ 0.4468, -0.5258, 1.0679, -1.3597, -0.4119, -0.9884, 0.5265, 0.2220],
  37985. [ 0.4100, -0.5387, 1.1669, -1.3780, -0.4079, -1.1665, 0.2400, -0.0464],
  37986. [ 0.4981, -0.4564, 1.6810, 0.3457, -0.4863, -0.4005, 0.3559, 0.3938],
  37987. [ 0.6217, -0.4062, 1.7781, -0.7077, -0.5463, -0.6803, 0.2540, -0.0090],
  37988. [ 0.4976, -0.4992, 1.7516, -0.4075, -0.5858, -0.2284, 0.3990, 0.2871],
  37989. [ 0.5423, -0.4508, 1.3129, -1.3409, -0.4253, -0.7361, 0.7393, 0.1741]],
  37990. device='cuda:0', grad_fn=<AddmmBackward>)
  37991. landmarks are: tensor([[[ 0.5435, -0.4045, 1.7557, 0.0851, -0.5307, -0.2844, 0.0172,
  37992. 0.1980],
  37993. [ 0.5762, -0.3840, 1.7268, 0.2622, -0.4210, 0.1313, 0.4277,
  37994. 0.3007],
  37995. [ 0.5992, -0.3968, 0.9307, -1.3497, -0.4730, -1.0465, 0.5259,
  37996. 0.2930],
  37997. [ 0.5037, -0.4375, 1.1032, -1.2390, -0.4499, -1.3159, 0.1301,
  37998. -0.0791],
  37999. [ 0.6055, -0.3393, 1.6575, 0.2545, -0.5942, -0.5461, 0.2949,
  38000. 0.4778],
  38001. [ 0.5532, -0.4008, 1.6575, -0.7155, -0.5942, -0.8309, 0.0890,
  38002. -0.0340],
  38003. [ 0.5757, -0.3917, 1.7095, -0.4768, -0.6346, -0.4229, 0.3931,
  38004. 0.3238],
  38005. [ 0.5809, -0.4011, 1.2533, -1.3313, -0.4557, -0.8079, 0.6298,
  38006. 0.1621]]], device='cuda:0')
  38007. loss_train_step before backward: tensor(0.0106, device='cuda:0', grad_fn=<MseLossBackward>)
  38008. loss_train_step after backward: tensor(0.0106, device='cuda:0', grad_fn=<MseLossBackward>)
  38009. loss_train: 1.3348985011689365
  38010. step: 86
  38011. running loss: 0.015522075594987634
  38012. Train Steps: 86/90 Loss: 0.0155 torch.Size([8, 600, 800])
  38013. torch.Size([8, 8])
  38014. tensor([[ nan, nan, 0.6793, 0.2110, 0.4012, 0.2167, 0.5112, 0.5583],
  38015. [0.6250, 0.4131, 0.8688, 0.2983, 0.4275, 0.2367, 0.6162, 0.5367],
  38016. [0.6169, 0.4108, 0.6821, 0.2722, 0.3825, 0.2550, 0.5550, 0.5717],
  38017. [0.6059, 0.4002, 0.7562, 0.2767, 0.3538, 0.3033, 0.5529, 0.5455],
  38018. [0.6271, 0.4020, 0.8375, 0.6083, 0.3925, 0.4867, 0.6037, 0.4626],
  38019. [0.6228, 0.4119, 0.7938, 0.2233, 0.4674, 0.1773, 0.6188, 0.5433],
  38020. [0.6102, 0.4001, 0.7738, 0.3583, 0.3463, 0.3800, 0.5524, 0.5689],
  38021. [0.6199, 0.4102, 0.8950, 0.4417, 0.4012, 0.5367, 0.6112, 0.5967]],
  38022. device='cuda:0', dtype=torch.float64)
  38023. predictions are: tensor([[-2.6211, -2.5212, 1.0486, -1.1651, -0.4450, -0.9817, 0.1317, 0.2193],
  38024. [ 0.6544, -0.3648, 1.8017, -0.8586, -0.2291, -1.0695, 0.6752, 0.0898],
  38025. [ 0.6067, -0.3832, 0.9838, -1.1507, -0.4227, -1.1178, 0.2797, 0.2733],
  38026. [ 0.4072, -0.5191, 1.3219, -1.0060, -0.5335, -0.8429, 0.3747, 0.1266],
  38027. [ 0.4585, -0.5462, 1.6707, 0.4150, -0.4383, 0.0116, 0.4843, -0.0506],
  38028. [ 0.5655, -0.4500, 1.4585, -1.1970, -0.0922, -1.2743, 0.6101, 0.1232],
  38029. [ 0.3720, -0.5204, 1.4266, -0.6348, -0.6016, -0.5184, 0.2348, 0.2190],
  38030. [ 0.3983, -0.5229, 1.9028, -0.1224, -0.4020, 0.3903, 0.5228, 0.3090]],
  38031. device='cuda:0', grad_fn=<AddmmBackward>)
  38032. landmarks are: tensor([[[-2.2859, -2.2859, 0.8516, -1.3112, -0.4326, -1.2851, 0.0755,
  38033. 0.2930],
  38034. [ 0.6011, -0.3778, 1.7268, -0.9079, -0.3113, -1.1928, 0.5605,
  38035. 0.1929],
  38036. [ 0.5633, -0.3883, 0.8648, -1.0288, -0.5192, -1.1081, 0.2776,
  38037. 0.3546],
  38038. [ 0.5129, -0.4374, 1.2072, -1.0080, -0.6520, -0.8848, 0.2679,
  38039. 0.2335],
  38040. [ 0.6105, -0.4293, 1.5824, 0.5239, -0.4730, -0.0380, 0.5025,
  38041. -0.1492],
  38042. [ 0.5908, -0.3832, 1.3804, -1.2543, -0.1270, -1.4671, 0.5721,
  38043. 0.2237],
  38044. [ 0.5327, -0.4381, 1.2880, -0.6308, -0.6866, -0.5307, 0.2658,
  38045. 0.3417],
  38046. [ 0.5772, -0.3913, 1.8480, -0.2459, -0.4326, 0.1929, 0.5374,
  38047. 0.4701]]], device='cuda:0')
  38048. loss_train_step before backward: tensor(0.0132, device='cuda:0', grad_fn=<MseLossBackward>)
  38049. loss_train_step after backward: tensor(0.0132, device='cuda:0', grad_fn=<MseLossBackward>)
  38050. loss_train: 1.3481103568337858
  38051. step: 87
  38052. running loss: 0.015495521342917078
  38053.  
  38054. Train Steps: 87/90 Loss: 0.0155 torch.Size([8, 600, 800])
  38055. torch.Size([8, 8])
  38056. tensor([[0.6255, 0.4017, 0.8688, 0.3217, 0.3638, 0.3550, 0.6344, 0.4901],
  38057. [0.6218, 0.4098, 0.7238, 0.1850, 0.4238, 0.2517, 0.6288, 0.5350],
  38058. [0.6275, 0.4111, 0.8463, 0.2617, 0.4487, 0.1983, 0.6211, 0.4904],
  38059. [0.6258, 0.4038, 0.8750, 0.4883, 0.3900, 0.5500, 0.6375, 0.5217],
  38060. [0.6173, 0.4013, 0.8713, 0.4640, 0.4800, 0.5700, 0.6025, 0.5600],
  38061. [0.6292, 0.4010, 0.8988, 0.4800, 0.3638, 0.3817, 0.6357, 0.5051],
  38062. [0.6143, 0.4055, 0.8150, 0.2767, 0.3825, 0.2567, 0.5173, 0.4901],
  38063. [0.6179, 0.3998, 0.8396, 0.3505, 0.3552, 0.3768, 0.6064, 0.5329]],
  38064. device='cuda:0', dtype=torch.float64)
  38065. predictions are: tensor([[ 0.2906, -0.6409, 1.6966, -0.7035, -0.6151, -0.6604, 0.4858, 0.0571],
  38066. [ 0.6713, -0.3423, 1.0141, -1.4186, -0.3000, -1.2024, 0.4727, 0.3262],
  38067. [ 0.7239, -0.2935, 1.5762, -1.0051, -0.1290, -1.1708, 0.5538, 0.0640],
  38068. [ 0.5389, -0.4618, 1.8069, 0.0371, -0.4642, 0.2881, 0.5613, 0.1595],
  38069. [ 0.3526, -0.5561, 1.7288, -0.0225, -0.1932, 0.3346, 0.4374, 0.2839],
  38070. [ 0.3348, -0.6103, 1.7632, -0.0713, -0.5542, -0.5833, 0.5219, 0.1729],
  38071. [ 0.6104, -0.3794, 1.3945, -0.9225, -0.5040, -1.0243, 0.0740, 0.1107],
  38072. [ 0.6039, -0.4215, 1.6051, -0.6760, -0.6397, -0.5594, 0.4211, 0.2860]],
  38073. device='cuda:0', grad_fn=<AddmmBackward>)
  38074. landmarks are: tensor([[[ 0.6031, -0.4307, 1.7268, -0.8002, -0.6058, -0.6462, 0.6442,
  38075. -0.0220],
  38076. [ 0.5861, -0.3929, 1.0570, -1.4314, -0.3286, -1.1235, 0.6182,
  38077. 0.1852],
  38078. [ 0.6126, -0.3871, 1.6229, -1.0773, -0.2132, -1.3698, 0.5829,
  38079. -0.0209],
  38080. [ 0.6047, -0.4209, 1.7557, -0.0303, -0.4845, 0.2545, 0.6587,
  38081. 0.1236],
  38082. [ 0.5652, -0.4325, 1.7388, -0.1429, -0.0688, 0.3469, 0.4970,
  38083. 0.3007],
  38084. [ 0.6204, -0.4336, 1.8654, -0.0688, -0.6058, -0.5230, 0.6503,
  38085. 0.0472],
  38086. [ 0.5515, -0.4129, 1.4785, -1.0080, -0.5192, -1.1004, 0.1034,
  38087. -0.0220],
  38088. [ 0.5680, -0.4393, 1.5920, -0.6672, -0.6453, -0.5457, 0.5149,
  38089. 0.1753]]], device='cuda:0')
  38090. loss_train_step before backward: tensor(0.0106, device='cuda:0', grad_fn=<MseLossBackward>)
  38091. loss_train_step after backward: tensor(0.0106, device='cuda:0', grad_fn=<MseLossBackward>)
  38092. loss_train: 1.3586613987572491
  38093. step: 88
  38094. running loss: 0.015439334076786921
  38095. Train Steps: 88/90 Loss: 0.0154 torch.Size([8, 600, 800])
  38096. torch.Size([8, 8])
  38097. tensor([[0.6182, 0.4099, 0.7812, 0.3000, 0.3937, 0.2367, 0.5325, 0.5750],
  38098. [ nan, nan, 0.7515, 0.2708, 0.3987, 0.2267, 0.5162, 0.5567],
  38099. [0.6078, 0.4033, 0.8019, 0.3055, 0.3450, 0.4200, 0.6025, 0.5550],
  38100. [0.6168, 0.4081, 0.8438, 0.3367, 0.3500, 0.4667, 0.6212, 0.5400],
  38101. [0.6239, 0.4174, 0.8425, 0.5733, 0.4825, 0.4500, 0.5625, 0.5933],
  38102. [0.6346, 0.4092, 0.7712, 0.5917, 0.4037, 0.4767, 0.7343, 0.5725],
  38103. [0.6186, 0.3967, 0.7337, 0.1992, 0.4120, 0.2508, 0.6105, 0.5395],
  38104. [0.6179, 0.4118, 0.7278, 0.4237, 0.3588, 0.3400, 0.5675, 0.5917]],
  38105. device='cuda:0', dtype=torch.float64)
  38106. predictions are: tensor([[ 0.7738, -0.2730, 1.4386, -1.0415, -0.3418, -1.2038, 0.2729, 0.1546],
  38107. [-2.5829, -2.4818, 1.2752, -1.0553, -0.4747, -1.0027, 0.1306, 0.1917],
  38108. [ 0.3546, -0.5509, 1.5533, -0.9022, -0.6610, -0.4159, 0.4727, 0.1213],
  38109. [ 0.2965, -0.5853, 1.7268, -0.7251, -0.6103, -0.2529, 0.5206, 0.1225],
  38110. [ 0.4197, -0.4712, 1.7078, 0.2913, -0.0776, -0.2989, 0.2674, 0.2726],
  38111. [ 0.4849, -0.4927, 1.4642, 0.1457, -0.4687, -0.2381, 0.8454, 0.2116],
  38112. [ 0.5784, -0.4329, 1.1811, -1.4583, -0.3245, -1.2026, 0.5778, 0.1081],
  38113. [ 0.6367, -0.3579, 1.3764, -0.5841, -0.5169, -0.8698, 0.3198, 0.3117]],
  38114. device='cuda:0', grad_fn=<AddmmBackward>)
  38115. landmarks are: tensor([[[ 0.5695, -0.3927, 1.3226, -0.9002, -0.4672, -1.1928, 0.1737,
  38116. 0.3700],
  38117. [-2.2859, -2.2859, 1.1854, -1.0352, -0.4441, -1.2390, 0.0986,
  38118. 0.2853],
  38119. [ 0.5215, -0.4232, 1.4182, -0.8747, -0.6924, -0.3460, 0.4970,
  38120. 0.2776],
  38121. [ 0.5631, -0.4008, 1.6113, -0.7309, -0.6693, -0.1304, 0.5836,
  38122. 0.2083],
  38123. [ 0.5959, -0.3579, 1.6055, 0.3623, -0.0573, -0.2074, 0.3122,
  38124. 0.4547],
  38125. [ 0.6454, -0.3960, 1.2764, 0.4470, -0.4210, -0.0842, 1.1057,
  38126. 0.3585],
  38127. [ 0.5713, -0.4538, 1.1028, -1.3659, -0.3831, -1.1273, 0.5340,
  38128. 0.2058],
  38129. [ 0.5680, -0.3840, 1.0756, -0.3290, -0.6289, -0.7155, 0.3353,
  38130. 0.4470]]], device='cuda:0')
  38131. loss_train_step before backward: tensor(0.0193, device='cuda:0', grad_fn=<MseLossBackward>)
  38132. loss_train_step after backward: tensor(0.0193, device='cuda:0', grad_fn=<MseLossBackward>)
  38133. loss_train: 1.3779842429794371
  38134. step: 89
  38135. running loss: 0.015482969022240866
  38136. Train Steps: 89/90 Loss: 0.0155 torch.Size([8, 600, 800])
  38137. torch.Size([8, 8])
  38138. tensor([[0.6129, 0.3925, 0.8720, 0.5246, 0.4534, 0.5515, 0.6026, 0.5114],
  38139. [ nan, nan, 0.7553, 0.2722, 0.3875, 0.2550, 0.5125, 0.5633],
  38140. [0.6186, 0.4060, 0.8750, 0.5050, 0.3538, 0.4367, 0.5813, 0.6083],
  38141. [0.6261, 0.3987, 0.9045, 0.4208, 0.3600, 0.4633, 0.6570, 0.5162],
  38142. [0.6132, 0.4066, 0.7259, 0.2402, 0.3588, 0.3300, 0.6000, 0.5600],
  38143. [0.6286, 0.3977, 0.9038, 0.4733, 0.3900, 0.4150, 0.7074, 0.5320],
  38144. [0.6203, 0.4021, 0.8780, 0.5031, 0.3667, 0.3882, 0.5842, 0.5405],
  38145. [0.6261, 0.4066, 0.8325, 0.2150, 0.4763, 0.2667, 0.7002, 0.5633]],
  38146. device='cuda:0', dtype=torch.float64)
  38147. predictions are: tensor([[ 0.5745, -0.4073, 1.7057, 0.1399, -0.1932, 0.2375, 0.3406, 0.0806],
  38148. [-2.1859, -2.2379, 1.2122, -1.0418, -0.5679, -0.9760, 0.1158, 0.2787],
  38149. [ 0.5295, -0.4313, 1.7288, 0.0070, -0.5881, -0.3866, 0.3565, 0.3972],
  38150. [ 0.4754, -0.4928, 1.8224, -0.3816, -0.6022, -0.2334, 0.5597, 0.0926],
  38151. [ 0.5929, -0.3855, 1.1233, -1.2087, -0.5744, -0.8558, 0.3877, 0.1927],
  38152. [ 0.5226, -0.4764, 1.7320, -0.1790, -0.4929, -0.5515, 0.6222, 0.1111],
  38153. [ 0.6722, -0.3593, 1.7096, -0.0041, -0.5225, -0.5976, 0.3466, 0.1197],
  38154. [ 0.4967, -0.4938, 1.5374, -1.2878, -0.1328, -1.1846, 0.8814, 0.1625]],
  38155. device='cuda:0', grad_fn=<AddmmBackward>)
  38156. landmarks are: tensor([[[ 0.5450, -0.4730, 1.7420, 0.1372, -0.1919, 0.2614, 0.4976,
  38157. 0.0764],
  38158. [-2.2859, -2.2859, 1.2030, -1.0288, -0.4961, -1.1081, 0.0813,
  38159. 0.3161],
  38160. [ 0.5713, -0.4105, 1.7557, 0.0467, -0.6520, -0.2690, 0.3988,
  38161. 0.5239],
  38162. [ 0.6059, -0.4442, 1.8921, -0.3422, -0.6230, -0.1461, 0.7489,
  38163. 0.0985],
  38164. [ 0.5466, -0.4080, 1.0668, -1.1764, -0.6289, -0.7617, 0.4855,
  38165. 0.3007],
  38166. [ 0.6174, -0.4490, 1.8885, -0.0996, -0.4845, -0.3691, 0.9814,
  38167. 0.1715],
  38168. [ 0.5791, -0.4289, 1.7694, 0.0379, -0.5923, -0.4927, 0.4126,
  38169. 0.2107],
  38170. [ 0.6059, -0.4080, 1.5594, -1.2928, -0.0861, -1.0542, 0.9485,
  38171. 0.3157]]], device='cuda:0')
  38172. loss_train_step before backward: tensor(0.0084, device='cuda:0', grad_fn=<MseLossBackward>)
  38173. loss_train_step after backward: tensor(0.0084, device='cuda:0', grad_fn=<MseLossBackward>)
  38174. loss_train: 1.3864263747818768
  38175. step: 90
  38176. running loss: 0.015404737497576409
  38177. Valid Steps: 10/10 Loss: nan 6.8743
  38178. --------------------------------------------------
  38179. Epoch: 10 Train Loss: 0.0154 Valid Loss: nan
  38180. --------------------------------------------------
  38181. Training Complete
  38182. Total Elapsed Time : 463.1469588279724 s
  38183.  
Advertisement
Add Comment
Please, Sign In to add comment
Public Pastes
We use cookies for various purposes including analytics. By continuing to use Pastebin, you agree to our use of cookies as described in the Cookies Policy.  OK, I Understand
Not a member of Pastebin yet?
Sign Up, it unlocks many cool features!